Backed by: state/lib/sweep.ts + state/lib/xano.ts

commitment-audit

Reads ray-todo + sweep snapshot + git log across active client repos. For each open commitment, checks whether a matching commit or Xano row exists. Flags orphans (promised, no delivery) and silent-ships (delivered, never told the client).

Steps

1. Load open commitments from ~/.claude/cache/krisp/action-items.json.
2. git log --since="$since" across ~/projects/snappy-client-* repos.
3. For each commitment, fuzzy-match commit message keywords + Xano

row-created timestamps.

1. Emit {orphaned, silent_shipped, matched} triage to

state/log/commitment-audit/<date>.md.

Eval

score("commitment-audit", run_id, {
  score: orphaned.length + silent_shipped.length > 0 ? 1.0 : 0.5,
  orphaned: orphaned.length,
  silent_shipped: silent_shipped.length,
  matched: matched.length,
});

0.5 for empty output (valid - nothing to audit). 1.0 when surfaces a real triage item. 0.0 on shape failure.

Gotchas

• git log across 10+ client repos is slow; cache per-repo HEAD hash and

skip unchanged ones.

• Fuzzy match is keyword overlap, not semantic - tune threshold per repo.

Rubric

criteria:
  - name: output_file_present
    kind: deterministic
    check: "The file 'state/log/commitment-audit/<date>.md' exists after execution."
  - name: output_file_format
    kind: judge
    check: "The content of 'state/log/commitment-audit/<date>.md' clearly separates 'orphaned', 'silent_shipped', and 'matched' items as specified."
  - name: commit_xano_check_performed
    kind: judge
    check: "The output in 'state/log/commitment-audit/<date>.md' demonstrates an attempt to match commitments against both git commits and Xano rows."
  - name: eval_score_correctness
    kind: deterministic
    check: "The `eval` score reflects `orphaned.length + silent_shipped.length > 0 ? 1.0 : 0.5` based on the content of 'state/log/commitment-audit/<date>.md'."

commitment-audit - loader

Per-turn rules for the commitment-audit skill. Full reference: state/skills/commitment-audit/SKILL.md. Do not skip these.

Critical Rules

• ALWAYS cache per-repo HEAD hash and skip unchanged repos - git log across 10+ client repos is slow
• ALWAYS surface BOTH orphaned (promised, no delivery) AND silent_shipped (delivered, never told the client) - silent ships are as bad as orphans
• 0.5 is a valid score when there's nothing to audit - empty output is not a failure
• Fuzzy-match is keyword overlap (Jaccard-style), NOT semantic - tune threshold per repo if it under-matches

Commands

| ui dashboard | state/skills/commitment-audit/resources/ui.openui | |invoke: load Krisp action-items, parallel git-log across ~/projects/snappy-client-*, fuzzy-match against commits + Xano timestamps, write state/log/commitment-audit/<date>.md |verify: shape-gate; score: orphaned + silent_shipped > 0 ? 1.0 : 0.5 (0.0 only on shape failure) |eval log: state/log/evals.ndjson (skill: "commitment-audit")

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/commitment-audit/resources/ui.openui. Read it before rendering or editing this skill's generated component surface.
• Treat this resource as a first-class artifact of the skill, not a generic chat response. Improve it when the skill's user-facing output needs to become richer.
• System resources compose OpenUI primitives and inherit SnappyChat tokens. Use ui_contract: branded in SKILL.md only for deliberate platform or client visuals.

Known Pitfalls

• Per-repo Jaccard threshold drift - what works for snappy-client-ray may under-match for snappy-client-orbiter. Tune per repo, not globally.
• Krisp action-items cache lives at ~/.claude/cache/krisp/action-items.json - refresh it via state/bin/krisp/refresh.sh before audit if stale
• Silent-ship detection is the load-bearing surface - if you're tempted to drop it because output looks redundant, don't (memory: feedback_audit_before_push)

Self-Test

An agent reading this should correctly:

1. [ ] Skip an unchanged repo via HEAD-hash cache rather than re-running git log
2. [ ] Score an empty audit at 0.5, not 0.0
3. [ ] Surface silent-ships even when orphans is empty

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] commitment-audit: <what was missing>" >> state/log/loader-feedback.log

<!-- kernel-ok: historical attribution for the PID-loader pattern. -->

Self-correcting loader (PID feedback)

This is the rule that makes loaders learn. Ported from snappy-kernel, where it demonstrably worked: the snappy-dom-cartographer skill mapped Skool admin UI once, every later ship-agent inherited the full map for free.

Before you finish, do two things:

1. Fix gaps (P - proportional). MANDATORY. If this loader didn't cover your case - if you had to read another file, run an undocumented command, work around a wrong selector, or discover a quirk - you MUST attempt an Edit to this AGENTS.md before you log.

• Default action: edit this file. Surgical. Add the missing line, table

row, or code example where it belongs. No restructuring.

• Only LOGGED is allowed when: the fix needs >10 lines, spans multiple

files, or requires a structural rewrite. In that case the state/regen/drain.sh queue picks it up asynchronously.

• "I didn't have time" / "it's minor" / "the next agent will figure it out"

are NOT valid reasons. Minor gaps compound - that's the whole point of this loop.

• Concrete example of a valid P-fix: discover agent-browser screenshot

takes path positionally not as --output <path>. One line in a table. Edit it in. Do not just log it.

• The goal: the next agent never has to leave the loader.

2. Log the result.

echo "[$(date -u +%FT%TZ)] commitment-audit: <what was missing or fixed> [FIXED|LOGGED]" >> state/log/loader-feedback.log

• FIXED = you patched this loader inline (P-fix).
• LOGGED = too large for inline; the PostToolUse enqueue + Stop-hook drain

will rewrite the loader from scratch on next session-end.

Do not skip this. Every agent run must leave the system better than it found it. The loader is the setpoint; you are the sensor; the gap is the error signal; closing the gap is the correction.