Keyed preamble commentary was appended after the whole tool loop, so it relied solely on the final visible-time sort for placement and lost the insertion-order tiebreaker against tool cards. Splice each keyed commentary segment into the items list before the first item with a strictly-later timestamp, so a preamble that arrived before a later tool renders above that tool while the run is live (not only after final materialization). Tools sharing the commentary timestamp that are already visible stay above it. Adds a buildChatItems regression covering a keyed preamble between two tools.
Add a per-viewer 'Keep commentary' toggle (UiSettings.chatPersistCommentary,
default true) that controls whether keyed Codex preamble/commentary blocks
stay after the final answer or clear with it.
- Persist (default): keyed commentary materializes as durable blocks, current
behavior, existing proof unchanged.
- Transient (toggle off): commentary stays live during streaming but is never
materialized, so it disappears as the final message arrives. This is the
transient-only behavior from #92236, now user-selectable instead of a
maintainer-level either/or policy choice.
Single gating point in materializeVisibleStreamState (skip itemId-keyed parts
when persistCommentary is false); threaded from settings through the chat
event handler. Adds desktop + mobile header toggles and an en.ts label
(locale bundles regenerated via ui:i18n:sync, English fallback).
Tests: reconciliation persist/transient coverage, final-event handler honors
the setting, settings round-trip + header button assertions updated.
* fix(system-prompt): move exec-approval + Authorized Senders below cache boundary
buildExecApprovalPromptGuidance (channel-varying: CLI /approve vs native
approval UI) and buildUserIdentitySection / "## Authorized Senders"
(owner/identity-varying, dropped when minimal) were emitted into the static,
cacheable prefix *above* SYSTEM_PROMPT_CACHE_BOUNDARY. They fork the cacheable
prefix at ~token 1,460, invalidating client-side prefix caching for the rest of
the ~17.8K-token system prompt — causing minutes-long cold prefills on local
models (llama.cpp / MLX / Ollama) after any channel-varying cron/heartbeat turn.
Follow-on to #40256, which moved Messaging/Voice/Reactions below the boundary
but missed these two. Relocates both into the existing below-boundary
channel-guidance block. Pure cache-stability change with no behavioural change
(the guidance is position-independent for correctness). Extends the boundary
test to assert both sections sit below SYSTEM_PROMPT_CACHE_BOUNDARY.
Measured on a local deployment: shared cross-channel prefix grew ~1,460 ->
~15,486 tokens; post-cron interactive turns went from minutes to ~10s.
Fixes#98261
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H6Hz9UEpxQ4W3d8XecvupH
* fix(system-prompt): suppress relocated exec-approval line under tool_call_style override
Addresses review (clawsweeper): the exec-approval guidance lived inside the
`tool_call_style` fallback, so a provider override of that section previously
replaced it. Relocating it below the boundary emitted it unconditionally, which
changed behaviour for providers/plugins that override `tool_call_style`. Gate the
relocated line on the absence of a `tool_call_style` override, restoring the
original "override replaces the whole section" contract. Extends the
provider-override test to assert the default approval line is suppressed when
the section is overridden.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01H6Hz9UEpxQ4W3d8XecvupH
* fix(system-prompt): tighten cache boundary proof
* fix(system-prompt): tighten cache boundary proof
* fix(system-prompt): tighten cache boundary proof
---------
Co-authored-by: headbouyJB <23249268+headbouyJB@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* fix(device-pairing): don't churn requestId on subset re-requests
A reconnect that re-requested a subset of an already-pending device
pairing request still superseded it with a fresh requestId. This is the
root cause of the "unknown requestId" failures during device approval:
1. A TUI connect files a broad scope-upgrade request; the owner copies its
id from `openclaw devices list`.
2. `openclaw devices approve <id>` reconnects as a CLI probe that only needs
`operator.pairing` — a subset of the pending scopes.
3. That subset re-request superseded the pending request with a new id, so
the originally-listed id no longer existed and approve failed.
Refresh the existing request in place when the incoming request only asks
for roles/scopes a single pending request (same device key + role) already
covers. Escalations that request *more* than the pending request still
supersede with a fresh id, so the requestId stays bound to at least the
scope snapshot the owner saw (the existing security-motivated behavior and
its test are preserved).
AI-assisted (Claude Code).
* fix(device-pairing): align subset pairing with scope coverage
* test(telegram): add regression test for forum topic message_thread_id with streamed reasoning
After thorough code tracing of all delivery paths (draft stream, durable,
non-durable, preview hooks), the current main branch already correctly
passes message_thread_id through all paths for the streaming reasoning +
final answer scenario in forum topics. The bug reported in #89352 may have
been resolved by earlier merged changes (media-path preservation,
draft/progress streaming, etc.).
This commit adds a focused regression test covering the combined
scenario to prevent future regressions.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(telegram): assert forum thread at draft-stream boundary for both lanes
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(memory): detect unindexed session transcripts in status mode (fixes#97814)
The status-purpose MemoryIndexManager init skips ensureSessionStartupCatchup(),
so openclaw memory status reports dirty=false while unindexed session files
exist on disk. This is a false-clean state: the operator sees no backlog,
but memory search silently misses unindexed transcripts.
Fix: run markSessionStartupCatchupDirtyFiles() in status mode. It checks
for on-disk session files without corresponding memory_index_sources rows
and sets sessionsDirty=true if found, without triggering a full sync.
The full catchup+sync runs on the next non-transient manager cycle (CLI
sync or normal runtime init).
Co-Authored-By: Claude <noreply@anthropic.com>
* fix(memory): await status dirty detection before status() reads the flag (fixes#97814)
The status-purpose MemoryIndexManager init skips ensureSessionStartupCatchup(),
so openclaw memory status reports dirty=false while unindexed session files
exist on disk. This is a false-clean state: the operator sees no backlog,
but memory search silently misses unindexed transcripts.
Fix: move status-mode markSessionStartupCatchupDirtyFiles() from the
constructor (fire-and-forget via void) into the create() factory method
where it is awaited. This guarantees sessionsDirty is set before any
caller reads manager.status(). The detection does NOT trigger a sync,
so status remains lightweight.
Review: verified manually — unit tests 21/21 pass, compilation 0 errors.
* test(memory): add regression test for status-purpose dirty detection (fixes#97814)
Adds a regression test that exercises the actual purpose:'status' manager
flow: create a session file without index row, create status manager, verify
status().dirty is true. This addresses the ClawSweeper P1 finding requesting
a test that exercises the real status path, not only the protected helper.
Also fixes the timing issue: move dirty detection from constructor (void)
to create() factory (await), guaranteeing sessionsDirty is set before any
caller reads manager.status().
---------
Co-authored-by: Claude <noreply@anthropic.com>
* fix(utils): keep reply directive ids unicode-safe
* test(utils): catch trailing lone surrogate in hasUnpairedSurrogate helper
Per PR #96938 review: the test helper missed a high surrogate at end of
string because charCodeAt(out-of-bounds) returns NaN, and NaN comparisons
are always false. Guard bounds explicitly so the truncation test actually
proves what it claims, plus two cases pinning helper behavior.
* chore(utils): rerun QA smoke to confirm memory-index timeout flake on #96938
---------
Co-authored-by: ly-wang19 <ly-wang19@users.noreply.github.com>
Two-step decode in decodeLiteralEscapes:
- Step 1: regex anchored to high-surrogate range (U+D800–U+DBFF) so a
preceding BMP escape (e.g. \u0041) cannot consume the high-surrogate
half of a valid pair like \uD83D\uDE00 (😀), leaving \uDE00 lone.
- Step 2: decode remaining BMP codepoints; preserve lone surrogates as
six-character literals instead of corrupting them to U+FFFD in the
outbound IRC UTF-8 stream.
* fix(discord): bound happy-path API response reads to prevent OOM
Replace the unbounded res.text() call in requestDiscord's success path with
readResponseTextLimited capped at 4 MiB. Discord channel message lists and
attachment payloads can accumulate to large sizes; without a cap the process
can exhaust available memory. The error path already used readResponseTextLimited
with DISCORD_API_ERROR_BODY_LIMIT_BYTES — this applies the same guard to the
happy path using a separate DISCORD_API_RESPONSE_BODY_LIMIT_BYTES constant
sized appropriately for valid API payloads.
* test(discord): upgrade to real HTTP server proof for bound requestDiscord
* fix(discord): remove unnecessary type assertion in bound test
Gap finding: tempdir-8h9i0j1k, tempdir-9i0j1k2l
Replace bare fs.mkdtemp/mkdtempSync calls in 4 test files with the
shared createSuiteTempRootTracker() helper, adding proper cleanup
where none existed.
Background: the project has standardized on shared temp dir helpers
since 88b87d893d (withTempDir) -> 13df67ebc8 (createSuiteTempRootTracker)
-> 06431fd99b / #87298 (CI guard + docs guidance). This PR continues
that migration for the highest-risk files.
Files changed:
- src/skills/lifecycle/install-fallback.test.ts: variable overwrite risk
- src/auto-reply/reply/session-hooks-context.test.ts: helper leaks temp dirs
- src/auto-reply/reply/abort.test.ts: createAbortConfig leaks root dir
- src/auto-reply/inbound.test.ts: test cases without cleanup
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(cron): clear agentTurn thinking override when patched with null
Cron agentTurn patches could clear model/fallbacks/toolsAllow overrides by
sending an explicit null, but thinking had no clear path: the patch schema and
normalizer dropped thinking:null before it reached the merge logic, and the
payload merge only handled string values. Blanking the Thinking/Effort field in
the Cron Control UI therefore silently preserved the old value.
Add thinking:null support across the patch schema, exported type, normalizer,
and payload merge (mirroring model). The Control UI now sends an explicit clear
for model/thinking when an edited job blanks a previously stored override, and
the CLI gains --clear-thinking for parity with --clear-model.
* docs(cron): document --clear-thinking beside sibling clear flags
postJson reads the Anthropic OAuth token endpoint response body with
an unbounded await response.text(). A compromised or hijacked OAuth
endpoint can stream an arbitrarily large body and force the runtime
to buffer the entire payload before parsing — an OOM/DoS vector.
Replace with readResponseWithLimit at 16 MiB cap + TextDecoder decode
to match the sibling bounded-read pattern (provider-http-errors.ts:308).
Co-authored-by: Claude <noreply@anthropic.com>
handleScannedLink does not set self.step = .connect after scanning a QR
code from the welcome step. The scanner sheet dismisses and the UI
returns to the welcome screen instead of showing connection progress.
The sibling handleScannedSetupCode path correctly advances the step;
this aligns the QR scan path.
Fixes#98297
Co-authored-by: Cursor <cursoragent@cursor.com>
* [AI] fix(memory-wiki): gracefully handle unparsable YAML frontmatter in vault scans (#96125)
toWikiPageSummary now catches YAML parse errors during vault-wide scans
(compile/lint/query/status) and returns a degraded WikiPageSummary with
frontmatterError set instead of letting the error propagate through
readPageSummaries -> compileMemoryWikiVault, which previously caused one
bad page to crash the entire vault.
Edit paths (apply, chatgpt-import) still throw on bad frontmatter to
prevent silent metadata loss on write.
- markdown.ts: add frontmatterError field to WikiPageSummary, extract
splitWikiFrontmatterBlock, catch YAML.parse in toWikiPageSummary
- lint.ts: add invalid-frontmatter lint code, skip frontmatter-derived
checks when frontmatterError is present
- markdown.test.ts: test degraded parse preserves body, empty frontmatter
- lint.test.ts: test invalid-frontmatter lint on broken page, healthy
page unaffected
Related to #96125
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
* fix(memory-wiki): isolate invalid frontmatter scans
* fix(memory-wiki): keep malformed lint reports fail-closed
* fix(memory-wiki): reject non-mapping frontmatter
* fix(memory-wiki): validate generated index targets
---------
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@golden-gate.local>
* fix(doctor): merge colliding model-ref map keys instead of dropping
When two retired model-ref keys in a models map upgrade to the same
current ref, rewriteModelRefMapKeys dropped the second entry silently
and still logged a success line, so doctor --fix lost one entry's
params/streaming/alias/agentRuntime tuning while falsely reporting it
as upgraded.
Mirror the sibling provider-models-array merge path (#90047): deep-merge
the colliding entries preserving disjoint fields, keep the existing
value deterministically on a true field conflict, and push a merge or
collision change-log line instead of dropping.
* fix(doctor): block prototype keys when merging colliding model-ref map entries
mergeModelRefMapEntries copied arbitrary incoming config fields into a
plain object, so a colliding model entry carrying a __proto__ field could
mutate the merged entry prototype during legacy config migration. Skip
isBlockedObjectKey fields before assigning or recursing, matching the
shared config merge guard. Disjoint-field merging is unchanged.
* fix(doctor): preserve canonical model migration values
* fix(doctor): filter blocked keys on both merge sides and report retired source key
* fix(doctor): preserve colliding retired model settings
---------
Co-authored-by: Vincent Koc <vincentkoc@ieee.org>
Co-authored-by: Peter Steinberger <steipete@golden-gate.local>
* fix(plugins): apply output text transforms to toolcall_delta and toolcall_end events
toolcall_delta and toolcall_end events bypassed the output replacement
pipeline, leaking masked tokens (e.g. PII placeholders) into tool call
arguments and external systems.
Closes#97761
* fix(plugins): transform nested tool call arguments in output replacements
Add recursive transformToolCallArgumentText to handle strings, arrays,
and nested objects in tool call arguments, not just the name field.
* fix(plugins): keep tool names unchanged and cover result path
- Remove toolCall.name transformation to prevent breaking tool routing
- Add arguments transform to stream.result()/done.message via
transformContentText for tool-call content blocks
* fix(plugins): narrow argument transform to toolCall content blocks only
Only transform arguments on content blocks with type=toolCall to avoid
touching non-tool-call blocks that happen to have an arguments field.
* test(plugins): assert stream.result() tool-call content block is transformed
* test(plugins): fix result-path fixture to actually contain toolCall block
The previous fixture used makeAssistantMessage('final') which produces
a plain text content block, not a toolCall block. The stream.result()
assertion was vacuously passing.
* style(plugins): fix type cast in result-path test assertion
* fix(agents): preserve tool argument transforms after repair
* refactor(agents): keep tool transform boundary narrow
---------
Co-authored-by: Peter Steinberger <steipete@golden-gate.local>
