Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Transforms raw meeting notes, transcripts, or recordings text into a concise, scannable summary with clear sections for decisions made and owned action items (owner + due date). Use this skill when the user asks to "summarize this meeting", "turn these notes into a summary", "extract action items", "who agreed to what", "write meeting minutes", "TLDR this transcript", "pull decisions and next steps", or pastes a Zoom/Teams/Google Meet transcript and wants it condensed. Also applies to standups,
.claude/skills/jayrha-meeting-summarizer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 198% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 289% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 373% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 361% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 439% | 0% |
This skill converts messy meeting input (chat transcripts, dictated notes, bullet dumps, recording transcripts) into a structured, decision-oriented summary that a busy reader can absorb in under a minute. The output prioritizes the two things people actually need after a meeting: what was decided and who owes what by when.
Keywords: meeting minutes, summary, recap, action items, decisions, owners, due dates, TLDR, standup, retro, 1:1, transcript, notes.
The non-negotiable rule: every action item must have an owner. If the source does not name one, the action item is flagged as OWNER: UNASSIGNED rather than silently dropped or guessed.
Use this skill whenever the user wants raw meeting content distilled. Triggers include pasting a transcript, asking for "minutes", "a recap", "key takeaways", "next steps", or "the TLDR". Do not use it for live note-taking guidance or scheduling — only for processing content that already exists.
Follow these steps in order.
Date: not stated.UNASSIGNED.templates/summary.md. Lead with a 1–3 sentence TLDR, then Decisions, then Action Items (as a table), then Discussion, then Open Questions. Omit any empty section except Action Items (which always renders, even if to state "None captured").references/checklist.md before delivering. Most importantly verify the no-orphan-actions rule and that no decision was downgraded to a discussion bullet.# <Meeting Topic> — Summary
Date: ... | Attendees: ...
## TLDR
1–3 sentences.
## Decisions
- ...
## Action Items
| # | Action | Owner | Due |
## Discussion Highlights
- ...
## Open Questions / Risks
- ...See templates/summary.md for the full fill-in template and examples/standup-recap.md for a complete worked example.
Use this quick test on each notable line:
When something is both a decision and spawns work (e.g., "We'll launch Tuesday — Sam, prep the release notes"), record the decision in Decisions AND the task in Action Items.
For tone, owner-resolution edge cases, and a helper script, see the references below.
scripts/extract_actions.py is a zero-dependency Python tool that scans a notes/transcript file for likely action-item and decision lines using cue-word heuristics, and prints them grouped — a fast first pass before you write the polished summary. Run:
python3 scripts/extract_actions.py meeting.txt
python3 scripts/extract_actions.py meeting.txt --jsonIt is a draft aid, not a replacement for reading the full input.
UNASSIGNED instead.See references/checklist.md for the full pre-delivery QA list and references/owner-resolution.md for handling tricky attribution cases.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-12 | pass→pass | 3,139 | 5,333 | +70% | 1 | 1 | 0% | 599 | 2,466 | +312% | 0 | 0 | — |
case-07 | pass→pass | 1,959 | 2,407 | +23% | 1 | 1 | 0% | 343 | 1,914 | +458% | 0 | 0 | — |
case-01 | fail→pass | 4,543 | 6,243 | +37% | 1 | 1 | 0% | 899 | 2,678 | +198% | 0 | 0 | — |
case-02 | fail→fail | 5,130 | 4,471 | -13% | 1 | 1 | 0% | 920 | 2,407 | +162% | 0 | 0 | — |
case-03 | fail→fail | 6,444 | 9,435 | +46% | 1 | 1 | 0% | 1,097 | 2,047 | +87% | 0 | 0 | — |
case-04 | fail→pass | 3,020 | 3,566 | +18% | 1 | 1 | 0% | 550 | 2,138 | +289% | 0 | 0 | — |
case-05 | fail→pass | 2,584 | 3,870 | +50% | 1 | 1 | 0% | 477 | 2,254 | +373% | 0 | 0 | — |
case-06 | pass→pass | 2,865 | 5,633 | +97% | 1 | 1 | 0% | 589 | 2,682 | +355% | 0 | 0 | — |
case-08 | fail→pass | 2,571 | 3,089 | +20% | 1 | 1 | 0% | 449 | 2,068 | +361% | 0 | 0 | — |
case-09 | pass→pass | 2,543 | 3,515 | +38% | 1 | 1 | 0% | 435 | 2,180 | +401% | 0 | 0 | — |
case-10 | pass→pass | 3,106 | 6,877 | +121% | 1 | 1 | 0% | 585 | 2,252 | +285% | 0 | 0 | — |
case-11 | fail→fail | 1,449 | 4,031 | +178% | 1 | 1 | 0% | 231 | 2,156 | +833% | 0 | 0 | — |
case-13 | fail→fail | 7,377 | 5,447 | -26% | 1 | 1 | 0% | 1,249 | 2,411 | +93% | 0 | 0 | — |
case-14 | pass→pass | 2,934 | 5,487 | +87% | 1 | 1 | 0% | 632 | 2,559 | +305% | 0 | 0 | — |
case-15 | fail→pass | 2,388 | 4,154 | +74% | 1 | 1 | 0% | 402 | 2,168 | +439% | 0 | 0 | — |
case-16 | fail→pass | 2,252 | 3,669 | +63% | 1 | 1 | 0% | 349 | 2,080 | +496% | 0 | 0 | — |
case-17 | pass→pass | 2,561 | 4,078 | +59% | 1 | 1 | 0% | 478 | 2,279 | +377% | 0 | 0 | — |
case-18 | pass→pass | 5,027 | 4,848 | -4% | 1 | 1 | 0% | 844 | 2,425 | +187% | 0 | 0 | — |
case-19 | pass→pass | 2,679 | 2,990 | +12% | 1 | 1 | 0% | 469 | 2,075 | +342% | 0 | 0 | — |
case-20 | pass→pass | 2,067 | 4,364 | +111% | 1 | 1 | 0% | 288 | 2,289 | +695% | 0 | 0 | — |
case-21 | pass→pass | 4,942 | 4,213 | -15% | 1 | 1 | 0% | 690 | 2,147 | +211% | 0 | 0 | — |
case-22 | fail→pass | 3,547 | 3,380 | -5% | 1 | 1 | 0% | 559 | 2,040 | +265% | 0 | 0 | — |
case-23 | pass→pass | 9,954 | 7,735 | -22% | 1 | 1 | 0% | 1,884 | 2,791 | +48% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +30 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | +27% |
Other measured skills in the registry, with their headline benchmark lift.