agentleFS
Wang-CankunWang-Cankun/cankun-skills/skills/meeting-audio-report/SKILL.md

meeting-audio-report

Skill2 starsChanged 32 days agoMITView on GitHub

Transcribe a meeting recording and write a report from it. Use when the user has an audio or video file of a meeting, webinar, seminar, or Q&A session — .m4a/.mp3/.wav/.mp4/.mov — and wants a transcript, meeting notes, minutes, or a written report; also on 转录, 会议纪要, 会议记录, 会议报告, transcribe this, write up this meeting.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read it end to end before writing anything.

    line 51

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Meeting audio report

A recording becomes two things: a verbatim transcript, and a report someone who missed the meeting can act on.

Two words govern the run.

Coverage — did every second of audio reach the transcript? Transcription drops blocks on transport errors, and the loss is silent. The one block lost is as likely as any other to hold the decisive sentence. Coverage is a gate, not a statistic.

Fidelity — does every name, number, and quote in the report trace to the audio? A transcription model guesses at proper nouns; a summarising model guesses at who said what. Both guesses arrive fluent and confident. The report carries what the recording supports and flags the rest.

1. Transcribe

scripts/transcribe.py <audio-or-video> [--out-dir DIR]

Writes <stem>-transcript.md, <stem>-coverage.json, and <stem>-blocks.json beside the input, or into --out-dir.

Done when the script has exited and the manifest exists.

2. Pass the coverage gate

The script exits non-zero when any block failed, and names each gap with its time range.

scripts/transcribe.py <audio> --retry-missing [--out-dir DIR]

Repeat until it exits zero. Failures are transport, not audio, so a second attempt usually lands.

Done when blocks_ok equals blocks_total in the manifest. Carry which blocks were recovered into step 5 — a recovered block that turned out to hold something material is itself a finding.

A gap that survives repeated recovery goes into the report as a stated gap, with its time range and what was being discussed either side of it. Never write around it silently.

references/transcription.md holds the manifest format, tuning flags, cost, and the speaker-label limits.

3. Read the whole transcript

Read it end to end before writing anything. An hour of meeting is 40 to 60 KB — read it in full, not in excerpts.

Done when you can state, without looking back: the subject, every number that was quoted, what was decided, what was deferred, and what a speaker said that contradicts a document.

4. Write the report

Author it as Markdown against references/report-structure.md, which carries the section skeleton and the fidelity rules.

Two things drive the report's value, and both live in the transcript rather than in a template:

  • The Assessment section states asymmetries — what is free and commonly skipped, what is stated once in passing and silently disqualifying, what looks like latitude and is risk transfer.
  • Sourcing and Caveats is mandatory. It carries coverage, speaker attribution limits, transcription distortion of acronyms and names, provisional answers, and internal inconsistencies.

Attribute by role. A name enters the report only when a speaker states their own on the recording, quoted as heard and flagged as unverified.

Done when every section of the skeleton is present and every number carries its unit and basis.

5. Render

scripts/render.py <report>.md [--out <stem>]

Markdown → pandoc → DOCX → LibreOffice → PDF, styled to match.

Done when the PDF exists and pdfinfo reports a non-zero page count.

6. Hand over

State: where the files are, the coverage result including any recovered or unrecoverable blocks, and the fidelity caveats a reader would otherwise take on trust.

Deliver the DOCX and PDF. The transcript stays alongside as the evidence the report was built from.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.