feat(harness): cron-less loop daemon under systemd (issue #102) - #103
Merged
Conversation
Replaces the session-scoped CronCreate loop with a systemd-supervised daemon: loop-event.sh runs the deterministic tick and parses its verdict (never re-deriving it, per issue #81) without touching any model process; loop-daemon.sh owns the setsid+timeout driver containment, adaptive FAST/WATCH/IDLE sleep from the census cadence line, and the .claude/state/loop-runs.log run ledger. action=none spawns zero drivers. Scaffolds systemd unit templates (pr-loop.service, claude-rc.service) and an arm-loop.sh installer through the existing @orchestrator-managed marker machinery (scaffold.sh/sync.sh/MANIFEST.md), and wires setup's "arm the loop" step to offer the daemon (Linux/WSL2-aware) alongside the legacy /pr-loop cron, which now points at the daemon as the recommended replacement. Co-Authored-By: Claude Fable 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01EtQ3Zexq39jTL1TSYmoGxi
USAGE.md gains a "Cron-less loop (daemon)" section covering the pr-loop.service/claude-rc.service architecture, adaptive FAST/WATCH/IDLE cadence, the loop-runs.log ledger, read-only supervision (ledger + live JSONL transcript + --resume --fork-session), kill-and-let-it-re-advance intervention, WSL2's systemd=true prerequisite, and the failure-contract table -- including the one non-self-healing state (a driver's abandoned feat/issue-N-* branch wedges ADVANCE for that issue until a human deletes the branch or opens the PR). GETTING_STARTED.md's onboarding summary and "arm the loop" checklist step now lead with the daemon as the recommended default, with the legacy /orchestrator:pr-loop cron kept as the fallback for environments without systemd. Co-Authored-By: Claude Fable 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01EtQ3Zexq39jTL1TSYmoGxi
robercano
approved these changes
Jul 9, 2026
robercano
approved these changes
Jul 9, 2026
4 tasks
robercano
added a commit
that referenced
this pull request
Jul 10, 2026
…f-hosted repo (issue #102 residue) PR #103 shipped these as setup templates; sync only re-stamps existing copies, so the self-hosted repo needs them materialized once. Plain copy_if_absent semantics, markers intact (v1). Co-Authored-By: Claude Fable 5 <[email protected]>
This was referenced Jul 10, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #102.
Summary
Replaces the session-scoped
CronCreatePR loop with asystemd --user-supervised daemon, plus docs for the whole architecture.Harness (previous phase, already reviewed/approved):
.claude/scripts/loop-event.sh— one firing: runs the deterministic tick (loop-tick.sh), parses its LAST-line verdict byte-identical (issue Harden PR-loop tick: deterministic loop-tick.sh with single verdict, in-flight detection, spawn lock #81 contract — never re-derived), and emits aloop-event: action=...block. Touches no model/driver process itself..claude/scripts/loop-daemon.sh— the forever loop. Each iteration runsloop-event.sh; onaction=noneit sleeps and loops with zero drivers spawned; on an actionable verdict it spawns exactly one contained driver (setsid timeout --kill-after=30s <90m default> claude -p ... --output-format json, whole-process-group timeout kill), ledgers it to.claude/state/loop-runs.log, then sleeps for an adaptive duration read off the census'scadence=FAST|WATCH|IDLEline (60/300/900s, 300s fallback).pr-loop.service,claude-rc.service) +arm-loop.shinstaller, wired through the existing@orchestrator-managedscaffold/sync marker machinery, and offered from/orchestrator:setup's "arm the loop" step (Linux/WSL2-aware) alongside the legacy/pr-loopcron.This phase (docs):
docs/USAGE.md— new "Cron-less loop (daemon)" section: architecture (pr-loop.service→loop-daemon.sh→loop-event.sh/loop-tick.sh;claude-rc-<repo>.service→ tmux →claude remote-controlfor spawning planning sessions from claude.ai/mobile), the adaptive cadence table, theloop-runs.logledger line format, read-only supervision (ledger + live append-only JSONL transcript tail +claude --resume <id> --fork-session), the kill-and-let-it-re-advance intervention model (never steer a running driver), Linux vs WSL2 setup differences (systemd=truein/etc/wsl.conf+ optionalschtasksWindows-logon autostart), and a failure-contract table — including the one non-self-healing state: a driver pushesfeat/issue-N-*but dies before opening the PR → census reportsNasin_flight→loop-tick.sh's advance check refuses every tick (logged each time,action=none) with no TTL/self-heal → manual fix is to delete the abandoned branch or open the PR by hand.docs/GETTING_STARTED.md— the onboarding summary and the "arm the loop" checklist step now lead with the daemon as the recommended default; the legacy/orchestrator:pr-loopcron stays documented as the fallback for environments without systemd. Never run both firing sources against the same repo at once (safe —loop-tick.sh's spawn lock prevents a double-spawn — just wasteful).Gates & review
GATES_FILE=.claude/self/gates.json gate.sh build|lint|testall green (docs-only diff this phase; harness phase's own gates — includingloop-daemon.test.sh/loop-event.test.sh— passed and were reviewed in the prior phase).correctnessandtestslenses, both approve, consensusall.🤖 Generated with Claude Code
https://claude.ai/code/session_01EtQ3Zexq39jTL1TSYmoGxi