meeting-triage

The feel-it-now companion to /snappy-ops (the idle-moment proof of life). /snappy-ops tells you the system is breathing; meeting-triage makes you feel the system DO something - within seconds of a Krisp cache refresh the byline shows you what you just committed to, ranked by urgency.

Usage

The script is fired by launchd (com.snappy.meeting-triage.plist, WatchPaths = ~/.claude/cache/krisp/action-items.json) - it runs the instant new action items land, not on a polling cadence. Invoke directly when you want an ad-hoc triage:

npx tsx state/bin/meeting-triage/run.ts

Output is a single JSON line on stdout:

{"ok":true,"refreshed":true,"pending":8,"new":3,"skipped_seen":5,"followups_log":"...","evals_log":"..."}

What it does

1. Read ~/.claude/cache/krisp/action-items.json. If older than 60 min, call

fetchActionItems() from state/lib/krisp.ts and refresh the cache.

1. Filter to items where Robert owns it (first-name-fuzzy OR title starts with

"Robert to …") AND completed == false.

1. Rank urgency by the rules in rankUrgency() - explicit keywords ("today",

"this week", "asap") win; otherwise fall through to due_date delta, and finally to meeting recency (anything from a meeting <48h old defaults to "this week").

1. Dedupe against IDs already written to state/log/followups.ndjson. Krisp

IDs are stable, so re-firing is cheap.

1. Append one row per new commitment. Eval row logged either way.

followups.ndjson row shape

{
  "id": "019d8e0b6cec7275924fceb0813a058a",
  "meeting_id": "019d8dcb56bf739dac4778f4c53b54a2",
  "meeting_title": "Robert Boulos <> Mark",
  "committed_by": "robert",
  "text": "Robert to announce in Slack when the PR has been pushed.",
  "urgency": "this week",
  "status": "open",
  "ts_created": "2026-04-18T18:42:12.108Z",
  "ts_resolved": null
}

/snappy-triage (the picker) appends a new row with the same id and an updated status when the operator acts on an item. Read-time logic takes the latest row per id.

Byline row

The byline (state/bin/statusline.sh) reads followups.ndjson, takes the latest row per id, counts rows where status == "open" AND ts_created is within the last 2 hours. When the count is > 0 it emits:

📬 meeting: N items · → /snappy-triage

After the operator works through them, the rows flip to status != open and the byline row drops.

Eval

Shape-gate (what this skill guarantees per run):

• An eval row is appended on every run, success or failure.
• On success the row contains {score: 1, pending_count, new_count, refreshed}.
• On failure (no cache, Krisp auth expired) the row contains

{score: 0, error: "..."} so the eval-coverage lint surfaces it rather than hiding it behind an empty byline.

Outcome-gate (over time):

• Median time from meeting end → first 📬 meeting: byline render ≤ 3 min

on the Mac Mini (cron owner). Measured by diffing ts_created on the oldest open followup against the meeting's meeting_date.

Gotchas

• assignee: null is common - Krisp drops it. pendingActionItemsFor already

handles the title-prefix fallback (titleAssignedTo) so items like "Robert to review …" still surface even without a structured assignee.

• The cache refresh trips over expired Krisp OAuth (the same way

fetchLinkedInDMs did). fetchActionItems refreshes the token automatically; if that fails, this script falls back to the stale cache and logs error: "refresh-failed" on the eval row so inbox-sweep-style silence is impossible.

• rankUrgency is deterministic. No dispatch call, no LLM. If we later need

nuance, add a --model=gemini flag and route the ranking through state/lib/dispatch.ts, but the v1 bar is "ship today, feel it tomorrow."

Graduation

Already graduated to script - state/bin/meeting-triage/run.ts + state/bin/meeting-triage/watch.sh. No prose-to-script ladder because the classifier already has a working reuse target (ray-todo.md's filter).

Part of the plan

Commit 2 of ~/.claude/plans/read-the-plan-nifty-wolf.md. Paired with /snappy-triage (the picker) and the 📬 meeting byline row.

Rubric

criteria:
  - name: eval_row_appended
    kind: deterministic
    check: "The command `tail -n 1 state/log/evals.ndjson` shows a new line with an 'eval' field after script execution."
  - name: robert_items_filtered
    kind: judge
    check: "All entries in `state/log/followups.ndjson` generated by this skill refer to Robert's open commitments, either by fuzzy first name match or by title starting with 'Robert to …' and `status == 'open'`."
  - name: new_open_followups_logged
    kind: judge
    check: "New action items for Robert from the Krisp cache, not previously in `state/log/followups.ndjson`, are appended to `state/log/followups.ndjson` with `status: \"open\"`."
  - name: cache_refresh_conditional
    kind: deterministic
    check: "If `~/.claude/cache/krisp/action-items.json` is older than 60 minutes, `fetchActionItems()` from `state/lib/krisp.ts` is called to refresh the cache."

meeting-triage - loader

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

Critical Rules

• assignee: null is common - Krisp drops it. pendingActionItemsFor already

Commands

| ui dashboard | state/skills/meeting-triage/resources/ui.openui | |invoke: see state/skills/meeting-triage/SKILL.md Steps section |eval log: state/log/evals.ndjson (skill: "meeting-triage")

Self-Test

An agent reading this should correctly:

1. [ ] Know which lib/bin artifact backs this skill (or that it is prose-only)
2. [ ] Know what to write to state/log/evals.ndjson after invoking
3. [ ] Know the eval mode (auto / shape / manual) from the .md frontmatter

Self-report

If this loader fell short, append a line:

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

OpenUI Resource

• Skill-owned OpenUI Lang resource: state/skills/meeting-triage/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.