meeting-feedback

# Meeting Feedback ## Consequential advice Before recommending a behavior to practice, follow the `Advice gate` in `dissenter` when the choice is consequential for the user's next meeting. When the gate applies, first say that you are using `/dissenter` and why. Review a meeting from Granola and give a short, calm coaching note. Be specific and evidence-based, but keep the output easy to act on. ## Fetch Meeting Data Read and follow [transcript resolution](../transcript-resolution.md) with these options: - Permit a named Granola meeting, attendee, or topic, and the most recent Granola meeting when no argument is supplied. - Use **Notes allowed** mode. Use the notes or summary for context and decisions. - When using notes, say: "Transcript unavailable, so this is based on the meeting notes only." ## Review Focus Look for the few points that matter most. Do not run a full rubric. - Identify 1-2 communication strengths that affected the meeting positively. - Identify 1-2 communication issues that made the meeting less clear, less decisive, or harder to follow. - Choose exactly one behavior the user should practice in the next meeting. Use a direct, calm tone. Do not be harsh for effect, do not flatter, and do not use red/yellow/green scoring. ## Evidence Include 1-2 short transcript quotes total, only where they clarify the feedback. If using notes/summary only, do not invent quotes; refer to the relevant note or summary point instead. ## Output Format Return only this structure: ```markdown Meeting: [title] What worked - [specific strength, grounded in the meeting] - [optional second strength] What got in the way - [specific issue, grounded in the meeting] - [optional second issue] Evidence - "[short quote]" [timestamp if available] - "[optional second short quote]" [timestamp if available] Try next time - [exactly one concrete behavior to practice in the next meeting] ``` If transcript quotes are unavailable, replace the `Evidence` bullets with 1-2 concise references to the meeting notes. Do not save files by default. If the user asks to save the feedback, save the exact note you produced. ## Automatic Jev check When the ordinary review identifies a candidate communication strength or issue, send only that candidate, its one or two supporting transcript passages, and stable IDs after reading the existing `typesafe-ai` skill and its current API documentation. Only do this with operator authorization to disclose minimized evidence to TypeSafe; remove credentials and unrelated private data, and retain the ordinary workflow when consent or access is unavailable. Ask a Noul: `the supplied passages support this specific coaching finding`; high means the stated finding follows without adding unstated context. Use it internally to recheck evidence selection; the agent writes the coaching note. On ambiguity, unavailability, or disagreement, use the original focused review. Preserve the exact visible output structure and do not add scores.