agentsclimarketplace
Sign in

Loop health

Skill thepictishbeast/claude-tools/skills/loop-health

Single-shot lint diagnostic across all active + paused loops and the history file. Reports missing canaries, oversized prompts, stale entries, orphan history events, malformed state files, and rotation candidates. Use when the user says "check loop health", "audit my loops", "lint my loop state", or "anything wrong with my loops?"From its SKILL.md

Install
npx -y skills add thepictishbeast/claude-tools --skill loop-health

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.6 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

/loop-health — lint diagnostic across all loop state

What this checks

Single pass, no mutation. Reports findings; user decides what to fix.

A. Active cron loops (from CronList)

For each:

  • A1. Untracked in history — no created / resumed / discovered event matches this cron's ID. Suggest /loop-track <id> or just call /loops (which auto-discovers as a side effect).
  • A2. Missing canary — prompt doesn't contain the "if you canceled or stopped this loop, you should NOT be seeing this message" pattern. Suggest /loop-edit prompt-append: "Note: if you canceled or stopped this loop, you should NOT be seeing this message.".
  • A3. Oversized prompt — prompt > 800 chars. Verbose prompts make each fire context-heavy; cleaving steady-state instructions into the prompt body wastes tokens on every fire.
  • A4. Self-firing canary triggered — if any recent history event shows this cron was paused or stopped AFTER its latest fire, but the cron still exists in CronList — the cancel didn't take. This is a real bug and warrants user attention.

B. Paused loops (from .paused-loops.json)

For each:

  • B1. Stale paused — paused_at is more than 5 days ago. The user probably forgot. Suggest /loop-resume or /loop-stop.
  • B2. Missing canary — same as A2.
  • B3. Oversized prompt — same as A3.
  • B4. inflight_tasks orphaned — entry has inflight_tasks but those tasks (by recorded ID) don't exist in current TaskList. Expected when paused state predates the current session. Just a note, not a problem.
  • B5. Multiple paused with identical prompt — paul merged twice. Suggest dedupe.

C. History file (loop-history.jsonl)

  • C1. Size > 1 MB — rotation candidate. Suggest archiving to loop-history-YYYY-MM.jsonl.gz and starting a fresh log.
  • C2. Malformed lines — non-JSON or missing required fields (event, at). Report line numbers; user manually fixes.
  • C3. Future-dated events — at timestamp in the future. Clock skew or paste error.
  • C4. Discovered-truncated events — when /loops auto-discovery found a cron with truncated prompt and couldn't capture full text. Suggest manual /loop-track <id> with full prompt.

D. Lock files (.paused-loops.lock, loop-history.lock)

  • D1. Stale lock — flock file present and older than 5 minutes with no holder process. Suggest rm to clear.

E. Auto-update freshness

  • E1. Toolkit out of date — if the locally-cloned claude-tools repo exists, check git fetch --dry-run output for incoming commits. Report count.
  • E2. Stale skill rename leftovers — ~/.claude/skills/ contains a directory that's no longer in upstream's skills/ (e.g. old pause/, resume/). Suggest removal.

Output format

Loop health: <N> findings (<M> warnings, <K> info)

A. Active loops:
  A2 [warn] cron 04b00904: missing canary
  A4 [bug]  cron 04b00904: previously paused 2026-05-17 06:55Z but still appears in CronList

B. Paused loops: (no findings)

C. History (124 KB): (no findings)

D. Lock files: (no findings)

E. Toolkit:
  E2 [info] ~/.claude/skills/pause/ — stale rename leftover (upstream uses loop-pause)

Run /loops to see current state. No mutations were made by this check.

Steps

  1. Snapshot all sources: CronList, ~/.claude/.paused-loops.json, ~/.claude/loop-history.jsonl (last 200 lines + size), ls ~/.claude/skills/, ls ~/.claude/*.lock.
  2. Run each check in order. Each is independent and read-only.
  3. If TaskList available, also include B4 inflight-task check.
  4. If a local clone of claude-tools exists, git fetch --dry-run for E1.
  5. Emit findings in the format above.
  6. Exit summary line: "Run X to fix Y, Z to fix W" guiding the user.

Don't

  • Don't mutate state. This skill is read-only.
  • Don't ask the user to confirm anything mid-run; just report and exit.
  • Don't try to auto-fix findings; emit suggestions, let user decide.

See also

  • /loops — current state view (also auto-discovers, which is a mild mutation; /loop-health is purely read-only)
  • /loop-track — fix A1 findings explicitly
  • /loop-edit prompt-append: … — fix A2 / B2 missing-canary findings

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most quality gates skills give in ~1.2k tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-07

  • Read the output and check the exit codein 54 of 1195, across 14 files
  • Verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
  • Identify the verification command proving the claimin 51 of 1195, across 12 files
  • Run the full verification commandin 50 of 1195, across 11 files
  • Verify output confirms the claimin 49 of 1195, across 12 files
  • Check version control diff after agent delegationin 46 of 1195, across 6 files
  • State claim with evidencein 44 of 1195, across 4 files
  • Run the test suitein 33 of 1195, across 26 files
  • Keep state in memory by defaultin 27 of 1195, across 6 files
  • Make prototype runnable with one commandin 26 of 1195, across 5 files
  • Produce a verification reportin 25 of 1195, across 14 files
  • Detect the package manager from lockfilesin 24 of 1195, across 5 files

Said here and by no other author read

  • snapshot all sources first
  • run each check independently
  • check for missing canary patterns
  • check for oversized prompts
  • check for stale paused loops
  • check history file size

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,851. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.