…ompaction

Per swim-35 convergent finding (Elliott journal evidence + Cael+Ronan
source trace): request_compaction() always failed because the call sites
in agent-runner-execution.ts and followup-runner.ts dropped the session's
active provider/model on the floor. resolveEmbeddedCompactionTarget then
fell through to DEFAULT_PROVIDER/DEFAULT_MODEL ('openai'/'gpt-5.4') which
no prince has auth for, producing 'Unknown model: openai/gpt-5.4' that
classified as 'reason=unknown' in the journal and an instant <1s failure.

Three coupled defects, all fixed:

1. Defect 2 (root cause): pass run.provider + run.model into
   compactEmbeddedPiSession at both call sites.

2. Defect 3 (caller lies to tool): both sites previously discarded the
   compactEmbeddedPiSession return value and unconditionally returned
   { ok: true, compacted: true } even on real failure. This caused
   incrementVolitionalCompactionCount to fire on phantom-successful
   compactions, lying to /status telemetry. Now honor the real result.

3. Defect 1 (silent failure): add log.warn on the resolve-with-failure
   branch in request-compaction-tool.ts so volitional failures are
   visible. Add 'unknown model' classifier branch in compact-reasons.ts
   so the journal stops saying 'reason=unknown' for this exact case.

Tests: 15/15 green on request-compaction-tool + compact-reasons.
tsc --noEmit clean across whole project.

Closes karmaterminal/openclaw-bootstrap#639

…del + skip-reason log level

Three follow-up fixes from PR #191 review (copilot pass relayed by figs-as-ronan):

1. followup-runner.ts:283 — read inner-scope provider/model from the
   run: async (provider, model, ...) fallback dispatcher (line 207),
   not the outer run.provider/run.model (persisted primary). Otherwise
   a fallback-selected model still routes compaction through a primary
   that may be in cooldown — half-fix.

2. agent-runner-execution.ts:1020 — same shadowing fix; closure opens
   at line 805. Same failure mode.

3. request-compaction-tool.ts:169 — downgrade resolved-failure WARN
   to info-level when the reason matches a legitimate skip case
   (nothing to compact / below threshold / already compacted / no
   real conversation messages). Mirrors isCompactionSkipReason from
   commands-compact.ts:39 with a local copy to avoid cross-package
   coupling (agents/tools → auto-reply/reply). Real failure cases
   still WARN.

Findings 1 & 2 are correctness; finding 3 is observability polish.

…alls

Scribe self-review pass (Cael's claude_session_start /pr-review-toolkit
on 37aae9c) caught the next failure mode after the original openclaw#639 bug
and the Copilot fold:

compactEmbeddedPiSession accepts authProfileId and uses it via
resolveEmbeddedCompactionTarget to pick credentials. Both call sites
(agent-runner-execution.ts:~1027, followup-runner.ts:~285) were not
threading it — so a session on a non-default auth profile (per-user
API key) would route compaction with the right provider/model but the
wrong credentials.

Conditional pattern mirrors agent-runner-execution.ts:841-844: thread
the persisted authProfileId only when the inner-scope provider matches
the persisted primary. On fallback to a different provider, the
persisted profile is wrong (keyed to primary), so leave undefined and
let resolveEmbeddedCompactionTarget pick the default profile for that
provider.

This is symmetric to findings 1+2 from the Copilot fold: same fallback-
shadowing class, one parameter deeper. Pre-existing bug, but the lane
is open and the fix is +2 conditional + 1 field per call site.

…ompaction-provider

fix(openclaw#639): thread session provider/model through volitional compaction

…le + status restoration

Updates RFC docs/design/continue-work-signal-v2.md to reflect the totality of changes since 107ca2b (the prior RFC edit) plus the two ship-gate PRs about to land:

- §4.3: document session provider/model threading through volitional compaction (openclaw#191 / bootstrap#639). Three coupled defects: root cause, caller-honesty (phantom-counter), visibility (`unknown_model` classifier + `isLegitSkipReason` helper + `log.warn` on resolve-with-fallback + scope-aware `authProfileId`).
- §6.1: add `[context-pressure:noop]` log anchor with reason taxonomy (window-zero / below-threshold / band-dedup); document the bootstrap#580 investigation cycle (`:reach`/`:skip` instrumentation, root cause = sentinel collision on band 0, fix = -1 sentinel).
- §6.3: clarify Discord/agent path through src/auto-reply/status.ts was reconnected at openclaw#187 + tested at #188 (the line had been silently dropped in an earlier refactor); note `volitional: N` is honest only after #191.
- §6.4: replace 'instrumentation is not currently in place' note with status of distinguishing-instrumentation work (openclaw#164/171/172/173).
- Appendix C.1: add 'Closed failure modes' table — phantom-counter, hedge-timer ref leak, band-0 dedup, precondition-skip blindness, Copilot summarization headers, dist-bundle satellite chunks, subagent-announce runtime path mismatch.
- Appendix D.2: add evidence-location rows for the new file paths (volitional threading sites; armHedgeTimer; status renderer; request-compaction-tool tests; context-pressure noop sites; agent-runner runtime promotion; subagent-announce co-location; F-NOISE scheduler test).
- Header: bump test count (~180 across 13 files, was '172 across 8') to reflect additions in #165, #170, #188, #193.

Skip-list (no RFC mention): #174 sessions/config raw-key sweep (internal hygiene); #173 Copilot log-enabled nits (micro-hygiene); 86134af removal of investigation breadcrumbs (cycle is folded into §6.1 narrative).

Refs:
  openclaw#191 head fc3f415 (in-flight, MERGEABLE/UNSTABLE, APPROVED)
  openclaw#193 head 14483a6   (in-flight, MERGEABLE/UNSTABLE, APPROVED x2)
  openclaw#187, #188 (merged d787890)
  openclaw#160, #162, #164, #165, #169, #170, #171, #172, #173, #174

🍆 in 🩲: this is a docs PR; if either #191 or #193 changes shape pre-merge the affected paragraph here will need a one-line touch-up.

Co-Authored-By: dandelion cult - ronan 🌊 <karmafeast@gmail.com>

)

The empty catch in agent-runner.ts post-compaction lifecycle silently
swallowed spawnSubagentDirect rejections — staged TaskFlow records were
released by consumeStagedPostCompactionDelegates, then any throw left
zero trace in logs, journals, or operator events. Same bug class as
the openclaw#639 ship-gate we just fought.

Extract the per-delegate spawn loop into
continuation/post-compaction-dispatch.ts so the failure path is
testable without agent-runner integration fixtures. On failure, emit:

  - log.warn '[continuation:post-compaction-spawn-failed]' with
    error message, sessionKey, and task slice (mirrors
    delegate-dispatch.ts:223-232)
  - enqueueSystemEvent so the operator surface sees dropped work

Failures are isolated per-delegate; one rejection does not abort the
remaining staged delegates.

Test coverage: 5 cases including success, named-anchor + system-event
emission on rejection, per-delegate isolation, non-Error throwables,
and empty-list short-circuit.

Closes #203

)

The empty catch block in agent-runner.ts was silently swallowing spawn
failures for post-compaction delegates. Since consumeStagedPostCompactionDelegates
had already marked the TaskFlow as released, these failures resulted in
lost work with zero trace.

This fix extracts the post-compaction dispatch loop into a dedicated
helper function (dispatchPostCompactionDelegates) with proper error
handling that mirrors delegate-dispatch.ts:223-232:

- Logs warn with anchor [continuation:post-compaction-spawn-failed]
- Enqueues a system event so failures are observable

Bug-class precedent: openclaw#639 (same silent-catch pattern in delegate-dispatch).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…th (#209)

Adds behavioral test coverage for the openclaw#639 fix arc. These tests guard
against regression when the next provider/model rename or fallback
refactor happens.

Coverage:
- Counter truthfulness: getVolitionalCompactionCount increments ONLY on
  {ok:true, compacted:true}; legit-skip reasons and failures do not
  increment.
- Log level correctness: legit-skip → log.info with [resolved-skip]
  anchor; real failures → log.warn with [resolved-failure] anchor.
- Call-site threading: verifies triggerCompaction closure passes the
  session's actual provider/model/authProfileId to compactEmbeddedPiSession
  (not DEFAULT_PROVIDER/DEFAULT_MODEL).
- authProfileId fallback-drop: when compaction provider ≠ persisted
  primary, authProfileId is dropped (set to undefined).

Refs: openclaw#639

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…/error paths (#205)

- Route request-compaction-tool's resolved-failure warn log through
  classifyCompactionReason so journal lines carry the structured code
  (e.g. unknown_model, timeout, provider_error_4xx) alongside the raw
  reason string. Before: operators had to grep the freeform 'reason=...'
  field; after: they can pivot on a closed-taxonomy 'code=...' field.
- Mirror the same classification on the background-error branch so
  transport-level rejects (AbortError / module-not-found / network)
  surface a structured code instead of only the raw throw message.
- Add request-compaction-tool.classifier-emission.test.ts (4 cases):
  resolve-with-unknown_model warn includes code=unknown_model, generic
  'Compaction cancelled' warn emits some code= field (pinning emission
  contract for future classifier extensions), promise-rejection error
  path includes code=<classified>, legit-skip takes info path (not warn).
- Pre-existing isLegitSkipReason routes 'nothing to compact' /
  'below threshold' / 'already compacted' / 'no real conversation
  messages' to the info path unchanged; the classifier only runs on
  real failures. This preserves the openclaw#639 legit-skip-quiet posture
  while closing the #205 structured-logging gap.
- triggerCompaction catch sites in followup-runner.ts and
  agent-runner-execution.ts intentionally unchanged: their return flows
  through the warn sibling which now classifies, satisfying the #205
  acceptance criterion without extra touch on the catch itself.

pnpm tsgo clean. 14/14 request-compaction-tool + compact-reasons tests green.

Closes #205.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

…upstream

22 comment-edits across 13 files. All bare `#N` references that point
to karmaterminal/openclaw or karmaterminal/openclaw-bootstrap fork
issues now use the prefixed `karmaterminal/openclaw#N` (or
`karmaterminal/openclaw-bootstrap#N`) form so they cannot be silently
auto-linked to unrelated upstream openclaw/openclaw issues.

Citations changed:
- bug openclaw#639 -> karmaterminal/openclaw#639 (8 sites: compact-reasons.{ts,test.ts}, followup-runner.ts ×3, agent-runner-execution.ts ×3, request-compaction-tool.ts ×2)
- (#580) / See #580 -> (#580) / See #580 (4 sites: context-pressure.{ts,test.ts})
- (#531) -> (#531) (2 sites: scheduler.test.ts)
- issue #475 -> #475 (1 site: session-cost-usage.ts)
- bootstrap#475 -> karmaterminal/openclaw-bootstrap#475 (1 site: discoverAllSessions.test.ts; the line above already used the prefixed form, this was the laggard)
- Bug #473 fix / issue #473 -> #473 (3 sites: subagent-announce.continuation.runtime.{ts,test.ts})
- review gap on #188 -> review gap on #188 (1 site: status.test.ts; also dropped a named-prince prefix in the same test label, which separately satisfies part of #200)

Left alone (out of scope per issue):
- Pre-existing upstream openclaw/openclaw citations (openclaw#26905, openclaw#36142, openclaw#60552, openclaw#28491, openclaw#41981, openclaw#18264, openclaw#59428, openclaw#31311, openclaw#14869, openclaw#17971, openclaw#29683, openclaw#58409, openclaw#35481, openclaw#30115, openclaw#14396, openclaw#19537, openclaw#1205, openclaw#1435, openclaw#26881, openclaw#32228, openclaw#35074, openclaw#26881)  -- these were present at v2026.4.14 and resolve correctly.
- delegate-dispatch.ts:62 (`openclaw#189.`) -- already in the canonical prefixed form per the issue's exception note.

Refs #199