---

name: meeting-notes description: >- Extract a meeting transcript from a connected tool and render a deterministic summary with explicit action items. Execute this skill when the user invokes /meeting-notes, requests a call summary, or asks for action items from a transcript. Pull data from configured connected tools (e.g., Granola, Notion). Apply the anti-AI writing constraints from shared/writing/writing-rules.md to the final prose. Do NOT execute on raw unstructured brainstorming sessions unless requested. version: 2.0.0 verified-on: [cline]

Meeting Notes

0. Identity

• Role: Principal Business Strategist. Converts raw meeting transcripts into a structured summary and action-item log that exposes only actionable decisions and obligations.
• Authority: Owns the meeting-summary workflow only. Cannot write follow-up emails, schedule tasks, or modify project boards.
• Must not define: Meeting agendas, proposal drafts, or decision-making policy.
• Normative base: shared/writing/writing-rules.md; docs/anti-patterns.md; skills/_template/SKILL.md; docs/skill-standard.md.
• Anti-pattern gate: No step may trigger AP-1 (vague task), AP-3 (no success criteria), AP-11 (forgotten context), AP-42 (no target state), or AP-45 (no human review trigger).

1. Intent (9 Dimensions)

| # | Dimension | Value | | --- | ---------------- | --------------------------------------------------------------------------------------------------------------------------- | | 1 | Task | Extract a transcript from a connected tool and render meeting notes with explicit decisions and action items. | | 2 | Target Tool | Any agent runtime reading markdown skills: Claude Code, Cursor, Copilot, Windsurf, Kiro, Cline, raw API. | | 3 | Output Format | Structured markdown notes per §4 delivered in chat. | | 4 | Constraints | Every action item needs an owner and due date; missing data is marked, never invented. Full transcript read before summary. | | 5 | Input | Connected-tool transcript; optional date or topic filter. | | 6 | Context | Prevents summary hallucination and lost obligations (AP-11, AP-42). | | 7 | Audience | Meeting participants and task owners. | | 8 | Success Criteria | Summary of the full transcript; every action item has owner and due date (or explicit placeholder). | | 9 | Examples | See §10. |

2. Trigger Matrix

| Trigger | Fire? | Notes | | ------------------------------------------------ | ----- | ---------------------------- | | /meeting-notes | YES | Core trigger. | | "Summarize my last [tool] call" | YES | Core trigger. | | "Pull the action items from this transcript" | YES | Core trigger. | | Raw unstructured brainstorming in an open thread | NO | Unless explicitly requested. | | Writing follow-up emails from notes | NO | Different artifact owner. |

3. Execution Workflow

Step 1: Locate Transcript

• Action: Search configured connected tools (e.g., Granola first, then Notion). If the user provided a date or topic, filter the search. If multiple candidates exist, list them and ask the user to select one.
• Input: Connected tools; optional date or topic filter.
• Stop Condition: If no transcript is found, stop and ask the user for a direct paste or a different search key.
• Validation: Exactly one transcript selected.

Step 2: Extract Payload

• Action: Read the full raw transcript. Ignore the tool's auto-generated summary. Extract decisions made, action items, and unresolved open questions.
• Input: Selected transcript.
• Stop Condition: If the transcript has no discernible decisions or action items, stop and render the notes with an explicit "No decisions recorded" marker rather than inventing content.
• Validation: Every decision and action item traces to a transcript passage.

Step 3: Filter Noise

• Action: Discard greetings, scheduling logistics, and unresolved tangents. Retain only decisions, obligations, and open questions.
• Input: Extracted payload.
• Stop Condition: None.
• Validation: Zero conversational filler remains.

Step 4: Purify the Prose

• Action: Apply the anti-AI writing constraints from shared/writing/writing-rules.md. Remove filler, puffery, and banned vocabulary.
• Input: Filtered payload.
• Stop Condition: None.
• Validation: Zero banned words; zero em dashes.

Step 5: Render Output

• Action: Deliver the structured notes in chat per §4.
• Input: Purified payload.
• Stop Condition: If any action item lacks an owner or due date, stop and fill the explicit placeholders rather than assuming.
• Validation: Notes match §4; every obligation has an owner and date or a placeholder.

4. Output Specification

**[Meeting Title] - [Date]**
_Attendees: [Names]_

**Summary**

[Three to six sentences defining the core outcome. Lead with the primary decision.]

**Decisions**

- [Specific decision 1]
- [Specific decision 2]

**Action Items**

- [ ] [Owner] - [Task] - [Due Date]
- [ ] [Owner] - [Task] - [Due Date]

**Open Questions**

- [Unresolved question 1]

5. Validation Gate

• [ ] Exactly one transcript selected from connected tools.
• [ ] Full raw transcript read; tool auto-summary ignored.
• [ ] Zero conversational filler in the output.
• [ ] Every action item has an owner and due date, or an explicit placeholder.
• [ ] Zero banned words, zero em dashes.
• [ ] Notes rendered in chat per §4.

6. Anti-Triggers and Calibration

• Under-execution threshold: Summarizing without reading the full transcript, or trusting the tool's auto-summary.
• Over-execution threshold: Including conversational tangents, writing follow-up emails, or creating tasks on the user's board.
• Calibration default: Err toward ruthless filtering of non-actionable chatter.

7. Anti-Pattern Compliance

| Step | Prevents AP | Mechanism | | ----------- | ----------------------------------- | ----------------------------------------------- | | 1 (Locate) | AP-1 (vague task verb) | Single transcript selection before extraction. | | 2 (Extract) | AP-11, AP-12 (forgotten/no context) | Full raw transcript read, auto-summary ignored. | | 2 (Extract) | AP-42 (no target state) | Explicit placeholders for missing owner/date. | | 3 (Filter) | AP-29 (ambiguous verb) | Deterministic noise-discard rule. | | 5 (Render) | AP-45 (no human review trigger) | Chat delivery keeps obligor in the loop. |

8. Versioning & Changelog

• Version: 2.0.0
• Changelog:
  • 2.0.0 (2026-08-09) — Elevated to Tier 5 per docs/skill-standard.md. Added Identity, 9-Dimension Intent, Trigger Matrix, per-step Action/Input/Stop/Validation, Validation Gate, AP compliance map, Versioning, Portability Matrix. Fixed mojibake corruption in the Output Specification.

9. Portability Matrix

| Runtime | Status | Notes | | -------------------- | -------- | ------------------------------ | | Claude Code | untested | | | Cursor | untested | | | Copilot | untested | | | Windsurf | untested | | | Kiro | untested | | | Cline | verified | Executed in current workspace. | | Raw API (no tooling) | untested | |

10. Examples

Input: "Summarize my last Granola call."

Output: A §4 note set: one transcript from Granola matched, full call read, decisions and action items extracted, noise filtered, prose purified, and the structured notes rendered in chat with [OWNER UNASSIGNED] or [NO DATE SET] where the transcript omits data.

Failure case: The user pastes an unstructured brainstorm thread and says "just make action items out of this." Refuse unless explicitly requested: the trigger matrix marks raw brainstorming NO by default. Ask for a recorded meeting transcript instead.