foundation-meeting-recap

Files (5)

• HISTORY.md
• SKILL.md
• trigger-fixtures.json · evals/
• EXAMPLE.md · references/
• TEMPLATE.md · references/

SKILL.md (first lines)

<!-- PM-Skills | https://github.com/product-on-purpose/pm-skills | Apache 2.0 -->
# Meeting Recap

A meeting recap is a post-meeting topic-segmented summary produced for attendees and light distribution. It organizes content by topic rather than chronology, highlights decisions visually, and captures actions inline (with owner, due date, dependencies) per topic segment, plus a consolidated actions view at the end for scannability.

This skill absorbs what would otherwise be a separate "meeting-actions" skill. Actions in this family live alongside the context that makes them meaningful, not in a sibling artifact.

This skill belongs to the Meeting Skills Family. It conforms to the [Meeting Skills Family Contract](../../site/src/content/docs/reference/skill-families/meeting-skills-contract.md).

## When to Use

- After any internal meeting that produces decisions or actions affecting attendees
- When a sibling `foundation-meeting-agenda` exists and needs reconciliation (planned vs. actual topics)
- When the team needs a topic-organized reference rather than a chronological transcript dump
- When inputs include a transcript (Zoom, Meet, Otter, Fireflies, Krisp MCP) or mixed notes and transcript

## When NOT to Use

- Setting the pre-meeting structure attendees will see (topics, owners, time boxes) before the meeting happens. Use `foundation-meeting-agenda`. recap summarizes what happened after a meeting; agenda sets expectations before it.
- Preparing the user's private strategic positioning before a meeting that matters. Use `foundation-meeting-brief`. recap is a shared post-meeting record; brief is pre-meeting tactical prep never shown to attendees.
- Communicating outcomes to non-attendees. Use `foundation-stakeholder-update`. recap assumes reader context; stakeholder-update translates to readers without it.
- Cross-meeting synthesis (patterns across multiple meetings). Use `foundation-meeting-synthesize`.
- Synthesizing user research interviews across participants, not one meeting's attendees. Use `discover-interview-synthesis`. recap organizes a single meeting by topic; interview-synthesis finds patterns across many research conversations.
- Live meeting note-taking. This skill consumes finished inputs; it does not transcribe live.

## Zero-friction execution

Per the family contract, this skill never blocks on interrogation. Default flow:

1. Read all provided inputs (transcript, notes, or hybrid) and note input quality upfront
2. Auto-discover related agenda via filename-prefix match on same-directory `*_{title}_agenda.md`
3. Run inference: meeting metadata from content, decisions from language markers, actions from imperative-future patterns, owners from attendee context
4. Present a brief inference summary and accept one-word `go` or corrections
5. Produce the recap

If invoked with `--go`, skip the inference summary. If the user provides all metadata upfront, no checkpoint appears.

## Fabrication prohibition

This skill never fabricates owners, decisions, or actions. When an action lacks an explicit owner, it is captured as `[owner: unassigned, needs confirmation]` not invented. When a decision is implicit ("it sounded like we decided X"), it is flagged with a confidence marker rather than stated as fact. Trust decay from fabrication is worse than the mild friction of flagging.

## Ownership reconciliation threshold (v1.1.0)

When the ratio of unassigned actions to total actions exceeds **0.3** (30%), OR when any high-priority action lacks an owner, the skill surfaces a dedicated `## ⚠ Ownership reconciliation required` section at the top of the recap (above the topic segments) listing:

- All unassigned actions
- A suggested next step for each (who should probably own this based on topic context, flagged as inference)
- A recommended follow-up action (Slack thread, 15-min sync, async survey)

The shareable summary also leads with this flag when triggered: `⚠ Ownership reconciliation required: N of M actions lack owners.`

The `unassigned_action_ratio` frontmatter field (float 0.0-1.0) records the ratio for downstream tools.

Rationale: a recap with 60% ownerless actions is "non-fabricated" (per the prohibition above) but operationally broken. a pile of broken tickets. The threshold makes this visible instead of silently shipping.

## Instructions

When asked to create a meeting recap, follow these steps:

1. **Parse inputs and detect type**
   Transcript (timestamped speaker-attributed lines), notes (bullet or prose), or hybrid. Note input quality upfront. transcript plus structured notes is high; scrappy bullets is low.

2. **Auto-discover related agenda**
   Look in the same directory for a file matching the pattern `{YYYY-MM-DD}_{HH-MMtimezone}_{title-slug}_agenda.md`. If found, load it. its topic list is the recap's topic skeleton and its `desired_outcomes` drive the meeting-quality reconciliation.

3. **Present go-mode inference summary**
   Show detected meeting date, title, attendees (if inferred), input quality assessment. Accept `go` or corrections.

4. **Topic-segment the content**
   - If agenda was loaded, use its topic list as scaffold (plus any emergent topics)
   - If no agenda, identify topics from transcript discourse markers ("moving on to", "the other thing")

5. **Per topic segment, extract**
   - **Discussion summary**: 2-3 sentences capturing what was discussed
   - **Decisions made**: bold-flagged visually. Never fabricate. if uncertain, flag "appears to have decided X [confidence: medium]"
   - **Actions**: owner + due date + dependencies. Flag missing owners as `[owner: unassigned]` and missing dates as `[due: not specified]`. Never invent.
   - **Open questions**: unresolved items with confidence marker on whether they are truly unresolved or simply not re-raised

6. **Consolidate actions by owner**
   Regroup all actions under each owner. Enables single-owner scan ("what do I owe after this meeting?").

7. **Reconcile agenda** (if agenda loaded)
   - `topics_planned`: from agenda topic list
   - `topics_hit`: topics actually discussed
   - `topics_skipped`: planned but not discussed, with brief reason
   - `topics_emerged`: discussed but not on agenda

8. **Assess meeting quality signals**
   - `outcomes_achieved`: N/M ratio of agenda `desired_outcomes` met (when agenda present)
   - `started_on_time` / `ended_on_time`: from timestamps when available; skipped when not
   - `key_attendees_present`: flag if decision-makers were absent

9. **Surface next steps**
   When we reconvene, what needs to happen on the critical path before that.

10. **Render TEMPLATE.md and validate**
    - Every action's owner appears in `attendees` or is explicitly `unassigned`
    - `meeting_quality.outcomes_achieved` matches `N/M` pattern when populated
    - `agenda_reconciliation` fields present even when empty lists

## Project Memory Contract

Active only when `.claude/pm-skills.local.md` exists. With no file, ignore this section entirely
and behave exactly as described above.

- **Reads:** `active_initiative`, so decisions are recorded against the initiative they belong to.
- **Writes:** each decision actually reached in the meeting to the `## Decisions` section, dated and attributed, plus the recap itself as a `decision` artifact.
- **Posture:** propose the entry and wait for confirmation before writing, unless
  `memory_auto_append: true` is set, in which case append and echo what was written.
- **Write discipline:** re-read the file immediately before writing, never from the copy that
  produced the proposal. If it changed in between, merge your entry into the current state and
  re-propose rather than overwriting; add only your own entry and leave every other field and
  section byte-identical. Nothing enforces this at runtime and the file is gitignored, so a
  careless whole-file write loses another session's work with no way to recover it.

The fabrication prohibition above governs this write without exception: record only decisions that were stated. An unrecorded decision is recoverable; an invented one in durable memory is not. This complements the family's filename-based chaining rather than replacing it: filenames still locate the sibling artifacts of one meeting, while project memory carries the durable product context across meetings.
## Quality Checklist

- [ ] Input quality flagged honestly (high / medium / low)
- [ ] Agenda auto-discovery attempted; result noted in Generation context

From product-on-purpose/pm-skills, Apache-2.0.

More from product-on-purpose/pm-skills