mirror of
https://github.com/wu736139669/hapi.git
synced 2026-08-05 06:24:37 +00:00
fd457e030359527f7212d7df73dff28e939f9479
75
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 SSU-WEI HUANGandGitHub
|
5b91504263 |
Add read-only session status panel (#1301)
* feat(web): add session status panel * fix(web): include nested session activity * fix(web): preserve incomplete status details * fix(web): handle anonymous terminal completion * fix(web): classify pending status accurately * fix(web): retain ambiguous terminal candidates |
||
|
weishu
|
58ff91e5c4 |
fix(web): correct context window calculation for local-mode sessions
Remote sessions get an authoritative context_window injected into usage by the CLI (from SDK result modelUsage), but local-mode sessions forward raw transcript JSONL whose usage has no context_window, so the web fell back to getContextBudgetTokens — which could not tell Fable's 1M window from its bare id (only the "[1m]" suffix was recognized) and received session.model, which is usually null for local sessions, defaulting to a 200k budget. A 1M Fable session with ~256k of context showed 135% used with an over-limit warning instead of ~26%. - modelConfig: recognize Fable ids (fable, fable[1m], claude-fable-*) as 1M - reducer: LatestUsage carries the usage-bearing message's own model - StatusBar: new contextModel prop feeds the fallback heuristic only; the model prop's other semantics are unchanged - HappyComposer forwards it; SessionChat passes latestUsage.model ?? session.model |
||
|
Haoqing WangandGitHub
|
3a931e3c81 |
fix(web): count unseen messages by rendered block, not raw message (#1255)
* fix(web): count unseen messages by rendered block, not raw message The "N new messages" pill counted raw DecryptedMessages while the timeline renders folded blocks, so the two never agreed. A subagent run is dozens of sidechain messages but a single Task card; a tool_use and its tool_result are two messages and one card; consecutive tools collapse into one group. The pill could read "47 new messages" when scrolling down revealed two new rows. collectNewUnseenIds never inspected isSidechain, and it could not: the reducer's grouping is stateful (it needs the Task tool_use before it can map parentToolUseId), so a per-message predicate in the store cannot reproduce it. Adding an isSidechain check there would also invert the error for orphan sidechain messages, which tracer.ts falls back to emitting at the top level. Instead, drop the store's unseen bookkeeping entirely and count what the renderer actually produced. Watermark the visible blocks when the user scrolls away from the tail, then count the blocks past the last one they had seen. The count is anchor-based rather than timestamp-based because the blocks array is not monotonic in createdAt: messages sort by invokedAt ?? createdAt, so a queued message carries an old createdAt while sitting at the end. Anchoring also makes prepended history free, since older blocks land before the anchor. Known limit, documented at the call site: once the history window fills up, mergeIntoWindow trims incoming messages off the tail, so the pill reports 0 instead of a count. Under-reporting is preferable here, and returning to the tail force-refetches the latest page anyway. * fix(web): keep unseen watermark stable across optimistic id replacement The watermark snapshotted only block.id, but that id is not stable for the user's own messages: mergeMessages replaces an optimistic row with a stored row that keeps localId under a new server id, and the user block renders with the message id. Scrolling into history while an own message was still optimistic meant its echo anchored one block earlier and bumped the pill by one, with no new rendered row. Track localId alongside id in the watermark and match on either. Reported by HAPI Bot on #1255. * fix(web): count joined assistant cards, not pre-join blocks visibleBlocks is still not one-to-one with rendered rows: assistant-ui joins a run of adjacent assistant-role blocks into a single card, so a response made of reasoning + text + a tool call was reported as three new messages instead of one, and appending another block to an in-flight response bumped the pill without adding a row. Walk the blocks after the anchor and only start a new row where the assistant run breaks. Role assignment is the part that would drift, so rather than restating it, visibleBlockRole moves from assistant-runtime.ts to toolGroups.ts (next to the VisibleChatBlock definition it describes) and both the runtime and the counter import the one copy. Reported by HAPI Bot on #1255. |
||
|
Haoqing WangandGitHub
|
eaf5ac49bd |
fix: stop context/cache stats from jumping (subagent usage + stripped context_window) (#1256)
* fix(web): exclude subagent usage from the parent context indicator The status bar's `ctx N/M` and `cache N` come from latestUsage, which scans the normalized messages backwards for the most recent usage. That scan includes sidechain messages, so while a Task subagent runs its usage — describing the subagent's own, much smaller context — becomes the parent's numerator, then snaps back when the parent resumes. The existing `scope_role !== 'child'` guard never fired on any path. Claude never stamps scope_role (sdkToLogConverter.ts says so outright), and Codex drops child token_count events in the CLI before they can reach the web layer, so no producer ever emits 'child'. isSidechain is the signal that actually survives. sdkToLogConverter.ts:308-313 already documents this exact reducer behaviour, but works around only the denominator by forcing the main session's context_window onto sidechain messages. The numerator was left unguarded. * fix(cli): stop stripping context_window from local-session usage UsageSchema is a plain z.object, so Zod's default strip mode drops every undeclared key. sessionScanner forwards parsed.data rather than the raw line, so on the local-JSONL path usage is truncated to the five declared fields and context_window — injected on the SDK path by sdkToLogConverter — never survives. The web status bar then falls back to getContextBudgetTokens, which subtracts a 10k headroom, so the same model reports a 1.0M denominator on a remote session and 990k on a local one. RawMessageSchema right below already carries .passthrough() with a comment about losing message.model and messageId the same way; the nested usage object just never got the same treatment. |
||
|
SSU-WEI HUANGandGitHub
|
0e3b4394ff |
fix(web): label terminal cards with concise command names (#1201)
* fix(web): label terminal cards with command names * fix(web): honor concise titles across shell cards |
||
|
AnanovoandGitHub
|
b28088bfe3 | feat(web): show timestamps in conversation outline (#1111) | ||
|
Junmo KimandGitHub
|
deb5d63695 | fix(cli,web): group orphaned subagent trace by parentToolUseId (#1175) | ||
|
AnanovoandGitHub
|
3021c9cba0 | fix(web): zero-pad message and session timestamps (#1112) | ||
|
SSU-WEI HUANGandGitHub
|
aa5beb3af2 | feat(codex): preserve native exploration actions (#1139) | ||
|
weishu
|
74ad25ec57 | feat(codex): refine tool activity display | ||
|
SSU-WEI HUANGandGitHub
|
6bedd0d924 | feat(tooling): preserve native tool titles (#1133) | ||
|
SSU-WEI HUANGandGitHub
|
e69ca0782f | feat(web): make grouped tool summaries specific (#1134) | ||
|
Junmo KimandGitHub
|
289c9f2218 |
feat(cli,web): show Claude Code's away recap in local-mode chat (#1089)
* feat(shared,cli): whitelist away_summary so auto recap reaches the hub Claude Code's local TUI writes an automatic away-summary recap to the session transcript on window blur/focus (5min+ idle), but VISIBLE_CLAUDE_SYSTEM_SUBTYPES dropped it before it ever reached the hub. Add it to the whitelist so the local launcher forwards it like the other system subtypes, and cover the forwarding + Zod passthrough of the recap `content` field with tests. * feat(web): render Claude Code's automatic away recap in the chat Once away_summary reaches the hub (previous commit), the web chat still dropped it silently: normalizeAgent had no branch for the subtype, so it fell through to `return null`. Add a `recap` AgentEvent, a normalizeAgent branch mirroring the existing turn_duration/compact subtype branches, and a presentation entry that prefixes the text with `recap:` so it reads distinctly from the manual /recap assistant bubble (which already renders as a normal message). No new render component needed: it flows through the existing generic system-event row (SystemMessage.tsx + getEventPresentation) that every other system subtype already uses. * fix(web): drop inaccurate manual-/recap comparison from recap comments |
||
|
Junmo KimandGitHub
|
2ce6d3ef3a |
feat(web): show tool call duration in the detail dialog (#1036)
* refactor(web): export formatDuration for reuse * feat(web): show tool call duration in the detail dialog Show a completed tool's execution duration at the top of its detail dialog. The value is derived from the Claude entry's own timestamps (the execution machine's wall clock) rather than the hub's message-receive time, and is used only when both the tool_use and tool_result entries carry a real timestamp — otherwise it falls back to the hub receive times on both sides, so the two clocks are never mixed. Running/pending tools show nothing, the running-state live timer is unchanged, and clock skew is guarded against. Reuses the existing formatDuration formatter. No schema changes. * fix(web): backfill hub startedAt on reorder so duration isn't 0.0s When a tool_result entry is reduced before its tool_use, the tool block is created from the result, so the hub startedAt is the result receive time. The tool_use path only lowered the exec start, not the hub startedAt, so a timestamp-less pair (no exec duration available) fell back to startedAt === completedAt and the detail dialog showed 0.0s. Lower the hub startedAt to the earlier tool_use receive time as well. |
||
|
weishu
|
8782b8a110 |
fix(codex): align local transcript message projection
Use semantic events as the visible text source and preserve completed plans as ordered plan proposal cards. |
||
|
Junmo KimandGitHub
|
e45fde51e9 |
fix(claude): stop 1M/200k context-window flicker in the status bar (#992)
* refactor(claude): thread session's selected model into SDKToLogConverter
Adds an optional selectedModel field to the converter's context, wired
from session.getModel() in the launcher, so a later commit can seed the
turn-1 contextWindow estimate for presets whose system/init model
arrives without the "[1m]" suffix. No behavior change yet.
* fix(claude): key contextWindow cache by model to stop 1M/200k flicker
The remote launcher re-emits system/init on every turn for the same
converter instance. Its init-time estimate only checked whether the
model string ended in "[1m]", but current claude CLI versions strip
that suffix from system/init for some 1M presets (fable[1m] arrives as
"claude-fable-5"), so the estimate guessed 200k for them. The one
authoritative value is result.modelUsage[<model>].contextWindow, which
arrives after the heuristic has already injected 200k into that turn's
assistant message and then gets clobbered back to 200k by the very
next turn's init - producing the observed 200k<->1M oscillation in the
web status bar.
Cache the authoritative contextWindow per model id instead of a single
session-wide number, and only let system/init seed a heuristic guess
for a model that has no cached value yet, so a same-model re-init no
longer downgrades an already-learned value.
Two observed facts about the CLI's model ids drive the design:
system/init.model and the result.modelUsage keys always agree with
each other within a session (both bare for plain/fable[1m], both
suffixed for opus[1m]/sonnet[1m]), while each per-turn assistant
message reports its model bare and thus can't distinguish a 200k plain
preset from its 1M "[1m]" variant on tiers where they share a base id.
So the cache is keyed on the raw id (init/result agree, no
normalization) and assistant injection looks the value up via
resolvedModel (the last init id) rather than the lossy message.model.
Keying raw keeps a plain preset and its [1m] variant on distinct
entries; looking up via resolvedModel also means sidechain (Task
subagent) messages carry the main session window rather than the
subagent's own, since the web status bar picks the most recent usage
message without filtering sidechains and would otherwise flicker to
the subagent's smaller window while it runs.
For presets whose init model arrives bare even though they are 1M
(fable[1m]), the originally-selected preset - which preserves the
"[1m]" suffix - seeds the turn-1 estimate, kept live across mid-session
model switches via updateSelectedModel() (called from the launcher on
every turn) so it never goes stale.
* fix(web): recognize [1m] suffix on full Claude model ids in budget fallback
getContextBudgetTokens already special-cased "[1m]" for short preset
values (e.g. "opus[1m]") but fell through to the default 200k budget
for full model ids (e.g. "claude-opus-4-8[1m]"), which is what the CLI
now reports once context_window isn't available and this fallback is
consulted. Check the suffix on that branch too so it stays a correct
last-resort even without a session-provided context_window.
* refactor(web): merge duplicate Claude context-budget branches
isClaudeModelPreset(trimmedModel) and the startsWith('claude-') branch
below it had become byte-for-byte identical bodies after the [1m]
suffix check was added to both. Merge them into one condition; no
behavior change.
* fix(claude): distinguish fable from fable[1m] when the CLI reports both bare
The per-model contextWindow cache keyed on the raw system/init model id,
on the assumption that a 1M preset and its plain form always land on
distinct ids. That holds for opus[1m]/sonnet[1m] (the CLI reports the
"[1m]" suffix on their init and result ids) but not for fable: the CLI
reports both "fable" and "fable[1m]" with the same bare id
"claude-fable-5". So the "seed only if not already cached" guard would
skip re-seeding when switching fable[1m] -> fable, leaving the stale 1M
in place until fable's own result arrived - the same switch flicker this
change set out to remove, just for fable specifically.
Fold the selected preset's "[1m]" back into the cache key
(computeContextWindowKey): when the init model arrives bare but the
session selected an "[1m]" preset, key the entry as "<id>[1m]" so the 1M
and plain variants stay distinct; ids the CLI already suffixed are left
as-is. Seeding, lookups, and the current-model result entry all use this
resolved key. Subagent result entries (e.g. haiku) keep their own raw id
so the session's "[1m]" is never folded onto a model that isn't the
selected one.
|
||
|
Junmo KimandGitHub
|
f51e06e8f3 |
fix(web): stop pinning resolved permission cards to the bottom of the chat (#974)
agentState keeps an answered permission request in completedRequests. When its tool_use message is not in the loaded window, the permission-only synthesis appended a card to the end of the timeline — and there is no chronological re-sort, so the card stays pinned above the composer as a stale "answered" card that never moves to its place in history. With several answered asks this piles up at the bottom of the chat. Synthesize a card only for a *pending* request (the case that needs an answerable card when its message hasn't loaded). A resolved request is history and renders only via its own message when that message is in the window. |
||
|
|
02a0aa6733 |
fix(cursor): surface agent errors with warning styling in web UI (#871)
* test: reproduce issue #864 Assert Cursor error paths emit agent error payloads and web UI renders them as warning-styled events instead of neutral session messages. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(cursor): surface agent errors with warning styling in web UI (closes #864) Route Cursor stderr, init, prompt, and legacy exit failures through sendAgentMessage({ type: 'error' }) and teach the web chat layer to render error events with a warning icon instead of neutral info text. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
|
e23ae1b265 |
feat: add Pi Coding Agent support (#862)
* docs: spec for hapi-pi-agent-backend
* docs: spec retrospect for hapi-pi-agent-backend
* docs: plan for hapi-pi-agent-backend
* docs: plan retrospect for hapi-pi-agent-backend
* feat(pi): add hapi pi command with JSONL transport and event converter
- PiTransport: spawn pi --mode rpc, JSONL stdio, ENOENT/EPIPE handling
- PiEventConverter: Pi AgentEvent → HAPI AgentMessage conversion
- runPi: session lifecycle, dual-track event routing, model switching
- pi command: CLI registration with PI_PERMISSION_MODES
- Shared: add 'pi' to AGENT_FLAVORS, FLAVOR_CAPS, FLAVOR_LABELS
30 tests passing (15 transport + 15 converter)
* fix(pi): add Pi RPC types, fix double-cleanup/double-start/converter safety net
- Add cli/src/pi/types.ts with PiAgentEvent/PiResponseEvent discriminated unions
- PiTransport: constructor uses options object, double-start guard, drop log
- PiEventConverter: typed events via type assertions, top-level try/catch
- runPi: safeCleanup guard prevents double-cleanup race, sendAgentMessage
for converted events, keepAlive() for session pings
- 33 tests passing
* docs: dev phase reviews and test results for hapi-pi-agent-backend
- Business logic review: pass (0 must_fix)
- Standards review: pass (0 must_fix)
- Taste review: P0 types issue fixed in code
- Robustness review v2: pass (v1 3 MUST_FIX all fixed)
- Integration review: pass (0 must_fix)
- Test results: 33 passing, all type errors resolved
* docs: taste review v2 pass after type definition fixes
* docs: dev retrospect for hapi-pi-agent-backend
* test: test execution for hapi-pi-agent-backend (20/20 pass)
* fix: add taste_review symlink for gate pattern match
* docs: test retrospect for hapi-pi-agent-backend
* fix(web): add pi to MODEL_OPTIONS Record type
* ci: PR and CI evidence for hapi-pi-agent-backend
* docs: overall retrospect for hapi-pi-agent-backend (all 5 phases)
* test(pi): add buffer split, missing fields, and handleResponse tests
- PiTransport: buffer cross-chunk reassembly test
- PiEventConverter: tool_execution_end with missing result/toolCallId
- handleResponse: 10 tests covering all branches (error, get_state,
set_model, new_session, abort, prompt, unknown command)
- Extract handleResponse to accept onUpdate callback for testability
- Total: 46 tests passing (was 33)
* fix(pi): set requiresRuntimeAssets to false — pi runs as subprocess, no native tools needed
* refactor(cli): lazy import ensureRuntimeAssets to reduce startup overhead
* docs: add 15 manual E2E protocol test cases (TC-4-xx) based on real Pi RPC capture
- TC-4-01 to TC-4-15: manual tests covering tool execution, thinking
lifecycle, multi-turn, abort, error scenarios, model switch, cleanup
- Priority: P0 (tool fields, failure, thinking, multi-turn, abort)
> P1 (basic conversation, write tool, model switch, usage) > P2 (edge cases)
- Includes actual Pi RPC event sequence from live capture as reference
- e2e-test-plan.md updated with test environment setup instructions
- Total test cases: 35 (6 unit + 14 integration + 15 manual)
* test: E2E protocol test results for hapi-pi (11/15 pass)
P0/P1 automated tests (8/8 pass):
- TC-4-01: Basic text conversation ✓
- TC-4-02: Tool read (field names verified) ✓
- TC-4-03: Tool write (file created) ✓
- TC-4-04: Tool failure (isError=true) ✓
- TC-4-05: Thinking lifecycle + usage ✓
- TC-4-06: Multi-turn context retention ✓
- TC-4-07: Abort generation ✓
- TC-4-14: Token count ✓
- TC-4-15: Extension UI events ignored ✓
P2 results:
- TC-4-10: Invalid token → 401 ✓
- TC-4-12: Ctrl+C cleanup, no orphans ✓
- TC-4-08: ENOENT (harness issue, exit code correct)
- TC-4-11: set_model not supported by Pi (success=false)
- TC-4-13: Pi crash (harness output capture issue)
* test: fix TC-4-11 result — Pi set_model works with correct provider/modelId
Previous test used invalid provider='' + modelId='deepseek-chat'.
Re-tested with provider='deepseek' + modelId='deepseek-v4-flash':
- set_model success=true
- model switched glm-5.1 → deepseek-v4-flash
- subsequent prompt confirmed working
Final E2E results: 12/15 PASS, 2 FAIL (test harness), 1 SKIP
* chore: remove .xyz-harness/ from git tracking, add to .gitignore
Local harness workflow artifacts should not be tracked in the repo.
* fix(pi): resolve web UI bugs for hapi-pi integration
Five bugs fixed for end-to-end pi session via hapi web UI:
1. runner buildCliArgs: add 'pi' branch to spawn correct command
(was falling back to 'claude', launching wrong agent)
2. runPi: implement real keep-alive (2s interval) to prevent hub
30s timeout marking session inactive
3. runPi: bump keep-alive to active state during agent/turn_start
4. sessionResume: add 'pi' to flavor switch and resume condition
(was returning undefined, causing 'cannotResume' on inactive session)
5. PiEventConverter: emit codex-compatible {type:'message',message:...}
/{type:'reasoning',message:...} with streamId; dedup by skipping
text_start/text_end (only send deltas) to avoid triple-rendered text
6. PiTransport: fallback to stdout 'end' event when child process
close event doesn't fire (bun spawn quirk)
Verified end-to-end: web UI shows pi reasoning + reply correctly,
session stays online, no duplicate text.
* fix(pi): address 4 web UI display bugs in hapi-pi integration
Three of four follow-up bugs reported after the initial fix (6c28949):
1. Stuck in 'queued' status — fix
Pi's runner doesn't use MessageQueue2, so the base session's
onBatchConsumed hook never fires. Add a FIFO of pending localIds
in runPi and emit messages-consumed on agent_start. turn_start
is intentionally skipped (it can fire multiple times per agent
run after tool calls). A prompt rejection from Pi also consumes
the localId so the next prompt isn't poisoned.
2. AI thinking only displays ':' — fix
Pi emits pure incremental deltas (text_delta / thinking_delta)
per token. The web reducer dedupes reasoning by streamId WITHIN
one message's content array only — separate wire messages
produce separate renders. Without accumulation, 50 deltas = 50
reasoning renders, of which the reducer keeps only the last
delta (a single character like ':').
3. Output text on separate lines — fix
Same root cause as #2 but for text: the reducer appends each
text AgentMessage as a new agent-text block (no dedup), so 50
deltas become a 50-row character-by-character column.
4. Tool call execution status (in_progress -> completed)
The tool result wire CodexMessage type is 'tool-call-result'
(with callId + is_error?); the internal AgentMessage 'tool_result'
is converted to that. Status mapping is preserved.
Implementation: extract a PiMessageAccumulator class (testable in
isolation) that mirrors codex's ReasoningProcessor pattern:
- message_start resets state and streamId
- text_delta / thinking_delta append to internal text / reasoning
- text_start/thinking_start/text_end/thinking_end ignored (they
carry full partial state — would duplicate)
- message_end flushes (max 1 reasoning + 1 text message, in order)
- turn_end safety net flushes if active
- flushIfActive() exposed for transport close / crash
The converter now routes AgentMessage through convertAgentMessage
so the wire format is codex-shaped (matches opencode/gemini/kimi
path). AgentMessage 'text' and CodexMessage 'message' both gain
optional id; convertAgentMessage preserves caller-provided id for
streamId-based dedup on the web side.
Tests: 16 new PiMessageAccumulator tests + 5 updated
PiEventConverter tests + 4 messageConverter tests, all passing.
Full suite: 909/910 (1 unrelated macOS path normalization). tsc
clean.
* fix(pi): review round 1 - 1 must-fix issue
The web session-resume helper referenced metadata.piSessionId, but the
shared MetadataSchema does not define the field, and the back-end has no
path to populate it (Pi session resume is out of scope per spec.md).
This caused web typecheck to fail and would also have produced a
runtime 'resume_unavailable' from the hub if a user tried to resume a Pi
session that had any user messages (the stale 'flavor === pi' branch in
inactiveSessionCanResume claimed resume was supported).
Revert the two early Pi branches from the web resume helper. Add a
comment pointing at the spec and noting what to undo when back-end
resume ships (re-add 'case pi' + 'piSessionId' on MetadataSchema +
extend hub resolveAgentResumeId).
* fix(pi): review round 2 - 4 must-fix issues
1. cli/src/runner/run.ts buildCliArgs: stop forwarding --resume to the pi
binary. Pi session resume is out of scope (no piSessionId on
Metadata), so forwarding would create an orphan session the hub can't
track. Hub already returns null from resolveAgentResumeId for
flavor='pi' and falls through to fresh spawn; this just hardens the
runner layer to match.
2. cli/src/pi/runPi.ts: cache currentProvider from get_state and use it
for subsequent set_model RPCs. Pi's set_model requires both provider
and modelId, but the bootstrap-time code emitted provider: '' which
Pi rejects. The bootstrap-time model is still applied by Pi at
startup, so suppressing set_model until get_state arrives is a no-op
for same-model configs rather than a wrong-model emit.
3. web/src/components/AssistantChat/modelOptions.ts: add explicit pi
branches to getModelOptionsForFlavor and getNextModelForFlavor.
Without them, Pi sessions fell through to the Claude preset cycler,
which would push sonnet/opus ids into a Pi session via
set-session-config. Mirrors the opencode handling introduced earlier.
Tests added/updated: buildCliArgs covers pi + claude resume; handleResponse
mirror test covers provider caching; modelOptions tests cover pi
no-fallback behavior for both option list and cycler.
* fix(pi): add session resume support and fix review issues
- Add piSessionId to MetadataSchema (shared/src/schemas.ts)
- Persist piSessionId from get_state response to metadata (cli/src/pi/runPi.ts)
- Pass --session-id to Pi spawn on resume (cli/src/pi/runPi.ts)
- Add pi branch to resolveAgentResumeId (hub/src/sync/syncEngine.ts)
- Add case 'pi' to resolveAgentSessionIdFromMetadata (web/src/lib/sessionResume.ts)
- Replace pi resume skip guard with --session-id forwarding (cli/src/runner/run.ts)
- Preserve piSessionId in pickExistingSessionMetadata (cli/src/agent/sessionFactory.ts)
- Add pi badge to AgentFlavorIcon (web/src/components/AgentFlavorIcon.tsx)
- Fix transport.onClose crash-marking on normal shutdown (cli/src/pi/runPi.ts)
* fix(pi): review round 1 - 3 must-fix issues
- resume.ts: add pi branch to dispatchLocalResume() so hapi resume
dispatches to runPi instead of falling through to cursor
- runPi.ts: accept existingSessionId and use bootstrapExistingSession
when resuming, matching other agents' pattern
- agentCommandOptions.ts: parse --session-id in addition to --resume
so runner-spawned pi resume actually forwards the session ID
- types.ts: export PiPermissionMode alongside other agent permission
mode types for consistent import convention
* fix(pi): review round 2 - 2 must-fix issues
* refactor(workflow): improve pi-adaptation-review-loop robustness
- Switch from structured output to file-based JSON output for reliability
- Replace per-round file limit (20→30) with clear wording (remove misleading split-commits instruction)
- Return { data, error } from readResultFile() to surface parse/validation failures in abortReason
- Fix lastMustFix sentinel: initialize to null, use ?? for explicit N/A reporting
- Add getAgentDirs() to dynamically discover agent dirs from cli/src/
- Document rollbackTo() atomic-round design intent
- Add isValidIssue() validation, runFinalCleanup() helper, git repo pre-check
* test(pi): add coverage for pi flavor across shared, cli, and web
- shared/flavors.test.ts: pi/kimi capability, label, known, supports
- shared/modes.test.ts: PI_PERMISSION_MODES contract, per-mode checks
(7-mode allowed/denied matrix)
- web/AssistantChat/modelOptions.test.ts: pi shortcut vs Claude
cycler, normalize filter (auto/default/whitespace), kimi/cursor/
opencode cross-flavor consistency
- web/lib/sessionResume.test.ts: piSessionId resolver, cross-flavor
stale-id protection, inactiveSessionCanResume for pi, regression
coverage for all 6 other flavors
- web/components/AgentFlavorIcon.test.tsx: pi badge styling
(bg-[#5b21b6]), Un fallback, case/whitespace normalize,
className override
- cli/commands/agentCommandOptions.test.ts: --session-id
(pi-specific flag), --resume alias, PI mode validation,
--yolo vs explicit-mode priority
137 new test cases, all passing. Full suite: 96 files / 933 tests
green (unrelated apiMachine.test.ts macOS /private/var path issue
remains as documented in handoff).
* feat(pi): implement P0 — context budget bar + dynamic model discovery
P0-1: Context Budget Bar
- Add pi branch to modelConfig.ts getContextBudgetTokens()
- Conservative 200K default context window for Pi sessions
P0-2: CLI-side model discovery
- Add get_available_models to PiRpcCommand type
- Auto-send get_available_models after get_state in runPi.ts
- Cache model list and push to session metadata
- Register ListPiModels RPC handler with promise-based transport query
P0-3: Hub-side routing
- Add listPiModelsForSession to rpcGateway and syncEngine
- Add REST endpoint GET /sessions/:id/pi-models (pi sessions only)
P0-4: Web-side rendering
- Add PiModelSummary type to shared apiTypes
- Add usePiModels hook (TanStack Query, stale 60s)
- Add getSessionPiModels to API client
- Add sessionPiModels query key
- Wire piModelOptions into SessionChat availableModelOptions
- Model dropdown renders discovered models or falls back to Default
* fix(pi): address code review findings + pre-existing test issue
Review fixes:
- Fix race condition in sendPiRpcAndWait: use incremental id as key
instead of command type, preventing resolver overwrite on concurrent
calls (e.g. auto-discovery + ListPiModels RPC)
- Extract parsePiModels() to eliminate duplicated model parsing logic
between handleResponse and ListPiModels RPC handler (DRY)
- Add resolvePendingRpc() call in error response path to prevent
promise leaks when Pi rejects an RPC with an id
- Add piModelsState.error guard to onModelChange in SessionChat,
matching the pattern used by codex and cursor flavors
Pre-existing fix:
- Fix apiMachine.test.ts symlink assertion on macOS (/var vs
/private/var) by applying realpathSync to the expected path
* feat(pi): P1 — session rename sync, thinking level UI, skills/commands
P1-1: Session Rename → Pi notification
- Add set_session_name to PiRpcCommand
- Register RenamePiSession RPC handler in CLI
- Hub syncEngine.renameSession now forwards to Pi CLI for active sessions
- Hub rpcGateway + REST endpoint added
P1-2: Thinking Level support
- Add Pi thinking level constants to shared/src/piThinkingLevel.ts
(off/minimal/low/medium/high/xhigh)
- Add ThinkingLevel capability to Pi flavor in flavors.ts
- sessionConfigRpc now supports effortMode for Pi thinking level
- runPi captures thinkingLevel from get_state and forwards via
set_thinking_level
- Hub effort endpoint accepts pi sessions (was claude-only)
- Web: piThinkingLevelOptions.ts + HappyComposer renders Pi options
when flavor=pi
P1-3: Skills/Commands discovery
- Add get_commands to PiRpcCommand, auto-discover after get_state
- Register ListPiCommands + ListSlashCommands RPC handlers in CLI
(maps Pi commands to HAPI SlashCommand format)
- Hub: listPiCommandsForSession + REST GET /sessions/:id/pi-commands
- Web: usePiCommands hook + api client + query keys
Also fixes:
- Pre-existing ZodError.errors → ZodError.issues in hub/socket/server.ts
- Updated test expectation for effort endpoint error message
* feat(pi): implement P2 features — steer, queue modes, history, native images
P2-1: Steer/Follow-up
- Track piIsStreaming state from agent_start/turn_start/turn_end/agent_end
- When streaming, onUserMessage sends steer instead of prompt
- Added PiSteer/PiFollowUp RPC methods + hub routing + REST endpoints
P2-2: Queue modes
- Added set_steering_mode/set_follow_up_mode to PiRpcCommand
- CLI RPC handlers with mode state tracking
- Hub routing + REST POST endpoints
- Web API client methods
P2-3: History replay
- Added get_messages to PiRpcCommand
- CLI handler converts Pi AgentMessage to PiMessageEntry format
- Hub RPC routing + REST GET /sessions/:id/pi-messages
- Web usePiMessages hook + query key
P2-4: Native image passing
- Added PiImageContent type for base64 image data
- extractPiImages() helper reads attachment files as base64
- prompt/steer commands now include images field
- Falls back to @path text reference for non-image/unreadable files
* feat(pi): implement P3 advanced features — compact, fork, clone, switch, stats, export
P3 features for Pi agent integration:
- Compact: compact RPC with custom instructions, set_auto_compaction toggle
- Fork: fork at entry ID, get_fork_messages for fork context
- Clone: clone current Pi session
- Switch Session: switch Pi to a different session by path
- Session Stats: get token counts, message counts, cost
- HTML Export: export session as HTML file
All features follow existing P2 pattern:
- CLI: RPC handlers in runPi.ts with sendPiRpcAndWait
- Hub: rpcGateway + syncEngine routing + REST endpoints
- Web: API client methods + query keys + type exports + hooks (stats, fork messages)
Total: 8 new REST endpoints, 9 RPC handlers, 6 web API methods
Typecheck: all 3 packages pass (cli+hub+web)
Tests: 1155 pass (263 hub + 803 web + 89 shared), 0 failures
* refactor(pi): clean up runPi.ts imports and readability
- Replace require('fs') with top-level import { readFileSync } from 'fs'
- Extract handleGetState() as standalone function from handleResponse
switch case (get_state case: 35 lines → 4 lines dispatch)
Typecheck: all 3 packages pass
Tests: 1066 pass (263 hub + 803 web), 0 failures
* fix(pi): remove native image passing, fix version pollution
- Remove extractPiImages helper and PiImageContent type: all
attachments now use @path text references via
formatMessageWithAttachments, consistent with every other agent
- Remove images field from prompt/steer/follow_up RPC commands
- Remove unused readFileSync import
- Restore cli/package.json version from test pollution
(0.0.0-integration-test-should-be-auto-cleaned-up-51369 → 0.20.0)
Typecheck: all 3 packages pass
Tests: 1286 pass, 0 failures
* refactor(pi): extract hub helper, unify web hooks, fix import style
- Hub: extract withPiSession helper eliminating boilerplate across 15
Pi REST endpoints (~400 lines → ~150 lines)
- Web: unify usePiForkMessages and usePiSessionStats to return
destructured typed fields matching usePiModels/usePiCommands pattern
- Web: move 15 Pi response types from inline import() to top-level
named imports in api/client.ts
- CLI: remove duplicate PiCommandSummary/PiCommandsResponse from
types.ts, re-export from @hapi/protocol/apiTypes
Typecheck: all 3 packages pass
Tests: 1286 pass, 0 failures
* chore: untrack .agents/skills and .pi, fix .xyz-harness in gitignore
* refactor: remove unused text message id from converter layer, update gitignore
* fix: update tests for pi resume support and text id removal
* fix: restore cursor resume branch in buildCliArgs
* refactor: remove pi-specific rename from syncEngine, align with other agents
* refactor: remove effort field from sessionConfigRpc, Pi self-handles RPC
Pi agent now self-handles SetSessionConfig RPC (like Claude) using
the existing field, instead of adding a parallel
field to the shared sessionConfigRpc helper which only knows about
.
- Remove effort/effortMode from sessionConfigRpc types and logic
- runPi.ts: self-register RPC handler with PiThinkingLevel validation
- Reuse resolveSessionConfigPermissionMode from sessionConfigRpc
* refactor: consolidate Pi RPC layer from 36 methods to 3 generics
rpcGateway: 12 methods → callPiRpc<T>
syncEngine: 12 passthroughs → callPiRpc<T> delegate
web client: 12 methods → callPiEndpoint<T>
routes: use engine.callPiRpc with RPC_METHODS constants
hooks: use callPiEndpoint, add missing type imports
* chore: revert unrelated apiMachine test change
* refactor: remove unused ThinkingLevel capability from flavors
Pi's thinking level is an effort variant, not a separate capability.
The ThinkingLevel constant and supportsThinkingLevel() had zero callers
— the frontend uses flavor-based branching for effort option rendering.
* refactor: drop Pi prefix from generic RPC method names
* refactor: remove 13 Pi RPC methods with no UI consumers
Steer: already handled by onUserMessage auto-routing
Follow-up: redundant with HAPI message queue
ListPiCommands/GetMessages/ForkMessages/SessionStats: no UI
Compact/SetAutoCompaction/Fork/Clone/SwitchSession/ExportHtml: no UI
SetSteeringMode/SetFollowUpMode: no UI
Kept: ListPiModels (has UI), SetSessionConfig, ListSlashCommands, Abort, Switch
Deleted: 4 web hooks, 13 RPC handlers, 12 REST routes, 13 rpcMethods entries
Net: -730 lines
* refactor: extract session.ts and loop.ts from runPi.ts
Restructure Pi agent following Codex pattern (without Local/Remote
splitting since Pi only has remote mode):
- session.ts: PiSession class managing state + hub communication
- loop.ts: response parsing, RPC resolver, transport event wiring
- runPi.ts: thin entry (bootstrap, RPC handlers, lifecycle)
Changes from review:
- Encapsulate RPC resolver in PiRpcResolver class (session-scoped,
not module-level singleton)
- Remove unused extractTextFromPiMessage export
- Fix inline import('./types') → top-level import
* refactor: normalize Pi file naming and improve test coverage
- Rename PiTransport.ts → piTransport.ts, PiEventConverter.ts →
piEventConverter.ts, PiMessageAccumulator.ts → piMessageAccumulator.ts
(match project-wide camelCase convention)
- Delete handleResponse.test.ts (tested stale copy of inline function)
- Add loop.test.ts with 20 tests covering parsePiModels,
parsePiCommands, wireTransportEvents integration, and sendPiRpcAndWait
- Total Pi tests: 73 (was 53)
* test: add E2E harness with 4 core helpers and integration specs
Helper functions in e2e/harness.ts capture the four non-obvious
interactions discovered during the 2026-06-09 retest:
- longPress: SessionActionMenu is triggered by 500ms press, not click
- mockOffline: useOnlineStatus hook listens to navigator.onLine +
window offline event, not CDP Network.emulateNetworkConditions
- pollForText: thinking indicator flickers in <1s, 3s polling misses
- isVisible: element.offsetParent returns null for position:fixed
dialogs even when visible; use getBoundingClientRect
Plus Chrome lifecycle (startChrome/stopChrome, never pkill chrome)
and hub API helpers (loginWithToken, listSessions).
5 integration specs (e2e/integration/) cover:
- yolo-permission: toggle + localStorage persistence (4 cases)
- codex-dialog: pre-flight check + dialog render (3 cases)
- stress: 10 concurrent + invalid JWT + malformed + unknown
endpoint (5 cases, all PASS)
All 12 integration cases pass. Full E2E results in
.xzy-harness/2026-06-09-full-e2e-retest/ (67 cases, 0 functional
bugs found).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
* fix: resolve Pi model selection and thinking level issues
- Fix PiModelPanel: use provider+modelId composite for selection check
and React key, preventing duplicate highlights for same-name models
across different providers
- Fix PiThinkingLevelPanel: unify thinkingLevelMap filtering logic by
extracting shared isThinkingLevelSupported utility
- Fix HappyComposer: auto-reset effort to highest supported level when
switching models, update label to reflect effective level
* refactor: remove 29 dead exports from feat-pi-support
Remove unused types, methods, and re-exports identified by dead code audit:
shared/src/apiTypes.ts (19):
- SessionModelIdentifier, ListPiCommandsResponse
- PiSteeringMode, PiFollowUpMode, PiSteerResponse, PiFollowUpResponse
- PiQueueModeResponse, PiMessageEntry, PiMessagesResponse
- PiCompactResponse, PiSetAutoCompactionResponse
- PiForkResponse, PiForkMessageEntry, PiForkMessagesResponse
- PiCloneResponse, PiSwitchSessionResponse
- PiSessionStats, PiSessionStatsResponse, PiExportHtmlResponse
cli/src/pi/types.ts (6):
- PiSessionStats, PiCompactionResult, PiForkMessageEntry (dead local duplicates)
- PiCommandsResponse, PI_THINKING_LEVELS, PI_THINKING_LEVEL_LABELS (dead re-exports)
cli/src/pi/piMessageAccumulator.ts (1):
- flushIfActive() method (comment claimed runPi calls it, but it doesn't)
cli/src/pi/piTransport.ts (1):
- isRunning() method (never called in production code)
web/ (2):
- ProviderGroup, PiThinkingLevelOption (unnecessary exports, made local)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix: resolve 7 PR review issues in Pi support
#3 Remove duplicated PI_THINKING_LEVELS in schemas.ts, import from @hapi/protocol
#2 Add piAvailableModels field to MetadataSchema (schema-runtime consistency)
#6 Replace hardcoded flavor names with supportsEffort() in effort route
#1 Move PiRpcResolver from module-level singleton to PiSession instance
#4 Add piCachedModels fallback in piModelOptions useMemo
#7 Merge message_update dead branch into unified not-converted case
#10 Fix misleading Pi model list comments in modelOptions.ts
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix: normalize Pi model object to string in hub sessionCache (#5), remove extra blank line in rpcGateway (#8)
#5: applySessionConfig now extracts modelId from { provider, modelId }
before passing to setSessionModel / session.model, preventing
[object Object] from being stored in SQLite when Pi switches models.
#8: Remove double blank line before RpcGateway class declaration.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(pi): preserve piAvailableModels on resume, document SetSessionConfig divergence
- sessionFactory: preserve piAvailableModels in pickExistingSessionMetadata
so web shows cached models on inactive-session view without RPC round-trip
- sessionConfigRpc: extend resolveNullableSessionModel to accept
{provider, modelId} object form for schema consistency
- runPi: document why Pi manually registers SetSessionConfig instead of
reusing registerSessionConfigRpc (wire protocol needs separate fields)
- package.json: restore version to 0.20.0
* refactor: remove unused Pi types, extract JsonLineParser, clean up review findings
- Remove 13 unused PiRpcCommand variants and PiStreamingBehavior type (YAGNI)
- Remove unnecessary exports on 3 internal Zod schemas in pi/schemas.ts
- Extract JsonLineParser base class to utils/, shared by PiTransport,
CodexAppServerClient, and AcpStdioTransport (eliminates 3x duplicate
handleStdout buffer logic)
- Remove DEV-only duplicate session ID detection from SessionList.tsx
(debug code unrelated to Pi support scope)
- Add comments explaining key prefix rationale in SessionChat.tsx
* chore: remove unrelated E2E test harness from Pi support PR
E2E harness (codex-dialog, stress, yolo-permission, scratchlist specs)
was introduced in this branch but tests generic HAPI behavior unrelated
to Pi agent support. Should live in a separate PR.
* fix: wrap cursor model change handler for union type compatibility
* fix: apply startup --model to Pi and remove duplicate lockfile entry
1. --model startup bug:
- Add initialModel to PiSession to preserve startup model
- handleGetState preserves initialModel instead of overwriting with Pi default
- get_available_models handler resolves provider from cached models and sends set_model
2. bun.lock duplicate key:
- Remove duplicate @twsxtd/hapi-win32-x64@0.20.0 entry
- Fixes CI lockfile regeneration that caused hono type errors
* fix: update test expectation for effort endpoint error message
* fix(pi): resolve 8 link-review defects + abort session termination
- W1C-D-1: hasSameAgentSessionIds missing piSessionId/kimiSessionId
+ extractAgentSessionId also needs piSessionId recognition
- D-1: dispatchLocalResume pi branch missing effort param
- W1B-1-01: buildCliArgs only passes --effort for claude, not pi
- W2B-D-2: effort=null does not send set_thinking_level to Pi
- D-3: turn_start does not consume pendingLocalIds
- D-7: keep_alive falls into default case in convertPiEvent
- D-9: finally overwrites sessionEndReason set by Switch/Abort
- W2B-D-3: ListPiModels RPC does not update metadata
- Abort handler: remove cleanupAndExit, only cancel current turn
Also: Switch handler returns { success: true } for consistency
Test coverage: 13 new test cases across 5 files
* fix: restore cli version from integration test placeholder
* fix(pi): send restored thinking level to Pi subprocess on startup
opts.effort was stored in piSession.currentThinkingLevel but never
forwarded via set_thinking_level during the startup sequence, causing
runner-spawned and resumed sessions to show the restored effort in
HAPI while Pi kept its default.
* fix: restore cli package version from integration test residue
* fix(pi): switch-to-remote handler preserves session instead of terminating
Replace lifecycle.cleanupAndExit() with createModeChangeHandler + keepAlive
in the Switch RPC handler. Pi runs as a single long-lived subprocess
without BaseLocalLauncher's restart loop, so cleanupAndExit() permanently
destroyed the session on mode switch. The web handoff button now correctly
changes control mode while keeping Pi alive.
* fix(pi): remove permission mode selector (Pi RPC has no runtime switching)
Pi's --mode rpc is non-interactive and auto-approves all tool execution;
there is no set_permission_mode command in the protocol. The selector
reported success without changing Pi's behavior, misleading users.
Remove the concept across all four packages:
- shared: getPermissionModesForFlavor('pi') returns [] (cascades to
hub 400 + web UI auto-hide via length===0 guards); drop
PI_PERMISSION_MODES / PiPermissionMode
- cli: strip permissionMode from PiSession/runPi/pi command/resume;
drop the no-op SetSessionConfig permission branch that stored state
without forwarding to the subprocess
- web: delete PiPermissionPanel.tsx; remove panel block + imports
from HappyComposer
* fix(cli): realpath workspace root in apiMachine test assertion
The handler realpaths the cwd as a symlink-escape guard, so on macOS
/var/folders/... resolves to /private/var/folders/... The test compared
against the un-resolved path and failed on macOS. Use realpathSync on
the expected value for cross-platform consistency (no-op on Linux where
/tmp has no symlink prefix).
* fix(pi): keepalive reads current mode instead of constructor-time startingMode
The Switch handler updated controlledByUser but PiSession.pushKeepAlive()
still emitted the readonly startingMode every 2s, so a runner-started
session switched to local would flip back to remote on the next keepalive.
Replace readonly startingMode with a mutable mode field; add setMode()
that updates it and re-pushes keepAlive immediately. The Switch RPC
handler now calls setMode() before handleModeChange.
* fix(pi): runner no longer passes permission flags to Pi subprocess
After removing the Pi permission selector, the Pi command parser rejects
--permission-mode and ignores --yolo. But the shared buildCliArgs tail in
the runner still appended these flags for Pi sessions, making runner-
spawned Pi children exit before registering a session.
Guard the permission/yolo append with agent !== 'pi'.
* fix(pi): preserve provider identity when persisting selected Pi model
The hub's applySessionConfig normalized Pi's { provider, modelId } object
down to a plain modelId string for the shared session.model field, losing
the provider. On reload or next render, web's selectedPiModel lookup
matched by modelId alone — if two providers share a modelId, the wrong
one was highlighted, and subsequent model/thinking-level changes sent the
wrong provider to the Pi subprocess.
Add a provider-qualified piSelectedModel field to session metadata
(schema + persistPiSelectedModel mirroring persistPreferredPermissionMode).
Web's selectedPiModel now prefers the provider-qualified match and only
falls back to modelId-only matching when absent.
* fix(pi): model picker checkmark follows provider-qualified selection
selectedPiModel already resolves via provider+modelId, but the model
panel's currentPiModel still matched by modelId alone — so with two
providers sharing a modelId the checkmark pointed at the wrong row.
Reuse selectedPiModel directly.
* fix(pi): steer messages consumed immediately, not queued in pendingLocalIds
onUserMessage unconditionally pushed localId into pendingLocalIds, but a
steer (sent while piIsStreaming) does not start a new turn — so the
steer's localId was never drained by turn_start. The next normal prompt's
turn_start would consume the stale steer localId instead, leaving the
new prompt's bubble stuck in the queued bar.
Only queue localId for the prompt path. Steer path emits
messages-consumed immediately.
* fix(pi): clear stale thinking level when switching to non-reasoning model
The model-change effect early-returned when selectedPiModel.reasoning ===
false, leaving the previously-set effort (e.g. 'high') persisted on the
session. The UI hid the thinking picker for the non-reasoning model, but
the hub still forwarded the stale effort as set_thinking_level — with no
visible control to clear it.
Call onEffortChange(null) for non-reasoning models.
* fix(pi): return provider-qualified model in SetSessionConfig applied
The CLI handler returned only currentModel (bare string), so the hub's
applySessionConfig saw a non-object model and cleared
metadata.piSelectedModel via persistPiSelectedModel(session, null) —
undoing the provider that was just stored on the inbound config.
Return { provider, modelId } when both are known so the hub keeps the
provider-qualified metadata intact across active model changes.
* fix(pi): preserve piSelectedModel in bootstrapExistingSession metadata
The metadata whitelist rebuild kept piAvailableModels but omitted
piSelectedModel, so the first resume/local-handoff update dropped the
provider identity — after which web fell back to modelId-only matching
and could select the wrong provider for duplicate modelIds.
* fix(pi): await Pi confirmation before reporting model/effort applied
SetSessionConfig was fire-and-forget — transport.send wrote JSONL to
stdin and returned immediately. If Pi rejected an invalid provider/model
or thinking level, the hub still persisted the new value and the UI
reported success while Pi kept the old runtime state.
Use sendPiRpcAndWait so a failed set_model/set_thinking_level rejects
the web request and leaves the session config unchanged.
* fix(pi): resolve set_model RPC so awaited model switch does not time out
SetSessionConfig awaits sendPiRpcAndWait(set_model) before reporting the
model applied, but handleResponse's set_model branch updated state and
fell through without calling resolvePendingRpc. The pending RPC promise
then waited the full 10s timeout and rejected, making /sessions/:id/model
return 409 even though Pi accepted the change. Mirror every other branch
by resolving the pending RPC after updating currentModel/currentProvider.
* fix(pi): drain pending localId on turn_start only; throw when set_model suppressed
- loop.ts: split agent_start/turn_start branches. Pi emits both per prompt;
draining on both popped the FIFO twice and shipped an undefined localId to
the hub. agent_start now only sets thinking state; turn_start drains.
- runPi.ts: when set_model is suppressed (provider unknown), throw instead of
silently returning applied, so the hub returns 409 rather than persisting a
piSelectedModel Pi never received.
- loop.test.ts: assert agent_start does not drain; add regression test that a
single turn drains exactly one real localId.
* fix(pi): exclude Pi from generic Ctrl/Cmd+M model cycler
SessionChat fed piModelOptions into HappyComposer.availableModelOptions,
so the global Ctrl/Cmd+M shortcut ran getNextModelForFlavor over the Pi
list and called onModelChange with a bare modelId string. Pi needs
{ provider, modelId } to disambiguate duplicate model IDs across
providers; a bare string made runPi fall back to the first cached
provider match (wrong provider) or throw when the provider was unknown.
Drop the piModelOptions useMemo and pass undefined for Pi, mirroring
modelOptions.ts where the Pi branch already returns the current model
unchanged (no-op) when no custom options are supplied. Pi model changes
now go only through the dedicated provider-qualified picker (piModels).
* fix(pi): commit PiSession config only after Pi confirms the RPC
SetSessionConfig previously mutated piSession.currentModel /
currentProvider / currentThinkingLevel BEFORE awaiting
sendPiRpcAndWait(set_model / set_thinking_level). When Pi rejected the
value or the RPC timed out, the handler threw and the route returned
409, but PiSession kept the unconfirmed values; the 2s keepalive then
reported them back to the hub, where handleSessionAlive persisted a
model/effort Pi never accepted.
Resolve the requested model/effort into locals first, send the RPCs,
and only commit to PiSession after each await resolves. The null
(clear-model) path needs no RPC so it still commits immediately; the
unknown-provider path still throws without committing.
* fix(pi): apply startup model only after Pi confirms set_model
Two startup paths persisted the requested --model before Pi confirmed it:
1. handleGetState set session.currentModel = session.initialModel as soon
as get_state returned, using the unconfirmed startup model instead of
Pi's actual default. If the model was unavailable or rejected, the 2s
keepAlive reported it to the hub, which persisted/showed a model Pi
never accepted.
2. get_available_models then sent set_model fire-and-forget, so a Pi
rejection was never observed and currentModel stayed on the bad value.
Fix: handleGetState now reports Pi's real current model (newModel) while
a startup model is merely requested. get_available_models resolves the
provider from the cached list, awaits set_model, and commits
currentModel/currentProvider only on success — on rejection it logs and
keeps Pi's default. The await is fired detached so the
get_available_models RPC itself still resolves for ListPiModels.
* fix(pi): do not persist startup model before Pi confirms set_model
The startup --model still reached the hub unconfirmed via two paths the
previous Fix #13 left open:
1. bootstrapSession({ model: opts.model }) seeded the hub session model
at creation time, and SessionCache.handleSessionAlive persists every
non-undefined keepAlive model — so an unavailable/rejected model was
stored and shown before get_available_models/set_model ran.
2. PiSession constructor set this.currentModel = opts.model, so the very
first keepAlive (sent by startKeepAlive before any RPC confirms the
model) reported the unconfirmed value.
Pass model: undefined to bootstrapSession and start PiSession.currentModel
at null; opts.model is still captured as initialModel and applied/committed
only after get_available_models confirms it exists and set_model succeeds
(Fix #13). The hub now sees Pi's real current model from the first
get_state keepAlive and switches to the requested model only once accepted.
Also add sendPiRpcAndWait contract tests pinning the await<->resolve
symmetry (Fix #10): set_model/set_thinking_level/get_available_models must
resolve before timeout on a success response, and reject on a Pi error.
* fix(pi): apply startup effort only after Pi confirms set_thinking_level
runPi restored opts.effort straight into piSession.currentThinkingLevel
before startKeepAlive ran, and pushKeepAlive persists effort — so a
resumed/runner-spawned session could store/show a thinking level Pi
rejected or ignored. This is the effort analog of the startup-model
confirmation contract (Fix #13/#14).
Capture the requested effort into a local startupThinkingLevel instead of
mutating currentThinkingLevel up front. After transport.start() and the
get_state/get_available_models/get_commands sends, await set_thinking_level
and commit currentThinkingLevel + push a keepAlive only on success; on
rejection keep Pi's default (already reported by get_state). The await is
detached so the run loop is not blocked, and get_state is sent before the
set so its authoritative baseline lands first and cannot clobber the
confirmed value.
* fix(pi): omit unknown runtime config from keepalive, don't clear persisted state
Fix #14 changed PiSession.currentModel to start at null so the startup
--model was not leaked before confirmation. But the hub treats keepAlive
model:null as an explicit clear (sessionCache.ts only skips when the
field is undefined), so the first heartbeat (startKeepAlive runs before
get_state) now erased a resumed Pi session's persisted model/effort
before Pi reported its real state.
Distinguish "unknown" from "clear": currentModel/currentThinkingLevel
start undefined and keepAlive omits undefined fields (via
getKeepAliveRuntime), so the hub leaves persisted values alone until Pi
confirms. null remains an explicit clear and is still forwarded. Once
get_state/set_model/set_thinking_level confirm a value it is set and
reported normally.
* fix(pi): disable Ctrl/Cmd+M model cycler for Pi entirely
Fix #11 removed piModelOptions from availableModelOptions, assuming
getNextModelForFlavor('pi', model, undefined) was a no-op. It is not:
the Pi branch returns normalizeCurrentModel(model), i.e. the current
modelId as a bare string, so the shortcut still called onModelChange with
a bare modelId. That loses the provider and can pick the wrong cached
match, clear the model when session.model is empty, or hit 'provider is
not yet known'. Short-circuit the handler for Pi so model changes go only
through the dedicated provider-qualified PiModelPanel.
* fix(pi): persist piSelectedModel from get_state and startup set_model paths
Pi stores session.model as the bare modelId and relies on
metadata.piSelectedModel ({ provider, modelId }) to disambiguate
duplicate modelId values across providers in the web picker and
thinking-level filtering. But piSelectedModel was only written by the web
/sessions/:id/model path (hub persistPiSelectedModel). The runtime paths
that set currentModel/currentProvider — get_state, the startup
get_available_models set_model, and the set_model response — only
keepAlive'd the bare modelId, so a Pi session on Pi's default model,
resumed from CLI, or started with --model had no provider identity in
metadata and could render/filter against the wrong provider.
Add persistSelectedPiModel(session) (no-op unless both fields are known)
and call it after get_state, after a successful startup set_model, and
after the set_model response updates the fields. This mirrors what the
web picker already does.
* fix(pi): default startingMode to remote — Pi has no local TUI path
A terminal `hapi pi` launch defaulted to startingMode 'local' and marked
the session controlledByUser, but Pi only runs as `pi --mode rpc` with
piped stdio — there is no local terminal/TUI input path like Claude/Codex
have. The terminal user could not drive the session and the web treated
it as local-controlled, so the first terminal Pi session was stuck until
manually switched from the web.
Default to 'remote' so the session is immediately drivable from the web.
An explicit opts.startingMode (runner path) still takes precedence.
* fix(pi): resume with remote startingMode — no local TUI path
The previous Fix #19 changed the `hapi pi` default to remote, but
`hapi resume` still passed startingMode: 'local' into runPi for Pi
sessions, re-introducing the same unsupported local-control state on the
resume path: setControlledByUser publishes controlledByUser while Pi has
no terminal/TUI input, hiding/rejecting remote-only controls until a web
switch. Pass 'remote' here too and update the resume test accordingly.
* fix: restore e2e/scratchlist.spec.ts deleted from main by mistake
The earlier "remove unrelated E2E harness" commit (d1e5b4c) deleted the
whole e2e/ directory this branch had added, but scratchlist.spec.ts is a
main-branch Playwright spec (the only file under playwright testDir
./e2e). Its removal left `bun run test:e2e` with no tests to run while
the script and playwright.config.ts still point at that directory.
Restore scratchlist.spec.ts from main; the unrelated harness files
(HARNESS.md, harness.*, integration/*.mts) that were genuinely
branch-only additions stay removed.
---------
Co-authored-by: pi <pi@local>
Co-authored-by: Claude <noreply@anthropic.com>
|
||
|
|
3a8693f380 |
feat(cursor): migrate remote sessions to ACP with model/variant pickers (#799)
* feat(cli,web,hub): migrate Cursor remote sessions to ACP with model/effort pickers Move stream-json remote launcher to legacy path and add ACP launcher with set_config_option model/mode sync, optimistic keepalive on config changes, and shared catalog caching. Web gets dual base/effort Cursor pickers for session and new-session flows; hide composer status bar when Cursor sends no usage_update. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(cli,web,shared): Cursor model picker — ACP wires + CLI sku variants Enrich the web/mobile picker with agent --list-models SKUs grouped under ACP wire bases, fix session-open base highlight, and keep catalog discovery safe while the ACP transport holds the CLI lock. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(cursor-acp): apply ACP default model when web resets to Default Web sends model: null for Default; push session/set_config_option with the ACP default[] wire so Cursor backend matches hub state. Regression tests for setModel(null) and applyModelConfig(null). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(acp): clear stale agent-acp lock when owning process is gone Check lock pid with signal 0; remove orphaned lock dirs after SIGKILL or crash so listCursorModels can run cold probes again. Regression tests for guard and catalog discovery. Co-authored-by: Cursor <cursoragent@cursor.com> * test(cursor): use live pid for ACP lock handler tests Stale-lock cleanup clears dead pids; handler tests must simulate an active lock with the current process pid to avoid cold probes/timeouts. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(acp): scope agent CLI lock guard to Cursor agent command only Gemini/OpenCode/Kimi ACP sessions must not register agent-acp-active; that blocked listCursorModels while unrelated backends were running. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(hub,web): reject Cursor model changes for local sessions Hub returns 409 when controlledByUser is set, matching Codex. Web hides model and variant pickers for local Cursor sessions so users do not hit a dead RPC path. Document pre-push-review in AGENTS.md. Verified: bun typecheck; bun run test (919 cli + 243 hub + 768 web + 46 shared). Co-authored-by: Cursor <cursoragent@cursor.com> * fix(web): send stable ids for Cursor ask_question replies Parse and submit question.id and option.id so ACP receives keys like { approach: ['a'] } instead of index/label. Verified: bun typecheck && bun run test. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
hangerandGitHub
|
485bb46db8 |
fix(claude): propagate real contextWindow from SDK result to web (#720)
The "Default" model in NewSession sends no --model flag, so Claude CLI picks its own default (e.g. Opus 4.7 [1m] on Pro accounts). The web status bar then falls back to 200K - 10K headroom = 190K because the Claude SDK path never plumbs the real per-model contextWindow through to the wire-level `modelContextWindow`, unlike the ACP/Codex backends. Fix the gap in three places: - cli/src/claude/sdk/types.ts: declare optional `modelUsage` on SDKResultMessage to surface what Claude CLI already emits (`modelUsage[<model>].contextWindow`). - cli/src/claude/utils/sdkToLogConverter.ts: on system.init, capture the resolved model name (full form with `[1m]` suffix) and derive an initial contextWindow from the suffix. On every assistant message, inject the cached contextWindow into `usage.context_window` when absent. On result, refine the cache with the authoritative value from `modelUsage`. - web/src/chat/normalizeAgent.ts: forward `context_window` through the assistant usage normalization, so the existing reducer path (reducer.ts:175 → StatusBar.tsx:175) can render the real window. Closes #719. |
||
|
|
6f2bb7d32b |
feat(opencode): add plan mode, reasoning effort, and status telemetry (#688)
* feat(opencode): support plan mode * feat(opencode): support reasoning effort * feat(opencode): surface context usage in web Bridge OpenCode ACP usage updates into the existing token-count pipeline so the web status bar can show live context and cache information without a separate UI path. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(opencode): restrict plan mode to remote, rollback reasoning effort on failure - Block local OpenCode plan startup (tools not enforced in local path) - Allow remote OpenCode plan only (ACP permission handler denies tools) - Guard web /permission-mode endpoint for local OpenCode plan sessions - Rollback session reasoning effort when OpenCode rejects set_config_option - Wire rollback callback through opencodeLoop to runOpencode closure - Add tests: local plan rejected, remote plan allowed, web guard, effort rollback * fix(web): auto-retry OpenCode models query to populate model selector without refresh - Retry early failures (RPC may still be registering on new sessions) - Poll briefly until availableModels is non-empty - Stop polling once model options are discovered - Add tests for retry/poll/stop policy * fix(opencode): cap model discovery polling --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
|
c1eccc0de2 |
fix(web): guarantee unique assistant-ui thread message IDs (#706)
* fix(web): guarantee unique assistant-ui thread message IDs Suffix duplicate kind:id pairs before ExternalStore sync; skip duplicate hub rows in SessionChat normalization. Align outline scroll targets with the new user-text thread id shape. Closes #704. Co-authored-by: Cursor <cursoragent@cursor.com> * fix(web): reuse thread-id wrappers for useExternalMessageConverter cache WeakMap stable BlockWithThreadMessageId objects keyed on reconciled block refs so streaming appends do not invalidate assistant-ui converter caches (PR #706 review). Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: Cursor <cursoragent@cursor.com> |
||
|
聪andGitHub
|
9e35067aa9 | feat(web): show timestamps on chat messages (#664) | ||
|
|
763f45acdd |
feat: add support for Kimi Code CLI and fixed some bugs (#659)
* Add Kimi agent support via ACP protocol Add full integration for the Kimi Code CLI agent using the standard Agent Client Protocol (ACP). Includes: - kimi command and CLI registry wiring - Local launcher spawning kimi directly - Remote launcher with ACP stdio transport via AcpSdkBackend - Session management with resume support - Permission handler supporting all Kimi permission modes - Terminal UI display component - Runtime config resolving model from env and ~/.kimi/config.toml via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * Fix Kimi ACP tool call input decoding on web Kimi streams tool arguments as JSON text inside the content array (e.g. {\"command\": \"df -h\"}) instead of rawInput/kind. The handler now extracts input from three sources in priority order: 1. rawInput (Claude/Codex path) 2. kind + title fallback (Gemini path) 3. content JSON text (Kimi path) Also handles: - rawInput: null no longer blocks the kind+title fallback - Title prefixes like \"Shell: free -h\" are stripped to extract args - Stale placeholder inputs are re-derived when the title updates - Normalized kind aliases (shell, run, read_file, write, etc.) via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * Add kimi support to web UI * Fix some bugs * fix(kimi): dedupe repeated tool_call display in terminal UI * fix(web): keep tool block immutable so React detects input/state changes * fix(web): recognise Kimi subagent titles like 'Agent: ...' as subagent tools * fix(web): allow-for-session for ACP agents (kimi, cursor) PermissionFooter treated all non-codex sessions as Claude, sending Claude-specific acceptEdits/allowTools to ACP agents that don't support them. Hub rejected acceptEdits for kimi, and the ACP PermissionAdapter ignored allowTools. - Only show 'allow all edits' for Claude sessions - Send decision: approved_for_session for non-Claude ACP agents - Update status display to check decision field * fix(web): lookup subagent sidechains by tool-call id instead of msg id * fix(web): don't trim newest messages when loading older history fetchOlderMessages was using trimVisible(merged, 'prepend') which kept the oldest 400 messages and dropped the newest ones. This caused: 1. Latest messages to disappear when user loaded older history 2. User to see no visible change when new old messages were drowned in the 400-message window. Remove the incorrect trim so all fetched older messages are retained alongside the current window. Subsequent ingestIncomingMessages (append mode) will naturally keep the window bounded when new agent messages arrive. * fix(cli): route Kimi session resume to runKimi instead of runCursor Kimi was present in AGENT_FLAVORS but dispatchLocalResume had no branch for it, so resuming a Kimi session fell through to the Cursor launcher. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(cli): pass selected model to Kimi ACP backend via KIMI_MODEL env createKimiBackend was ignoring opts.model and only setting KIMI_PROJECT_DIR. Use buildKimiEnv so the selected model reaches the subprocess as KIMI_MODEL. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): bound message window on older loads with dedicated larger cap fetchOlderMessages was keeping all messages unbounded, causing sessionStorage bloat on repeated pagination. Reintroduce trimming with OLDER_LOAD_WINDOW_SIZE (800) so growth is capped while the newest messages are still preserved for far longer than before. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): revert sidechain lookup to message id, matching tracer/grouping pipeline tracer.ts sets sidechainId to the parent message id, and reducer.ts groups by sidechainId. A prior commit changed reducerTimeline.ts to look up by tool-call id (c.id), which broke sidechain attachment. Revert to msg.id so the lookup matches the actual grouping key end-to-end. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(cli): gate ACP title prefix stripping to known tool-kind labels extractTitleArgument stripped at the first colon unconditionally, corrupting commands/paths like curl http://localhost:3000 or Windows paths. Now it only strips when the prefix normalizes to the same tool kind as the event, verified via regex. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(shared): include kimi in isCodexFamilyFlavor for ACP permission UI Kimi is an ACP-style agent that supports the abort decision, but isCodexFamilyFlavor excluded it, so PermissionFooter rendered the non-Codex Allow/Deny UI without the Abort button. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> --------- Co-authored-by: HAPI <noreply@hapi.run> |
||
|
|
b06251bd53 |
fix: stream opencode reasoning updates (#661)
Emit throttled ACP reasoning snapshots with stable stream ids so OpenCode reasoning updates render live without one row per token. Collapse matching reasoning snapshots in the web timeline and avoid stale session 404 redirects during refetch. Tests: targeted Vitest suite and bun typecheck. Co-authored-by: twshe <twshe@outlook.com> |
||
|
SmallSpiderandGitHub
|
74e40b8a1a | fix(codex): stabilize goal status UI events (#652) | ||
|
SmallSpiderandGitHub
|
25631d971c | fix(codex): stabilize goal status updates (#651) | ||
|
weishu
|
0e594da84d | Render Codex review messages | ||
|
MapleStoryIdleandGitHub
|
86780e94d0 | feat(codex): preview generated images in chat (#630) | ||
|
weishu
|
089ddad476 | feat: support Codex goal slash command | ||
|
junesandGitHub
|
af3491e046 |
feat(web): group consecutive tool-use cards (#604)
* feat(web): group consecutive tool-use cards Add a web-only visible projection that groups consecutive root-level execution tools into expandable cards. Keep approval and question tools standalone, reuse older-history loading on expand, and add regression coverage for grouping and UI behavior. * fix(web): hydrate oldest visible tool group Mark needsOlderHistory on the first visible grouped tool run even when earlier visible blocks are non-tool content, and add regression coverage for the boundary. * fix(web): continue grouped history hydration Decouple ToolGroupCard older-history chaining from the shared loading flag, invalidate stale hydration runs safely, and add regression coverage for multi-page hydration. * fix(web): harden grouped tool hydration - retry incomplete group hydration after transient pagination contention\n- keep approved and denied permissioned tool cards eligible for grouping\n- cover both regressions with targeted web tests * fix(web): keep Codex permission cards standalone - treat CodexPermission as a semantic grouping boundary even after approval\n- keep permissioned execution tools groupable while preserving permission milestones\n- add regression coverage for Codex permission eligibility and boundary behavior * fix(web): narrow incomplete tool-group hydration - only mark groups at the oldest visible boundary as needing older history\n- avoid auto-paginating complete groups behind text, standalone tools, or permission milestones\n- add regression coverage for the adjacent boundary cases |
||
|
junesandGitHub
|
991b01cd10 |
fix(web): stabilize session history recovery (#593)
* fix(web): stabilize session history recovery * fix(web): split latest and older history generations Separate latest and older async guards in the message window store. Prevent refreshes from wedging load-more state and add a regression test for the overlap. |
||
|
SmallSpiderandGitHub
|
841b7cc035 |
Add Codex multi-agent timeline support (#588)
* checkpoint codex multiagent UI state * fix codex multiagent event scoping * fix: stabilize codex subagent timeline * chore: remove codex subagent nesting prompt * test(web): stabilize tool result rendering tests * fix: collapse codex agent trace rows by default * fix(web): keep chat scrolled to bottom * fix(web): backfill agent-run-heavy message loads * fix(codex): fail stuck subagent spawns * fix(codex): preserve wrapped child event scope * fix(codex): surface agent tool completions |
||
|
Junmo KimandGitHub
|
e17d7e5995 | fix(web): align Agent tool dialog with TUI ctrl+o expand (#585) | ||
|
Junmo KimandGitHub
|
d9d7ed6699 |
feat(web): show message metadata (invoke time, duration, model) on click (#555)
* feat(web): show message metadata (invoke time, duration, model) on click
* fix(cli): preserve model field on assistant messages forwarded to hub
`RawMessageSchema` validates the `message` object in Claude Code session
JSONL lines before the cli forwards each message to the hub. Zod's default
parse mode strips fields that the schema does not declare, so the
`message.model` value (e.g. `claude-sonnet-4-6`) was silently removed
before the message reached the hub. The web normalizer reads
`data.message.model` to label assistant blocks, so without this field
every assistant message fell back to a generic "AI Model" label —
defeating the per-message model attribution this PR adds.
Add `model` to `RawMessageSchema` so it survives parse and reaches the
hub intact.
* fix(web): drop dead model shorthand in result envelope normalize
The `result/success` branch in `normalizeAgentRecord` referenced a `model`
identifier that was never declared in the function scope, breaking
`bun typecheck`. The reducer that consumes the resulting `turn-duration`
event does not look at `model` on the event itself, so the shorthand was
dead code. Remove it to restore typecheck.
* refactor(web): simplify turn-duration matcher with findLastIndex
Replace the imperative reverse-scan loops in the `turn-duration` reducer
branch with `findLastIndex`. The previous fallback also had an awkward
double-loop that mutated the matched block in place; using an index plus
a single immutable update keeps the block reference clean and makes the
match priority (id-prefix > tool-call id > last assistant-like) explicit.
Behaviour is unchanged — existing reducer tests cover both the messageId
match and the fallback paths.
* fix(web): preserve per-message model across mid-session model switches
The metadata footer fell back to `Session.model` from chat context when a
message did not carry its own `model`. That session value mutates when
the user switches models mid-session, so older messages were relabeled
with the latest model — including Codex/local assistant paths
(`AGENT_MESSAGE_PAYLOAD_TYPE`) that don't populate `msg.model`.
Drop the mutable-context fallback: pass `messageModel ?? null` to
`MessageMetadata` and let it omit the model line when no per-message
value is available. This is correct behaviour for messages whose
producer didn't record a model, and avoids ever attributing a message
to a model that didn't generate it.
Also remove the now-unused `useHappyChatContext` import in this file.
Add reducer invariants to lock in the data flow:
- `preserves per-message model across mid-session model switches`
- `leaves model undefined when message lacks per-message model`
* fix(web): keep tool-block reference identity when applying turn-duration
`ensureToolBlock` stores the same `ToolCallBlock` instance in both
`toolBlocksById` and `blocks`. The earlier refactor cloned the matched
block via `blocks[foundIndex] = { ...b, durationMs }`, which left the
map pointing at the stale original. A subsequent permission/result
mutation through `ensureToolBlock` would then update the stale map
object while the rendered `blocks` entry never sees the completion or
result, causing tool cards to miss state transitions.
Mutate the matched block in place instead — same in-place pattern the
reducer used before — and gate the assignment on the kinds that carry a
`durationMs` field so TypeScript narrows correctly.
Add an invariant test that fires a `turn-duration` event at a tool-call
block and asserts the rendered block and `toolBlocksById.get(...)`
remain the same object reference.
* fix(web): do not render service_tier as the model id
`MessageMetadata` previously fell back to `usage.service_tier` as the
"model" when no per-message `model` was available, so messages without
their own model id could surface labels like `Model: standard_only` —
service_tier is tier metadata, not a model.
Render the model line only when a real `model` is present; if a
non-`standard` `service_tier` is the only signal, surface it as a
separate `Tier: <tier>` label so it is not mistaken for the model.
The standard tier is the implicit default and is never rendered alone.
Extract the label-building logic into `buildMessageMetadataLabels` so
it can be unit-tested without a DOM. Add tests covering: model present,
model missing with non-standard tier, default standard tier, model with
non-standard tier appended, and the empty-input case.
* fix(web): metadata toggle ignores clicks on nested interactive controls
The bubble-level click handler that opens the metadata footer wraps
interactive descendants — tool-card buttons, retry buttons, dialog
triggers (Radix `role="button"`), and the Markdown code-copy button.
Clicking any of those flips the metadata footer as a side effect, even
when the descendant is the actual target of the user's intent.
Extract the closest-ancestor check into a small `metadataToggle` helper
and route both `AssistantMessage` and `UserMessage` click paths through
it. The toggle bails out when the click target sits inside any
`button`, `a`, `input`, `textarea`, `select`, or `[role="button"]`
ancestor; plain message-body text still toggles as before.
Add unit tests covering: button target, nested span inside a button,
`role="button"` Radix-style trigger, anchor/input/textarea/select form
controls, plain message-body text (no toggle), and a non-HTMLElement
target.
* fix(cli): preserve messageId on system/turn_duration record
`web/src/chat/normalizeAgent.ts` matches each `turn-duration` event to
the assistant block carrying the same `data.messageId`. Claude code
emits that field on the `system/turn_duration` record, but
`RawJSONLinesSchema`'s system branch did not declare `messageId`, so
Zod stripped it before the cli forwarded the record to the hub. The
matcher then fell back to "the last visible block", which is wrong for
interleaved/tool-heavy turns and silently attaches the duration to the
wrong assistant block.
Add `messageId: z.string().optional()` to the system schema so the id
survives parse and reaches the web reducer. Tests cover the preserved
case, the legacy case without `messageId`, and the previously-fixed
`message.model` case so Zod strip regressions on adjacent fields stay
locked in.
* fix(web): metadata toggle accepts SVG event targets
`isClickOnNestedControl` only walked up via `closest` when the click
target was an `HTMLElement`. The copy / retry / Markdown code-copy
buttons render SVG icons, so clicking the icon makes the event target
an `SVGElement` (not an `HTMLElement`) — the guard returned false and
the bubble-level click flipped the metadata footer anyway.
Widen the type check to `Element`, which is the common super-class of
both `HTMLElement` and `SVGElement` and also exposes `closest`. Plain
text targets and non-Element targets still behave as before.
Add a regression test that mounts an icon-only button (`<button><svg>
<path/></svg></button>`) and asserts both the `<svg>` and `<path>`
targets walk up to the enclosing button.
* refactor(cli): rely on Zod passthrough for jsonl envelopes
`RawMessageSchema` and the `system` branch of `RawJSONLinesSchema` were
declared with Zod's default `strip` mode, so any field the cli did not
explicitly enumerate was silently dropped before the hub forwarded the
record. The metadata pipeline lost `message.model` and
`system/turn_duration.messageId` exactly that way, and each gap took a
separate fix.
Switch both schemas to `.passthrough()` so undeclared fields survive
parse and reach the web reducer verbatim. Future SDK additions no
longer require another schema patch.
Add tests asserting that unknown keys on assistant messages and
unknown keys on system records (alongside the existing `messageId`
case) are preserved end-to-end through the schema.
* refactor(web): clean up dead metadata propagation surface
Several knobs were added to thread metadata through the chat tree but
ended up unused or redundant; consolidate them so the data flow has a
single canonical path.
- Drop the unreachable `data.type === 'result' && data.subtype ===
'success'` branch in `normalizeAgentRecord`. Claude's `result`
records are consumed by `claudeRemote` as session-completion signals
and never forwarded to the hub; the cli `RawJSONLinesSchema`
discriminator does not include `result`, so these records are
rejected before they reach `normalizeAgentRecord` either way.
- Stop threading `invokedAt` through the inner `normalizeAssistantOutput`
/ `normalizeUserOutput` / `normalizeAgentRecord` calls. Every caller
in `normalizeDecryptedMessage` already overwrites it via the outer
spread, so the inner copies were dead writes. Set `invokedAt` only at
the outer boundary.
- Remove the `model?: string | null` field from `HappyChatContextValue`
and the `model` prop on `HappyThread` / `SessionChat`. Its only
consumer (`AssistantMessage` mutable-fallback) was removed when the
per-message model attribution fix landed; the prop has no readers
now.
- Match the existing `as Partial<HappyChatMessageMetadata> | undefined`
cast pattern in `AssistantMessage` and `UserMessage` instead of the
non-`Partial` cast that pretended every field was present even when
`custom` is undefined.
- Rename `AgentEvent.turn-duration.messageId` to `targetMessageId` so a
reader does not confuse the duration's target with the surrounding
envelope id; the wire field on Claude's `system/turn_duration` record
stays `messageId` (vendor name) and is mapped at the normalize
boundary.
No behaviour change. All existing tests pass.
* fix(web): turn-duration matcher and cli-output merge precedence
Two reducer-level metadata-correctness bugs surfaced during a hostile
self-review.
1. Turn-duration matcher silently dropped the duration when
`targetMessageId` resolved to a non-duration-bearing block. The
existing pipeline did `findLastIndex(b => b.id === targetId || ...)`
first; if that hit an `agent-event` or `user-text` block (id-prefix
collision), the kind guard at the assignment site failed and the
duration was never attached. The fallback search ran only when the
first pass returned -1, not when the kind check rejected the match.
Fold the kind filter into every search predicate via a typed
`isDurationTarget` helper so the priority `target-bearing match >
tool-call id > last duration-bearing block` is exhaustive.
2. `mergeCliOutputBlocks` had asymmetric metadata precedence between
the command-name block (`prev`) and the stdout follow-up (`block`):
`invokedAt` and `model` preferred prev, but `durationMs` and `usage`
preferred block. Only the command-name block carries first-class
metadata; the stdout follow-up is a synthetic split. Use prev as the
primary source uniformly and fall back to block only when prev is
missing the field.
Tests cover the fallback path on the matcher and both precedence
scenarios on the merger.
* fix(web): preserve tool-call invokedAt across tool-result update
`ensureToolBlock` is called twice for the same tool: first with the
seed from the assistant's tool-use block, then with the seed from the
matching tool-result message. The second call's `seed.invokedAt` came
from the tool-result message and was unconditionally overwriting the
tool-call's original invokedAt. The rendered "Invoke" timestamp on a
tool card therefore showed when the result was processed, contradicting
the column header.
Guard the assignment so the timestamp survives the second call —
`existing.invokedAt ??= seed.invokedAt` semantics — while still letting
the first call set the value when the tool is created. `durationMs`,
`usage`, and `model` continue to overwrite because their values come
from the result message's usage block and are intentionally newer.
Add a regression test that fires a tool-use followed by a tool-result
with a later invokedAt and asserts the tool block keeps the original.
* fix(web): metadata footer UX, accessibility, and label hardening
Bundle the remaining UI surface fixes for the metadata footer.
- Make the bubble interactive only when there is metadata to disclose.
Without the guard, every non-Claude session bubble (Codex / Cursor /
Gemini, none of which populate `model`/`usage`/`durationMs` in the web
layer) showed a pointer cursor and reacted to clicks even though
`MessageMetadata` rendered nothing — false-positive interactivity.
- Add keyboard support: when the bubble is interactive it now exposes
`role="button"`, `tabIndex=0`, `aria-expanded`, and an `Enter`/`Space`
key handler so screen readers and keyboard-only users can disclose
the footer the same way mouse users do.
- Fix nullish-vs-falsy bugs in the label builder: a 0 ms turn or a 0
unix-epoch invokedAt no longer hides their lines. Use explicit
`!= null` / `>= 0` checks.
- Rename the token total to "billable tokens" so the explicit exclusion
of cache I/O is signalled in the label rather than implied by the
number alone.
- Tag the queued/sending status spans with `role="status"` (and an
accessible label) so they are announced by AT and so the metadata
toggle's `closest('button, ..., [role="status"]')` filter does not
accidentally fire when a user clicks a status icon.
- Add the native `<summary>` element and `[role="status"]` to the
toggle's nested-control selector. Tool cards already render their
expandable bodies as `<details><summary>` — clicking the summary now
expands the disclosure without also flipping the metadata footer.
Tests cover: native `<summary>` target, `role="status"` target, the
billable label, durationMs=0 surfaced, invokedAt=0 surfaced,
invokedAt=null/undefined hidden.
* fix(web): expose cli-output metadata via dedicated toggle button
CliOutputBlock renders the entire card as a Dialog trigger <button>, so
the bubble-level click handler on the cli-output branch never opened the
metadata footer by mouse — every click landed inside that button and
isClickOnNestedControl bailed out. The wrapping div with role="button"
was also a nested-interactive a11y anti-pattern.
Drop the wrapper's role/onClick/tabIndex/keyDown on the cli-output
branch and render an explicit "Show metadata" / "Hide metadata" button
beneath the card. The dialog trigger keeps its full hit area; the
metadata footer is now reachable by both mouse and keyboard.
* fix(web): exclude toggle wrapper from nested-control guard
The bubble-level toggle wrappers in AssistantMessage / UserMessage
carry role="button" for keyboard accessibility. Without excluding
currentTarget, closest('[role="button"]') from any inner click matches
the wrapper itself and the toggle bails out — making the metadata
footer unreachable for mouse users (keyboard Enter/Space still worked,
which is why unit tests and bot review missed it).
Walk currentTarget out of the match: a nested control is one whose
closest matching ancestor is *not* the wrapper itself.
* fix(web): apply nested-control guard on keyboard activation too
The mouse path bailed via isClickOnNestedControl, but the keyboard path
on the metadata-toggle wrapper did not. Pressing Enter or Space on a
focused descendant control (e.g. Markdown code-copy button) bubbled the
keydown up and the wrapper toggled metadata alongside the descendant's
own activation.
Generalize the helper to isNestedInteractiveEvent over both
MouseEvent and KeyboardEvent and call it from onMetadataKeyDown in
AssistantMessage and UserMessage.
|
||
|
CoColateandGitHub
|
e76738aa5a |
fix(codex): improve web rendering for Codex events (#544)
* test(codex): add web event rendering harness * fix(codex): surface plan updates in web * fix(codex): render MCP tool calls in web * fix(codex): improve terminal and context display * fix(codex): format token usage events * fix(codex): show status context in web * fix(codex): preserve tool result errors |
||
|
NightWatcher314andGitHub
|
d66547ff46 | Add web conversation outline (#534) | ||
|
|
e7ba48c761 |
fix(web): normalize non-sidechain text-only array user output as user message (#409)
The CLI wraps array-content user messages as agent output because isExternalUserMessage rejects non-string content. On the web side, detect text-only arrays in non-sidechain user output and emit them as role:'user' so they display in the user lane. Also handle sidechain user messages with mixed array content (e.g. tool_result + text) by extracting text parts into a sidechain block. Closes #407's original scope on top of the #402 base. via [HAPI](https://hapi.run) Co-authored-by: HAPI <noreply@hapi.run> |
||
|
|
73fa846df3 |
fix(web): drop "No response requested." assistant messages (#402)
* fix(web): drop "No response requested." assistant messages When Claude Code injects system messages (task notifications, system reminders) as user turns, Claude responds with "No response requested." In the HAPI web UI this appears as a reply to the user's message, making it look like Claude is ignoring their input. Filter these out in isSkippableAgentContent() (catches the fallback path in normalize.ts) and in normalizeAssistantOutput() (catches the primary path). Both checks verify the assistant message contains only the text "No response requested." with no tool calls. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): filter text block instead of dropping entire message Address bot review: dropping the whole normalized record breaks sidechain UUID threading (parentUUID chain orphans). Instead of returning null, suppress only the "No response requested." text block during content extraction. The message record (uuid, parentUUID, usage) is preserved so the tracer's sidechain grouping continues to work. Also remove the isSkippableAgentContent check since we no longer need to drop the message at that layer. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): move "No response requested." filter to reducer layer Address bot review: filtering in normalizeAssistantOutput() produced empty content arrays, breaking traceMessages() which reads uuid and parentUUID from content[0]. Sidechain child messages whose parentUUID pointed to the filtered message became orphaned. Fix: revert the normalizer to always emit the text block (preserving the UUID chain for the tracer), and filter the sentinel text in reducerTimeline.ts where text blocks become visible AgentTextBlocks. At this point tracing is already complete. Also adds reducer-level tests for the filter and updates the normalize test to verify the text block is preserved. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): scope sentinel filter to single-block assistant messages only Address bot review: the previous filter suppressed any text block matching "No response requested.", which could hide legitimate replies. Now the filter only triggers when the message has exactly one content block (msg.content.length === 1) — i.e., the assistant response is purely the sentinel text with no tool calls or reasoning blocks. This prevents false positives while still catching the system-injection auto-reply case. Add test for the multi-block case (text + tool call) to verify the sentinel text is preserved when other content exists. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): add parentUUID structural check to sentinel filter Address bot review: raw text match alone could theoretically suppress a legitimate reply. Add c.parentUUID !== null as a structural guard: - Sentinel auto-replies always follow a prior assistant turn, so their parentUUID is set (pointing to the previous message in the chain). - A first message in a conversation has parentUUID: null and will never be filtered. Combined conditions: msg.content.length === 1 (sole block, no tool calls) AND c.parentUUID !== null (not the first reply) AND exact text match. Add tests for the parentUUID=null escape hatch. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): use injected-turn UUID tracking for sentinel filter Address review: use structural markers instead of broad text matching. 1. Pre-scan collects UUIDs from sidechain content blocks (system- injected user turns). The sentinel filter now only triggers when parentUUID points to one of these known injected turns. 2. Move task-notification event extraction from normalizer to reducer. Previously, task-notifications with summary were normalized as role:'event', losing their uuid. Now they stay as sidechain (uuid preserved for pre-scan), and the reducer extracts the summary as an agent-event block. 3. Remove redundant 'uuid' in c guard (always present on sidechain type). False positive analysis: a legitimate reply is only suppressed when ALL of: (a) sole content block, (b) parentUUID matches a sidechain-injected turn, (c) exact sentinel text. This combination cannot occur for real user-facing content. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): add parentUUID to sidechain content type for tracer linkage The sidechain content block was missing parentUUID, so traceMessages() could not chain system-injected user turns (task notifications, system reminders) inside a Task sidechain back to their parent. This caused later sidechain messages pointing to the injected turn's UUID to become orphaned and disappear from the Task card. Add parentUUID to the sidechain type definition and propagate it from normalizeUserOutput() in both the isSidechain and non-sidechain paths. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * fix(web): handle array-content sidechain user messages to prevent prompt leak Sidechain user messages can arrive with either string content or array content ([{type:'text', text:'...'}]) depending on how Claude Code serialises them. The previous fix only handled the string case, causing intermittent prompt leaks when array format was used. Now normalizeUserOutput extracts text from array-content sidechain messages and emits them as sidechain blocks, so the tracer can match them to their parent Task tool call. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> * test(web): verify parentUUID propagation from assistant output data Add integration tests confirming normalizeAssistantOutput correctly maps data.parentUuid to text block parentUUID (used by the reducer's sentinel detection). Tests cover both present and absent parentUuid. via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> --------- Co-authored-by: HAPI <noreply@hapi.run> |
||
|
Junmo KimandGitHub
|
00ba610ab0 |
feat: display rate limit warnings instead of raw JSON (#388)
* refactor(web): extract normalizeTimestamp helper in presentation
Extract the shared seconds-vs-milliseconds normalization logic into
a private `normalizeTimestamp()` helper. No behavior change —
`formatUnixTimestamp()` produces identical output.
* refactor(web): return AgentEvent from parseClaudeUsageLimit
Change return type from `number | null` to `AgentEvent | null` so
the caller doesn't need to construct the event object. No behavior
change — the same `limit-reached` event is produced.
* feat(cli): convert rate_limit_event to standardized text format
Parse undocumented Claude `rate_limit_event` JSON in the CLI adapter
layer (AcpMessageHandler) before it reaches the web.
Converted text format (pipe-delimited):
- "Claude AI usage limit warning|{ts}|{pct}|{rateLimitType}"
- "Claude AI usage limit reached|{ts}|{rateLimitType}"
Status handling:
- `allowed_warning` → warning text with utilization and limit type
- `rejected` → reached text with limit type
- `allowed` → silently suppressed (noise)
- unknown statuses → passed through as-is (forward-compatible)
* feat(web): display rate limit warnings with limit type
Parse standardized pipe-delimited text from the CLI adapter into
`limit-warning` and `limit-reached` events, displaying the rate
limit type (5-hour, 7-day) when available.
- `limit-warning`: "⚠️ Usage limit 90% (5-hour) · resets 2:00 PM"
- `limit-reached`: "⏳ Usage limit reached (5-hour) until 4/2/2026"
- Backward compatible: `limit-reached` without limitType still works
The `reached` regex uses `(?:\|([^|]*))?$` to optionally match the
limitType field, maintaining compatibility with the existing format.
* refactor(cli): move rate limit parsing out of flushText
Remove rate limit detection from flushText() back to plain buffer
flush. The next commit will re-add parsing at the chunk level
(handleUpdate) where it can intercept before buffer merging.
Includes failing tests that demonstrate the mixed-chunk bug:
when a rate_limit_event chunk arrives in the same turn as normal
text, the JSON leaks into the merged buffer.
* fix(cli): intercept rate_limit_event at chunk level, not flush
Move rate limit detection from flushText() to the agentMessageChunk
handler so it fires before the chunk enters the shared text buffer.
Previously, a rate_limit_event chunk arriving in the same turn as
normal text would merge into bufferedText and leak as raw JSON.
Now the chunk is intercepted individually, the existing buffer is
flushed first (preserving prior text), and the converted message
is emitted separately.
* fix(cli): skip flush when suppressing allowed rate_limit_event
Only flush the text buffer when the parsed event will actually be
displayed. Suppressed events (e.g. status: 'allowed') now return
immediately without flushing, preventing a text → allowed → text
sequence from splitting one answer into two agent-text blocks.
* fix(web): include limitType in limit-reached reconcile key
Without this, reprocessing a message from the old format (no
limitType) to the new typed format reuses the stale block and
the (5-hour)/(7-day) suffix never appears.
|
||
|
Haoqing WangandGitHub
|
36022a0a1a |
fix(web): filter system-injected XML tags from rendering as raw text (#387)
* fix(web): filter system-injected XML tags from rendering as raw text Claude Code injects internal messages (<task-notification>, <system-reminder>, <command-name>, <local-command-caveat>) as user-role messages. The web UI was rendering these as raw XML text visible to users. - Parse <task-notification> and display as agent-event with summary text - Silently drop <system-reminder>, <command-name>, <local-command-caveat> - Add tests covering all injection prefixes and edge cases * fix(web): scope system injection filtering to Claude sessions only Address review feedback: the XML tag filtering was applied at the generic timeline layer, which could incorrectly hide legitimate user messages in Codex/Gemini sessions. - Add isClaudeSession flag threaded from Session.metadata.claudeSessionId - Only filter system-injected tags when isClaudeSession is true - Add tests verifying non-Claude sessions pass through all messages * fix(web): treat all string user output as sidechain to prevent prompt leaks Restores the fix from 3cf96ab that was accidentally reverted in 2205e04. In normalizeUserOutput(), string-content user messages arriving through the agent output path are never real user input (real user text goes through normalizeUserRecord). Previously, non-sidechain string messages were emitted as role:'user', causing subagent prompts and system-injected messages to render as user text in the web UI. Now all string-content user messages in this path are: - <task-notification> with summary → converted to role:'event' - Everything else → marked as sidechain (matched to parent Task tool call by the tracer, or harmlessly skipped by the reducer) This provides a root-level fix that prevents ANY string user message from the agent output path from leaking as visible user text. * ci: retrigger CI * fix(web): remove superseded return-null filter from upstream PR #372 The upstream `return null` filter for <task-notification> and <system-reminder> (from PR #372) is now superseded by the comprehensive sidechain upgrade logic. Remove it to avoid short-circuiting the new task-notification → event conversion. * refactor(web): remove reducer-side system injection filtering System-injected messages are now fully handled in normalizeUserOutput() (normalize layer), so the redundant filtering in reduceTimeline() is no longer needed. Removing it also eliminates the risk of accidentally hiding legitimate user messages that happen to start with XML tags. - Remove SYSTEM_INJECTION_PREFIXES, isSystemInjectedMessage, parseTaskNotificationSummary from reducerTimeline.ts - Remove isClaudeSession plumbing from reducer.ts and SessionChat.tsx - Simplify reducerTimeline.test.ts to only test pass-through behavior |
||
|
Haoqing WangandGitHub
|
2c20b04bec | fix(web): suppress Task tool prompt text from leaking into chat (#372) | ||
|
Haoqing WangandGitHub
|
4c11fbe5f2 | fix(web): stop rendering sidechain prompt as user message (#340) | ||
|
weishu
|
a02f908dc5 | fix test | ||
|
weishu
|
32f05d99a0 | Release version 0.16.3 | ||
|
Junmo KimandGitHub
|
d76b1a6ac0 | refactor: introduce model-agnostic agent interfaces (#323) | ||
|
weishu
|
329d28a93c | remove , using instead | ||
|
weishu
|
ce5edc42e1 | Add support for claude 1M context. close #291 | ||
|
|
2f6bbcd400 |
fix(web): 修复切换会话时消息乱序问题 (#150)
* feat(web): add built-in Nerd Font for terminal icon support Add MesloLGLDZ Nerd Font as a built-in web font to ensure Powerlevel10k and other Nerd Font-dependent prompts render correctly in the web terminal, even on devices without local Nerd Fonts installed (e.g., mobile phones). Changes: - Add MesloLGLDZNerdFontMono-Regular.woff2 to web/public/fonts/ - Load font via FontFace API with BASE_URL support for subpath deployments - Merge font loading and terminal creation in same useEffect to ensure font is ready before terminal renders Fixes #121 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * refactor: use CDN for Nerd Font instead of bundled file Load MesloLGLDZ Nerd Font from jsDelivr CDN instead of bundling the 1.2MB font file. This reduces the bundle size significantly while still ensuring Nerd Font icons display correctly on all devices. CDN URL: https://cdn.jsdelivr.net/gh/mshaugh/nerdfont-webfonts@v3.3.0/build/fonts/MesloLGLDZNerdFontMono-Regular.woff2 Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com> * fix(web): prevent message ordering issues when switching sessions When switching between sessions or loading paginated messages, old tool blocks from agentState.requests were being mixed with current messages, causing display order issues. Root cause: reduceChatBlocks creates tool blocks for all permissions in agentState, including those older than the current message page. These old blocks (with earlier createdAt) would appear mixed with newer messages. Fix: 1. In reducer.ts: Skip creating permission-only tool blocks if their createdAt is older than the oldest message in the current view 2. In SessionChat.tsx: Clear caches synchronously in useMemo when session changes, since useEffect runs after render Closes #148 via [HAPI](https://hapi.run) Co-Authored-By: HAPI <noreply@hapi.run> --------- Co-authored-by: tfq <tfq@gmail.com> Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com> Co-authored-by: HAPI <noreply@hapi.run> |