tool-scout

Reads meetings cache, greps transcripts for known tool keywords (Xano, Fly, Neon, Cloudflare, Supabase, Vercel, Claude Code, etc.), and cross-checks state/lib/ + state/skills/ for coverage.

Steps

1. Read meetings cache from ~/.claude/cache/krisp/.
2. For each transcript, regex-match the tool keyword set.
3. For each mentioned tool, check whether a state/lib/<tool>.ts or

state/skills/<tool>.md exists.

1. Emit a gap report: {mentioned, covered, gaps}.

Eval

score("tool-scout", run_id, {
  score: 1.0,
  mentioned: mentioned.length,
  covered: covered.length,
  gaps: gaps.length,
});

No pass/fail - this is a telemetry skill. The score is always 1.0 if the shape gate passes.

Gotchas

• Keyword regex will false-positive on casual mentions ("I was thinking

about trying Supabase"). Dispatch a gemini pass to drop those.

Rubric

criteria:
  - name: transcript_processing
    kind: deterministic
    check: "The skill must read meetings cache from `~/.claude/cache/krisp/` and process all available transcripts."
  - name: keyword_matching
    kind: deterministic
    check: "The skill must correctly identify instances of known tool keywords (Xano, Fly, Neon, Cloudflare, Supabase, Vercel, Claude Code) within the transcripts using regex matching."
  - name: coverage_check
    kind: deterministic
    check: "For each mentioned tool, the skill must accurately check for the existence of `state/lib/<tool>.ts` or `state/skills/<tool>.md`."
  - name: gap_report_accuracy
    kind: deterministic
    check: "The skill must emit a gap report with correctly counted 'mentioned', 'covered', and 'gaps' fields as described in the 'Eval' section."

tool-scout - loader

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

Critical Rules

• ALWAYS post-filter regex matches with a Gemini dispatch - keyword regex false-positives on casual mentions ("I was thinking about trying Supabase") and pollutes the gap report
• ALWAYS check BOTH state/lib/<tool>.ts AND state/skills/<tool>.md for coverage - a tool with one but not the other counts as a partial gap, not full coverage
• This is a TELEMETRY skill - the score is always 1.0 if the shape gate passes; do not invent a pass/fail rubric on top
• ALWAYS read meetings cache from ~/.claude/cache/krisp/ (written by state/bin/krisp/refresh.sh); if the cache is empty, log primary_issue: "krisp-cache-missing" and emit zero-mention shape, never silently no-op

Commands

| ui dashboard | state/skills/tool-scout/resources/ui.openui | |invoke: prose skill - follow steps in state/skills/tool-scout/SKILL.md |krisp cache: ~/.claude/cache/krisp/ |coverage check: state/lib/<tool>.ts and state/skills/<tool>.md |eval log: state/log/evals.ndjson (skill: "tool-scout") - {score: 1.0, mentioned, covered, gaps}

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/tool-scout/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

• Casual mentions ("we should probably try Supabase someday") get counted as real demos without a Gemini filter pass
• Counting only library coverage and missing skill-page coverage (or vice versa) under-reports gaps
• Treating tool-scout as pass/fail instead of telemetry - the value is the gap list, not a 0/1 score

Self-Test

An agent reading this should correctly:

1. [ ] Run a Gemini dispatch over regex matches to drop casual mentions?
2. [ ] Check both state/lib/ and state/skills/ for a tool before declaring it covered?
3. [ ] Score 1.0 whenever shape passes, treating this skill as telemetry not pass/fail?

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] tool-scout: <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)] tool-scout: <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.