chore: delete six dead automation lanes (append compactor, commit review, crawl-remote deploy, finalize-open-PRs, proof nudges, state compaction)

[steipete]

Executes Stages 2-3 of the simplification proposal: delete six dead or dormant lanes outright, with the code, tests, docs, and workflow-shape invariants that existed solely for them. Net -15,820 lines (896+/16,716-) across 55 files.

Deleted lanes and evidence

Lane	Evidence	What was deleted
Append-window compactor	Latest scheduled runs all report drained=0 acked=0 errors=0 (e.g. run 30551689153); the git projection it fed was deleted in #936, and no producer posts /internal/state/append anywhere in the tree — state-append-client.ts only speaks the canonical /records/tuples + /records/commits endpoints	state-materializer.yml (a runner every 20 min), src/repair/state-materializer.ts, the four /internal/state/{append,drain,ack,dispose} Worker routes and DO endpoints, and the five state_append_* DO tables (dropped on upgrade) plus their tests and stats surface
Commit review	0 successes in the last 20 runs (11 failed, 9 cancelled)	commit-review.yml, repair-commit-finding-intake.yml, the hosted commit-sweeper CLI commands, commit-classifier.ts, commit-checks.ts, commit-finding-intake.ts, the commit_review worker lane in limits/scheduler/sweep budget accounting, and the dispatcher doc. The offline pnpm local-review engine and its --local-range reuse in the main sweeper are preserved; existing commit_finding repair jobs stay executable
crawl-remote deployment	3 runs ever, 0 successes (one still stuck waiting since 2026-07-29)	deploy-crawl-remote.yml (3,055 lines), the pinned Wrangler toolchain under .github/deploy/, the ci.yml integration job, and the 3,030-line workflow test
Finalize-open-PRs	Last success April 29; manual-dispatch only since	repair-finalize-open-prs.yml only — see skips below for why the module stays
Proof nudges + bot proof	Runs green but contributes nothing to weekly review coverage (the decisive SLO at 42.9%)	proof-nudges.yml, docs/proof-nudges.md, the proof-nudges/bot-proof CLI commands and their eligibility/marker/rendering/cursor machinery in clawsweeper.ts, and the 911-line policy test. PROOF_NUDGE_MARKER_PREFIX survives: the stalled-unproven close policy dates historical proof asks with it, and the comment-router/dispatcher keep ignoring old nudge comments
State compaction	Monthly history rewrite of openclaw/clawsweeper-state, whose written surface is nearly retired; removing this writer advances the archive goal	state-compaction.yml and src/repair/state-compaction.ts + test. state-repo-size.ts stays (still run by exact-review-reconcile.yml)

Skips (live load-bearing consumers found)

• src/repair/finalize-open-prs.ts module: repair-publish-results.yml (repair cluster, out of scope) runs repair:finalize-open-prs --write-report after every worker-result publish, and publish-result.ts reads results/finalize-open-prs.json. Only the dormant dispatch workflow was deleted.
• Idea-archive revival (idea-archive-revival.yml + src/idea-archive-revival.ts) — skipped entirely: it is the live reversibility half of the unsponsored_feature_request close policy. The apply lane (out of scope) actively parks ideas whose close comments promise "@clawsweeper revive ... will automatically reopen", the close policy imports its threshold/label/reaction helpers, and its run history is green every 6 hours. Deleting it would silently turn reversible parks into irreversible closes; pausing that close policy is a separate decision.

Invariant updates

test/state-writer-workflow.test.ts: setup-state sites 25->19, git publishers 23->19, slug-scoped/undefined hydration lists and non-fatal-bookkeeping lists shrunk to match, and all seven deleted workflows joined the stays-deleted assertion. Sweep/clawsweeper workflow-shape tests, queue-pressure and worker-limit lane tests updated; the internal-route HMAC test now proves the signature gate on the canonical /records/tuples route.

Proof

• pnpm run check green locally (build:all, lint, format:check, limits/active-surface/queue-boundary checks, full unit + coverage suites).
• test:repair green except the pre-existing Linux-only containment/Landlock failures on macOS (Landlock probe, procfs enumeration, containment runtime — unrelated to this change).
• autoreview (Codex, xhigh) clean: "no accepted/actionable findings", overall correct 0.98. autoreview's secret-scanner fails closed when bundling deleted workflow/test files containing secret-shaped env wiring, so the pure file deletions were committed as standalone deletion commits and the full modification diff was reviewed separately.