Fold the writer-claim fence into the four sync write ops at their
definition sites, collapse the logical-layer fence wrappers to plain
re-exports, delete the session-accessor.sqlite middle barrel (the only
unfenced escape hatch), and retarget its 26 importers to the focused
defining modules. replaceSessionEntrySync intentionally stays unfenced:
its two production callers own their freshness guards and its return
carries no rebound signal. Net -106 production LOC, behavior-neutral.
Part 1 of the session-accessor dual-layer collapse.
* perf(agents): keep turn-path model catalog reads off the full live build
First agent turns (embedded and cron) resolved thinking capability through
loadPreparedModelCatalogSnapshot without readOnly, which materialized the
full live model-runtime catalog: ambient synthetic-auth discovery fanned out
to every registered provider and loaded plugin discovery modules through
jiti source transform (3,172 TS modules, 36s event-loop block, +600MB heap,
58.7s model-selection on a cold gateway).
- add loadProviderScopedThinkingCatalog: manifest metadata first, then a
provider-scoped read-only static catalog, then scoped live discovery only
for runtime-discovery providers (preserves #116584 Ollama semantics)
- route scopedLiveProviderDiscovery through the scoped read-only loader
- scope live-mode ambient synthetic-auth refs to the requested providers
- bound the last-resort synthetic-auth sweep to discovery entry modules
- memoize per-turn plugin skill dir resolution/republish (single-slot,
lifecycle-cleared; was a full walk + symlink republish every turn)
Cold first turn 72.7s -> ~22s wall (remaining cost is provider prefill of
the ~19.5k-token default prompt); model-selection 58,726ms -> 124ms.
* test(agents): align model-catalog.runtime mocks with scoped thinking catalog seam
Explicit vi.mock factories must export every binding prod touches; the new
loadProviderScopedThinkingCatalog export is now mocked everywhere the module
is stubbed, and the live-model-switch Ollama hydration test asserts the new
provider-scoped seam instead of the retired unscoped snapshot call shape.
* test(agents): export scoped thinking catalog from every prepared-catalog mock; split synthetic-auth helpers
- add loadProviderScopedThinkingCatalog to all explicit prepared-model-catalog
and model-catalog.runtime mock factories (vi.mock factories must export every
binding prod touches)
- move synthetic-auth ref scoping/resolution into
prepared-model-runtime.synthetic-auth.ts; keeps facts under the max-lines cap
* test(agents): prove scoped thinking hydration for runtime-only models
Boundary proof for the ClawSweeper review gap: the three-tier helper stops at
manifest or scoped-static when they resolve, and runs provider-scoped live
discovery (no broad fanout) only for runtime-only models; cron selection
hydrates through the same scoped helper and skips it entirely for thinking=off.
* test(agents): accept rest args in scoped thinking catalog mocks
* feat(agents): record run-end worktree cleanup outcome
Persist removed, retained, and failed run-end cleanup outcomes on managed worktree records. Operators and QA can inspect the durable fact through worktrees.list and openclaw worktrees list --json.
Release note: Managed worktree run-end cleanup now records why a checkout was removed or retained in worktree list JSON.
* test(qa): prove dirty worktree retention outcome
* chore(protocol): regenerate swift gateway models
* fix(agents): harden worktree cleanup recovery
Register run_end_cleanup_json as a lazy compatible column so same-version v6 index repair and read-only doctor migration can recover databases created before the column existed.
Type removal contention at the registry boundary; unexpected claim failures now best-effort record a bounded failed outcome and rethrow the original error.
* fix(ci): clear repo-wide lint debt blocking merge gates
The red-main landing rule requires this PR to repair repository-wide merge-gate debt instead of bypassing it. Apply the current lint contracts mechanically and split turn-transition coverage into a concept-named sibling with per-file-safe test state.
Exact line delta: +676/-574 (net +102) across 44 test/support files.
* fix(ci): preserve cached health refresh proof
Require the public refresh call to exist before accepting that sensitive fields were omitted, so the boundary proof cannot pass on a missing call.
* fix(ci): correct test typing left by the lint sweep
Literal-widened totalTokensVersion fixtures, a WebSocket RawData overload
mismatch, and the protocol schema document cast broke check-test-types
after the repo-wide lint repair. Aligns the fixtures with SessionEntry,
narrows Buffer handling per RawData, and keeps the JSON-shaped undefined
omission under structuredClone.
* test(agents): reuse upstream resource-loader test support
The session-loop split and #120463's helper extraction landed the same
createResourceLoader/createCompactionHandlers twice; the rebase kept both,
orphaning main's agent-session-loop-resource-loader.test-support.ts and
failing the dead-code gate. Import the upstream helpers and delete the
duplicates.
* fix(agents): reject finalized rows at the worktree removal claim
Address the accepted ClawSweeper late-claim finding by rereading and rejecting missing or finalized worktree rows inside the synchronous removal-claim transaction.
Preserve the authoritative cleanup invariant: finalized contenders record nothing, while retained-busy is written only while the row remains live.
* refactor(agents): reuse registry update for busy outcomes
Keep the live-row conditional write in the canonical registry update path so the finalized-claim repair stays below the registry max-lines ratchet without weakening the authoritative-outcome invariant.
* test(agents): drop session test duplicates after rebase
Keep current main as the canonical owner of next-turn lifecycle coverage and correctness test support after replaying the older lint-debt split.
* fix(agents): guard post-abort cleanup outcomes against finalization
After abortWorktreeRemoval releases a stale remover's claim, its retained or
failed write raced a finalizing remover and could overwrite the authoritative
removed-lossless fact. Route every retained/failed write through the live-row
condition; only the finalizing remover's own removed-lossless write stays
unconditional.
* fix(agents): persist the removal outcome atomically with finalization
A delayed removed-lossless write after remove() finalized could race a
restore plus newer cleanup and overwrite the newer operator-visible fact.
The run-end outcome now rides remove()'s finalization update; every other
cleanup write stays live-row conditional, so no post-finalize write path
remains.
* test(qa): restore strict cached-health contract assertions
The lint sweep's Boolean() coercions let truthy non-booleans satisfy the
wire-typed cached-meta contract. Assert the literal boolean for unknown-typed
fields and use nullish-coalesced strict equivalents for boolean chains.
* fix(agents): clear the stale cleanup outcome when restoring a worktree
A restored checkout begins a new lifecycle; leaving the removed-lossless
fact on the live row showed operators a stale result until the next
cleanup. Restore clears the recorded outcome and the regression asserts
the cleared state before the next cleanup records fresh truth.
* fix(agents): scope stale cleanup outcomes to their observed lifecycle
A stale remover's retained/failed write raced a concurrent remove-plus-
restore: the revived row is live again, so the live-row condition alone
could stamp a prior-lifecycle outcome. Condition those writes on the
activity stamp the remover observed; restore bumps lastActiveAt, making
any prior-lifecycle write a no-op.
* fix(agents): advance the restore activity stamp within one millisecond
Stale cleanup writes fence on the activity stamp they observed; a restore
completing in the same millisecond could revive the row with an identical
stamp and let the fence match. Restore now always advances past the
stored value, and the ABA regression pins the clock to prove the
same-millisecond case.
* fix(worker): preserve Responses replay across handoff
Long OpenAI Responses cloud-worker sessions retain authoritative compaction replay through windowing and fail visibly when it cannot be preserved.
* test(worker): align replay fixture with frame limit
* fix(ai): preserve Responses server compaction state
Persist opaque Responses compaction items as fenced provider replay state so long stateless sessions can resume authoritative compressed history without exposing it in display or diagnostics. Carry state through worker transcripts and prune replay prefixes without splitting tool pairs.
Release note: Preserve long OpenAI Responses sessions across server-side compaction and worker restarts.
Related: #95788
* test(ai): align long-context fixtures with CI contracts
Make tool-result fixtures type-complete, use the canonical model selector helper, remove unused test-helper exports, and route the paid long-context live probe through the dedicated Gateway profile shard.
* test(ai): type mocked Responses terminal events
Give the mock SSE event collection an explicit open event shape so terminal response events coexist with output-item events under the root test typecheck.
* fix(ai): suppress rejected compaction replay
Persist a route-fenced suppression tombstone when encrypted-content recovery rejects a compaction item, so later turns do not retry the same opaque state. Preserve the tombstone through transcript redaction and cover successful fallback followed by the next turn.
* fix(ai): keep compaction suppression transport-private
Keep the suppression contract local to its sole Responses transport owner and make the regression fixture satisfy root type and lint checks without widening the Plugin SDK surface.
* refactor(ai): remove compaction suppression re-export
* fix(ai): scope compaction suppression to replay route
Keep foreign-route rejection tombstones from hiding the newest compatible Responses compaction while preserving same-route suppression.
* fix(ai): harden Responses replay recovery
Stage encrypted replay recovery so compaction is only suppressed after an attributable rejection. Preserve terminal ordering and keep provider replay within worker frame budgets without truncating opaque state.
* refactor(ai): centralize Responses output indexes
Keep normalized output identity tracking in the stream-slot owner, move response failure state to its diagnostic owner, and remove the obsolete replay clone export so exact-head static gates remain shrink-only.
* fix(ai): retain idless terminal tool identity
Use the canonical empty identity only when a provider supplies neither call nor item id, preventing terminal recovery from duplicating a done-only tool call while preserving stronger identities when available.
* fix(sessions): hide provider replay from public events
* fix(ai): stage encrypted replay recovery
* fix(ai): keep replay attempt kind internal
* fix(ai): route Azure through replay recovery
Use the shared encrypted-content retry owner for Azure Responses so compaction suppression and prompt-observer variants stay coherent across transports.
* fix(ai): harden replay persistence boundaries
Fence Azure replay by the resolved request endpoint, drop invalid replay during transcript sanitization, and surface worker-launch replay omissions through the existing redacted diagnostic path.
* refactor(agents): route scheduler tool-name consumers through canonical identity
Introduce AUTOMATIONS_TOOL_NAME + isAutomationsToolName() in
src/agents/tools/automations-tool-name.ts as the single source of truth for
the scheduler agent tool's name, and convert every exact-name consumer:
factory descriptors, deferred-followup availability detection, add-counting,
policy deny lists, mutation classification, trusted media set, tool catalog
id, system-prompt tool order and tool-line map, sandbox deny defaults,
delegation capability map, MCP loopback probes, and local-model lean deny.
Behavior-neutral: the constant still resolves to "cron". Prepares the
rename in RFC openclaw/rfcs#50 so the flip is a one-line change with no
scattered literals.
* feat(agents): rename scheduler agent tool cron -> automations
Flip AUTOMATIONS_TOOL_NAME to "automations" and register the legacy name:
- TOOL_NAME_ALIASES gains cron -> automations, so persisted toolsAllow/
toolsDeny lists, tool groups, and creator allowlists written before the
rename keep matching through the same shipped mechanism as bash -> exec.
No doctor rewrite needed.
- isAutomationsToolName() accepts legacy names so saved transcripts keep
their mutation/replay-safety classification; MUTATING_TOOL_NAMES retains
the legacy entry for the same reason.
- Tool label, catalog label, and tool-search keywords follow the rename
("cron" kept as a search synonym).
- Regression tests cover old-name policy matching (allow and deny), legacy
transcript replay classification, and legacy creator allowlists
normalizing to the canonical id.
Model-facing description strings still say cron; those move in the
follow-up strings PR. Part of RFC openclaw/rfcs#50 Phase 1.
* test(agents): update creator-cap expectations for canonical automations id
The creator tool surface derives from normalized live tool names, so derived
toolsAllow outputs now emit "automations". Passthrough paths without a
creator cap keep storing user input verbatim; those expectations stay on the
legacy name as stored-data coverage.
* fix(gateway): canonicalize legacy cron tool calls and restore scheduler deny protection
Review follow-ups from ClawSweeper and Codex on the rename (RFC 0026):
- MCP loopback tools/call resolves legacy "cron" names to the published
automations tool without re-advertising the old name in tools/list.
- Gateway /tools/invoke canonicalizes legacy names before core-id checks and
exact-name dispatch, so pre-rename integrations keep working.
- Security fix: dangerous-tools deny lists (owner-only HTTP deny and
control-plane set) were keyed on the literal "cron", so the renamed tool
silently lost default-deny and owner-only protection on the HTTP invoke
surface. Lists now use the canonical constant, and the gateway.tools.allow
un-deny filter normalizes both sides so legacy allow entries still lift it.
- Voice high-impact confirmation list and MCP serve creator allowlist follow
the canonical name.
Existing cron-regression suite now proves the legacy path end to end:
default deny 404 for both names, legacy allow entry lifts the deny, and
non-owner protection holds.
* fix(agents): cover stdio MCP legacy calls, probe prompts, and prompt snapshots for the rename
- stdio MCP servers (openclaw-tools-serve / plugin tools handlers) resolve
legacy "cron" callTool names to the published canonical tool, matching
the HTTP loopback behavior; listTools stays canonical-only.
- Live probe prompts instruct harnesses to load/call the automations MCP
tool (mcp__openclaw__automations) instead of the retired name.
- Prompt snapshot fixture filter follows the canonical name (the renamed
tool had silently dropped out of the Codex dynamic-tools snapshots);
snapshots regenerated as a clean rename.
- Type-cast the new mcp-http handler test payloads for check-test-types.
* test(agents): update tool-surface expectations for the automations rename
CI-surfaced fallout in shards not covered by the focused local runs: tool
availability, agent-config filtering, coding-tools construction,
model-provider lean policy, and skill dispatch all assert the scheduler
tool's surface name. Mock fixtures and expectations follow the canonical
id; legacy-name coverage stays in the dedicated policy/creator-cap/invoke
regression suites.
* test(gateway): update tool-resolution exclude expectations for automations rename
* test(security): update trust-model audit expectations for automations rename
* docs(agents): declare cron a permanent scheduler-tool alias per owner decision
Maintainer decision (Omar): cron is not being retired anywhere — config keys,
RPC methods, schedule syntax, and the CLI token all keep it, and the tool
alias follows the same permanent contract as bash -> exec. No doctor rewrite
and no removal window; comments updated to state the contract instead of a
deprecation plan.
* chore(agents): regen prompt snapshots after rebase onto main
* fix(agents): teach canonical automations tool in fallback guidance and reuse the identity source
Review follow-ups: the structured-list fallback still taught models the cron
tool; the cron-scope test echoed its own stub; MCP serve allowlist and voice
confirmation hardcoded the name instead of the canonical constant.
* fix(mcp): place automations identity import outside the header comment
---------
Co-authored-by: Omar Shahine <10343873+omarshahine@users.noreply.github.com>
* perf(sqlite): cache synchronous prepared statements
* fix(sqlite): refresh cached statement metadata
* fix(sqlite): bound cached statement retention
* fix(sqlite): close statement cache lifecycle gaps
* test: suppress SQLite warning in retention child
* test(sqlite): make transcript-load counter statement-cache-aware and allowlist raw test SQL
* test(sessions): type the wrapped iterate against its overloaded signature
* test: share a statement-cache-aware SQLite execution counter across prepare-count tests
* fix(pr): validate head SHA with bash regex so fork-CI runners without ripgrep pass the artifacts-init guard
---------
Co-authored-by: VACInc <3279061+VACInc@users.noreply.github.com>
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
* feat(sessions): retain reset history in sqlite with physical disk budget
* fix(sessions): satisfy test-types, knip export scan, and docs map
* fix(sessions): align behavioral suites and flip proof with retained history, route-aware cleanup plans
Measured breakdown of check-prompt-snapshots in run 29557851276 (5m03s):
- 37s checkout + Node env setup
- 4m25s inside one `pnpm prompt:snapshots:check` process:
- ~35s tsx import of the generator graph
- ~3min first createOpenClawCodingTools call: the message-tool schema
build resolves the scenario channel plugin, the empty loaded registry
falls back to the bundled-channel jiti loader
(extensions/telegram/index.ts), and jiti re-transpiles the core source
graph (136.6s jiti self time in --cpu-prof; second generation call in
the same process takes 0.3s)
- ~1min more for the cold provider-plugin lookup in
normalizeAgentRuntimeTools; no provider plugin owns tool-schema hooks
for the `codex` harness provider, so the jiti load only rediscovers a
no-op
- <2s actual rendering + oxfmt + drift compare
Split axis decision: the planned 2-3 way scenario shard matrix cannot
lower the pole because the cost is a one-time per-process transpile;
every shard would pay the full ~4min again (jiti fsCache does not help:
a warm-cache rerun still took 200s). Fix the generator instead:
- Preload each scenario's channel plugin surface (channel-plugin-api.js,
resolved via the bundled-plugin public-surface resolver) through the
ambient tsx/vitest ESM graph, and pin exactly that channel in a scoped
loaded registry while the scenario's tools are built. Snapshot bytes
are unchanged: the loaded plugin object is the same export the bundled
jiti fallback materializes, and cross-channel action discovery still
sees no other loaded channels.
- Pass allowProviderRuntimePluginLoad=false to normalizeAgentRuntimeTools;
registry-only resolution returns the same undefined plugin instantly.
Coverage guard: test/scripts/prompt-snapshots.test.ts now generates the
snapshot files and asserts zero jiti plugin-loader calls (with the
offending transform targets in the failure message), so a future scenario
or channel that misses the preload fails the suite instead of silently
restoring the 4-minute lane.
Proof: `pnpm prompt:snapshots:check` passes byte-identical against the
committed fixtures in 54s wall / 35s CPU locally (was 354s wall / 306s
CPU); focused vitest suites (32 tests incl. new guard), tsgo core-test +
extensions-test lanes, scoped oxlint/oxfmt, and `git diff --check` are
green. Expected CI pole when the lane fires: ~5.1min -> ~1.5-2min.
Peter Steinberger