feat(sync): sync v2 — offline env/labels/observability/deploy-lag checks (closes #141) by robercano-ghbot · Pull Request #191 · robercano/reCode

robercano-ghbot · 2026-07-24T14:27:28Z

Closes #141

Extends /orchestrator:sync into "sync v2" — the sanctioned migration command — by adding four new offline, detect-and-report check functions to sync.sh, in the same defensive style as the existing detect_stale_vendor_copies. sync.sh stays fully offline/non-interactive/bare-runnable; anything needing the network is emitted as an advisory command and documented as an agent step in SKILL.md (live execution via bot-gh.sh).

Added in this PR

Deploy-lag check (issue "Do" Document concurrent config-write safety under parallel workers (re: anthropics/claude-code#29217) #5 gap): reads the target's .claude/state/loop-runs.log and reports whether the last driver run looks active vs idle, reinforcing "re-arm/restart ONLY between drivers".
Environment check (docs: make 'bootstrap first' explicit in GETTING_STARTED #2): offline .env/GH_BOT_TOKEN= key presence (never prints the token value) + installed plugin version vs. the 0.2.2 (Release 0.2.2 — published plugin cache lacks the #131 driver working-directory fix #136 release) floor; prints the live bot-gh.sh api user advisory.
Missing-labels check (feat: seed-issues.sh example + GitHub-Issues-as-tickets #3): derives module:<name> (from the target adapter's modules[]) + needs-human and prints exact bot-gh.sh label create advisory commands, matching the setup skill's pattern; quiets in self-hosting mode.
Observability-plumbing check (gates.json: test_affected has no cheap pnpm-workspace equivalent — add per-stack guidance #4): detects whether .claude/state/worker-tools.jsonl / events.jsonl exist / have activity, surfacing the gap tracked by Ship the worker tool-mirror hook in hooks/hooks.json — cockpit live-workers panel is reCode-only #137 — detection only, does not fix Ship the worker tool-mirror hook in hooks/hooks.json — cockpit live-workers panel is reCode-only #137.
SKILL.md: Flow section documents the new verdict lines; a new "live steps the agent performs" subsection covers the bot-identity + label create-missing steps via bot-gh.sh.
sync-managed-files.test.sh: +19 offline checks (87 total) covering each new check's positive/negative paths, plugin-version at/above/below/malformed/missing, observability absent/empty/active, deploy-lag active/idle/unparseable-ts, token-value-never-leaked, and self-hosting quieting.

Already covered by prior #134 work (verified, NOT rewritten)

"Do" items 1 & 5's core: managed-file re-stamp ladder, detect_stale_vendor_copies, and the systemd-restart migration caveat were already implemented. This PR only added the genuinely-missing loop-runs.log inspection on top.

Out of scope (deliberately)

The underlying Ship the worker tool-mirror hook in hooks/hooks.json — cockpit live-workers panel is reCode-only #137 hook-shipping fix (this PR only detects the gap).
Loop: post-arm canary driver — verify the spawn environment before advancing real issues #132/Cockpit: no consumer-facing entry point — plugin ships cockpit.sh but nothing can invoke it outside the self-hosted repo #133/Escalation: create the needs-human label if missing, fail loudly otherwise #139.
Cross-repo acceptance against the external reDeploy/reDeFi repos — not reachable from this environment. Re-anchored to reCode's own offline test harness, which is self-hosting-safe.

Verification

GATES_FILE=.claude/self/gates.json bash .claude/self/checks.sh test → PASS (all 87 checks passed, test: harness smoke OK). build + lint gates green. Both review lenses (correctness, tests; consensus=all) APPROVE.

🤖 Generated with Claude Code

… observability (issue #141) Extends sync.sh with four report-only, offline, self-hosting-safe checks that close the genuine gaps in issue #141 left after #134's decision (stop vendoring, detect stale copies) was already implemented: (1) reads .claude/state/loop-runs.log to tell the operator whether a driver looks active or idle before they re-arm/restart the loop units, (2) verifies .env/GH_BOT_TOKEN presence (never the value) and the installed plugin version against the #136 release floor, printing the live bot-identity advisory for the agent to run via bot-gh.sh, (3) derives expected module:<name> + needs-human labels from the target's adapter and prints the exact bot-gh.sh create commands as advisories, and (4) detects whether worker-tools.jsonl/events.jsonl observability state exists, referencing issue #137 for the gap without attempting to fix it. Extends sync-managed-files.test.sh with offline coverage for every new check plus self-hosting quieting, and documents the paired live steps in SKILL.md. Co-Authored-By: Claude Sonnet 5 <[email protected]>

…bient GATES_FILE (issue #141) Blocking review findings on the sync v2 offline checks: - sync.sh's deploy-lag last_ts parse was unguarded under set -euo pipefail — a loop-runs.log last line with no ts= token made grep exit 1 and pipefail abort the whole script before the "could not parse a ts=" fallback ever ran. Add the missing `|| true`, matching the sibling parses in the same function. - sync-managed-files.test.sh scenarios 12/12b/15 read the target's adapter via sync.sh's default $GATES_FILE resolution without pinning it to their own fixture, so an ambient GATES_FILE (e.g. the real self-hosted test gate exporting GATES_FILE=.claude/self/gates.json before running *.test.sh) leaked in and broke 4 checks. Pin GATES_FILE explicitly per fixture, same convention as gate.test.sh's write_gates_file/run_gate. Also folds in the non-blocking asks: explicit coverage for an above-floor plugin version and a malformed/missing version field in the environment check, and present-but-empty / present-with-activity coverage for events.jsonl (previously only worker-tools.jsonl was exercised beyond the absent case). Co-Authored-By: Claude Sonnet 5 <[email protected]>

robercano-ghbot · 2026-07-24T14:29:46Z

feat(sync): sync v2 — offline env/labels/observability/deploy-lag checks (closes #141) (not yet reviewed)

robercano-ghbot · 2026-07-24T15:19:20Z

feat(sync): sync v2 — offline env/labels/observability/deploy-lag checks (closes #141) (not yet reviewed)

…ger alone (issue #141) loop-daemon.sh only appends a loop-runs.log line AFTER a driver exits, so the ledger has zero signal for an in-flight driver -- a stale OR missing OR just- appended entry is equally consistent with a new driver running right now. check_deploy_lag's recent/old-ts branching claimed the opposite ("may still be ACTIVE" / "looks safe to re-arm/restart"), which is exactly backwards. All three branches now report the ledger state plainly and defer to an independent liveness check (systemctl --user status) instead of asserting a verdict the log can't actually support. Also strengthens the s12 "labels are advisory-only" test from a vacuous assertion into a real regression guard (stubs bot-gh.sh/gh both via PATH and via the literal ${CLAUDE_PLUGIN_ROOT}/scripts/bot-gh.sh path the advisory string itself uses), and adds version_ge differing-length-component coverage.

robercano-ghbot · 2026-07-24T16:13:03Z

Stall-recovery update: this branch had partial work + an already-open PR with green CI, but no GitHub review yet. Ran both reviewer lenses (correctness, tests; consensus=all) fresh against the diff before publishing further.

correctness caught a real bug: check_deploy_lag() read a recent loop-runs.log entry as "driver may still be ACTIVE" and an old one as "looks idle — safe to restart". That's backwards — loop-daemon.sh only appends a ledger line after a driver exits, so the ledger has zero signal about an in-flight driver; a stale (or missing, or empty) ledger is equally consistent with a new driver running right now. All three branches (not-found / empty / age-based) now report the ledger plainly and defer to an independent check (systemctl --user status 'pr-loop-driver-*') instead of asserting a verdict the log can't support. Took two more review rounds to catch this in the not-found and empty-ledger branches too (same flaw, different branch).

tests caught a vacuous assertion (test 1 -eq 1, always true) standing in for "labels are advisory-only, no network call". Replaced with a real regression guard: stubs bot-gh.sh/gh both via PATH and via the literal ${CLAUDE_PLUGIN_ROOT}/scripts/bot-gh.sh path the advisory string itself uses (a PATH-only stub was proven to miss that exact construction). Also added version_ge differing-length-component coverage.

All 93 offline checks pass; build/lint/test gates green (GATES_FILE=.claude/self/gates.json); both lenses APPROVE on the final diff. Not merging — still waiting on your review.

robercano and others added 2 commits July 24, 2026 16:08

robercano-ghbot assigned robercano Jul 24, 2026

robercano-ghbot requested a review from robercano July 24, 2026 14:27

robercano-ghbot added the needs-human Loop escalated: attempt budget exhausted, manual triage required label Jul 24, 2026

robercano-ghbot added needs-human Loop escalated: attempt budget exhausted, manual triage required and removed needs-human Loop escalated: attempt budget exhausted, manual triage required labels Jul 24, 2026

robercano approved these changes Jul 24, 2026

View reviewed changes

robercano-ghbot removed the needs-human Loop escalated: attempt budget exhausted, manual triage required label Jul 24, 2026

robercano-ghbot merged commit 8141238 into main Jul 24, 2026
9 checks passed

robercano-ghbot deleted the feat/issue-141-sync-v2-checks branch July 24, 2026 18:39

robercano-ghbot mentioned this pull request Jul 24, 2026

Split needs-human into two labels: routine 'needs-review' vs genuine 'needs-human' escalation #192

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat(sync): sync v2 — offline env/labels/observability/deploy-lag checks (closes #141) - #191

feat(sync): sync v2 — offline env/labels/observability/deploy-lag checks (closes #141)#191
robercano-ghbot merged 3 commits into
mainfrom
feat/issue-141-sync-v2-checks

robercano-ghbot commented Jul 24, 2026

Uh oh!

robercano-ghbot commented Jul 24, 2026

Uh oh!

robercano-ghbot commented Jul 24, 2026

Uh oh!

robercano-ghbot commented Jul 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

robercano-ghbot commented Jul 24, 2026

Added in this PR

Already covered by prior #134 work (verified, NOT rewritten)

Out of scope (deliberately)

Verification

Uh oh!

robercano-ghbot commented Jul 24, 2026

Uh oh!

robercano-ghbot commented Jul 24, 2026

Uh oh!

robercano-ghbot commented Jul 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants