Backed by: state/lib/log.ts (reads ~/.claude/cache/krisp/ dump) + state/lib/eval.ts (writes the score row)

krisp-inbox

Reads ~/.claude/cache/krisp/meetings.json, groups by participant, and emits a "who's waiting on what" summary. Complement to ray-todo (which covers commitments); this covers follow-ups that never became commitments.

Steps

1. Read meetings cache (refresh if stale).
2. For each meeting, dispatch gemini with transcript (if available) +

prompt "list follow-ups Robert owes the other participant."

1. Dedupe against ray-todo output for the same date.
2. Write to state/log/krisp-inbox/<date>.md.

Eval

Auto. Shape gate on the produced summary file:

Score 1.0 if all three pass, 0.5 if file written but missing dedup line, 0.0 if no file. Eval row written via state/lib/eval.ts.

Graduation

Prose; no script until the dedup logic is stable.

krisp-inbox - loader

Per-turn rules for the krisp-inbox skill. Full reference: state/skills/krisp-inbox.md. Do not skip these.

Critical Rules

• ALWAYS dedup against ray-todo for the same date - krisp-inbox covers follow-ups that never became commitments, ray-todo covers commitments; overlap must be reported (even as 0)
• ALWAYS write the dedup count to the output file - the shape gate's dedup_run check requires explicit overlap mention
• ALWAYS use ~/.claude/cache/krisp/meetings.json - Krisp is the one documented MCP exception in program.md "Don't list", isolated to this daily cron dump
• NEVER skip writing the file - score 0.0 if no state/log/krisp-inbox/<date>.md is produced

Commands

|invoke: skill is prose, runs the steps in state/skills/krisp-inbox.md (read cache → dispatch gemini per meeting → dedup → write) |cache refresh: bash state/bin/krisp/refresh.sh (cron-driven dump) |eval log: state/log/evals.ndjson (skill: "krisp-inbox")

Known Pitfalls

• Cache may be stale if krisp/refresh.sh cron didn't run - refresh before scoring rather than emitting an empty inbox
• Output must contain at least one - [ ] checkbox row OR an explicit "no follow-ups" sentinel - silent empty fails the has_followups shape check

Self-Test

An agent reading this should correctly:

1. [ ] Cite the ray-todo overlap count in the output file even when the count is 0?
2. [ ] Read the Krisp cache from ~/.claude/cache/krisp/meetings.json?
3. [ ] Score 0.5 if file written but dedup line missing, 1.0 if all three shape checks pass?

Self-report

If this loader fell short, append a line:

echo "[$(date -u +%FT%TZ)] krisp-inbox: <what was missing>" >> ~/.claude/logs/snappy-os-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)] <skill-name>: <what was missing or fixed> [FIXED|LOGGED]" >> state/log/agents-md-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.