Attempt dispatch stopped emitting legacy `sqlite:` session-file markers in
#113233, which replaced SessionTranscriptRuntimeTarget.sessionFile with
storePath and started passing the canonical session key instead. The Codex
plugin still re-derived its own SQLite gate from that string, so
parseSqliteSessionFileMarker never matched and every Codex run bailed out
with reason "non-sqlite-session-target".
The host already owns SQLite target resolution and identity validation: it
returns a recorder only when a committed session row resolves. Drop the
stale marker gate and the now-dead trajectorySessionFile plumbing, leaving
the host-provided recorder as the single gate.
A user-home app-server keeps its native Codex account, and OpenClaw verified only
half of that contract. A subscription route was checked against the native
account, but a Platform (API-key) route was not: an operator signed in to Codex
with a ChatGPT subscription who selected an API-billed model silently spent their
plan. Both directions of the same billing boundary now share one account/read
check. An absent account is left alone, since a native home may serve a custom
model provider that reports no OpenAI account at all.
The home-scope rule behind #114397 was also expressed four times: two
prepared-auth call sites spread it in conditionally from raw plugin config, two
others read resolved start options, and the supervision connection applied its
own default inline. resolveCodexAppServerHomeScope now owns it, and homeScope is
a required argument of resolveCodexAppServerPreparedAuthHandoff, so a new call
site that forgets it fails to compile instead of failing every turn.
Docs dropped the stale user-home paragraph that still described the startup
failure #114397 removed and told operators to delete a working OpenAI profile.
* refactor(memory-core): remove orphaned shadow trials
* refactor(qqbot): migrate direct voice upload formats
Release note: QQBot configs using voiceDirectUploadFormats now migrate through openclaw doctor --fix to audioFormatPolicy.uploadDirectFormats at root and account scope. Existing nested uploadDirectFormats values win conflicts, and runtime reads only the nested policy.
* refactor(feishu): migrate tools base alias
Release note: Feishu configs using tools.base now migrate through openclaw doctor --fix to tools.bitable at root and account scope. Existing bitable values win conflicts, and runtime accepts only the canonical key.
* refactor(extensions): remove dead runtime state
* chore(config): refresh bundled channel metadata
* refactor(codex): keep realtime fallback type private
* test(ci): align QA compatibility scenario count
A prepared OpenClaw auth handoff was still resolved for connections that
target the operator's native Codex home, so every turn failed with
"Prepared Codex auth requires an isolated app-server home." once a stored
OpenAI profile produced a prepared model route. User-home mode now keeps
its native account for both subscription and Platform routes, matching the
documented contract; the shared-client guard stays as the backstop.
* fix(ui): keep stable chat rows in insertion order and only sort live tool/stream items by timestamp
* fix(ui): keep live rows within current turn
* fix(ui): keep current work above queued turns
* fix(ui): keep streamed replies above queued turns
* fix(ui): preserve reconnecting run order
* fix(ui): preserve causal terminal ordering
* fix(ui): bound replay rows to owning turns
* fix(ui): keep question summaries in owning turns
* fix(ui): scope question run ownership to session
* fix(ui): restore reconnecting chat run identity
* fix(ui): remove unused chatItemTimestamp import in chat-thread-build.ts
* fix(ui): correlate question summaries with agent runs
Co-authored-by: Peter Lee <li.xialong@xydigit.com>
* chore(i18n): refresh native source baseline
Co-authored-by: Peter Lee <li.xialong@xydigit.com>
* fix(ui): remove unused chatItemTimestamp export and split tool-stream test file
---------
Co-authored-by: Peter Steinberger <steipete@gmail.com>
* refactor(prompt): plain inbound context labels with a provenance marker
Replaces trust-worded inbound context labels ("(untrusted metadata)",
"(untrusted, for context)") with plain labels plus a fixed provenance
marker suffix appended to every OpenClaw-injected context header.
Detection keys on the marker, not label text, so strippers stay correct
across UI, TUI, replay, /trace segmentation, memory recall, and the Swift
chat preprocessor. Drops sanitizeInboundSystemTags in favor of the marker
boundary plus trusted system-prompt narration.
Renames the untrusted-named plugin SDK context identifiers to
channel-provenance names, keeping deprecated aliases registered for
removal after 2026-09-08.
Adds `openclaw doctor --fix` migrations that rewrite legacy inbound
labels in stored SQLite transcripts and purge legacy envelope-
contaminated LanceDB recall rows.
* fix(ci): resolve gate failures for plain inbound context labels
- doctor sqlite readers: open read-only connections via openNodeSqliteDatabase
so the Kysely connection-boundary guardrail holds; unexport the now-internal
transcript snapshot type (Knip unused-export gate).
- compat registry: split the record table into registry-records.ts and
plugin-sdk-subpath-records.ts. The new compat record pushed registry.ts past
the 700-line oxlint cap; suppressions are disallowed, so follow the existing
sibling record-module pattern. Public exports and PluginCompatCode literals
unchanged.
- acp-runtime test: assert current finalization behavior (newline normalization
only). The bracket de-fang and System: rewrite it expected were removed with
sanitizeInboundSystemTags; forged system lines are neutralized at the
system-event queue, the single chokepoint feeding the System:-per-line render.
- regenerate docs_map and the plugin SDK API baseline manifest.
* fix(prompt): harden inbound context label migration and drop in-band sanitizer
Review follow-ups on the plain-label + provenance-marker change:
- Remove src/security/system-tags.ts. Rewriting inbound text to neutralize
look-alike `System:`/`[System]` markers corrupted legitimate user text and is
not a real injection boundary; role separation plus external-content wrapping
is. Explicit product decision, recorded at the system-event queue.
- Narrow the LanceDB legacy-row purge so it cannot delete benign memories. It
now requires a complete known legacy sentinel line, a legacy label followed by
a fenced JSON body, or the complete legacy external-content header. The prior
predicates matched ordinary prose such as `Notes (untrusted metadata):`, and
deletion is irreversible.
- Make explicit-empty canonical ChannelStructuredContext win over the deprecated
alias via a present/absent result instead of collapsing `[]` to undefined.
- Keep `\r?` in the active-memory doctor rule. It is the only rule spanning the
header's line break, migrated assistant rows skip newline normalization, and
without it the marked-header replace wins and the body strips to empty. Added
a CRLF regression test.
- Fix stale comments that described removed behavior, and cover the Swift
prose-block strip path.
Claude-Session: https://claude.ai/code/session_01WNzsPddQmxy9Y7jKD4wAxH
* test: consolidate OpenClaw test state fixtures
* test(plugin-sdk): expose isolated test state
Promote the isolated OpenClaw test-state lifecycle through a narrow published Plugin SDK subpath so extension tests no longer import private core helpers. This intentional SDK surface addition is maintainer-approved.
* test: use SDK test-state seam in extensions
Route bundled extension suites through the focused repo-local Plugin SDK test-state entrypoint and remove the Codex projector harness exports made stale by fixture consolidation. Keep the seam out of production builds and published package artifacts while auditing its real consumers in the full-tree deadcode scan.
* test(plugins): map test-state in package boundaries
* fix(agents): honor run abort signal in image and pdf tools
The image and pdf agent tools declared `execute: async (_toolCallId, args)`
and dropped the run abort signal that `wrapToolWithAbortSignal` supplies as
the third execute argument. The wrapper only races the execute promise, so an
aborted run kept sequentially downloading images/PDFs (up to the per-tool cap,
each up to the byte cap) and still issued a paid vision/PDF-model call for a
dead run.
Thread the signal into the existing `requestInit: { signal }` seam (which the
media fetch layer already merges into the download fetch) and add
`signal.throwIfAborted()` between sequential loop items and before the paid
model call. No new media-options signal field; non-abort behavior is unchanged.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(agents): keep the pdf test suite under the lint ceiling; guard model dispatch
Follow-up on the abort-signal change, resolving the check-lint failure and a
review finding on the same seam.
max-lines
---------
Adding the two pdf abort tests pushed src/agents/tools/pdf-tool.test.ts to 1018
effective lines against a ceiling of 1000, failing check-lint. Resolved without
touching config/max-lines-baseline.txt — a suppression there would also have
tripped the max-lines ratchet.
- The new tests declared their own `describe` with `beforeEach`/`afterEach`
hooks identical to the existing `describe("createPdfTool")`. They exercise
that same tool, so they now live in it and the duplicated scaffolding is gone.
- `stubPdfToolInfra`, `createPdfModelRegistry` and `FAKE_PDF_MEDIA` moved to
pdf-tool.test-support.ts, which already exists for exactly this. They go
through `createPdfToolInfraStub(completeMock)` rather than being exported
directly, because the stub wires the suite's own `complete` mock into the
model registry and vi.mock handles are file-scoped. All 21 existing call
sites are unchanged.
Abort propagation into model dispatch
-------------------------------------
The previous revision stopped cancellation at download boundaries and before
the first model call, but `runImagePrompt`/`runPdfPrompt` never saw the signal.
A run cancelled while the first provider request was in flight could still
issue the remaining ones — the image path dispatches `describeImage` once per
image in a sequential loop, so a dead run kept paying for every later image.
Both now take an optional `signal` and check it immediately before each
provider dispatch (3 sites in image-tool, 4 in pdf-tool).
Forwarding the signal further, into the provider transports themselves, is a
different seam and is deliberately left out of this PR.
Also declares `requestInit` on `ImageToolLoadWebMediaOptions`. The local facade
omitted it while the underlying loader accepts it (web-media.ts declares it and
forwards it to readRemoteMediaBuffer), so the option worked at runtime and only
compiled because spread properties skip excess-property checking. A non-spread
call site would not have.
tsgo core + core-test, oxlint, oxfmt clean; pdf-tool and image-tool suites 266
tests pass.
* test(agents): prove in-flight media aborts
* test(agents): narrow PDF loader options
* test(agents): reject abort mocks with errors
* fix(agents): propagate media cancellation
* test(agents): type PDF abort fixture
* test(agents): type prepared runtime snapshot
* fix(agents): normalize abort rejection errors
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Peter Steinberger <steipete@gmail.com>
