agentleFS
ofcRSofcRS/kleoth/.claude/skills/summarize-meeting/SKILL.md

summarize-meeting

Skill8 starsChanged 51 days agoApache-2.0View on GitHub

Summarize a Kleoth meeting in this Claude Code session — you read the transcript and write summary.json (the MeetingSummary shape) and summary.md yourself, with no API key and no API cost. Use when the user wants a Kleoth meeting directory summarized here, by you, rather than through one of Kleoth's AI providers.

  • Reads credentials

9 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 9 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    If the meeting has no transcript yet (neither transcript.md nor transcript.json: the app shows it as *Untranscribed*.

    line 20
  2. Very specific:

    Read the transcript: Read <dir>/transcript.md (speaker-tagged Name: text lines).

    line 27
  3. Distinctive:

    Write every value (title, tldr, overview, each task, each highlight) in the same language as the transcript.

    line 33
  4. Distinctive:

    Keep speaker and owner names exactly as they appear in the transcript.

    line 35
  5. Very specific:

    If the binary isn't built yet, run swift build first.

    line 63
  6. Distinctive:

    No API key is needed; this uses your Claude Code session.

    line 70
  7. Very specific:

    Never read or echo .env or any API key.

    line 73

“Common” is measured: kept by at least 6 in 10 of the 9 edited versions of this file we could read. “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

Summarize a Kleoth meeting in this session

Kleoth normally summarizes through an AI provider: the app does it after each transcription, and the CLI does it with kleoth summarize <dir> --provider openrouter|local|claude-code|codex. This skill is the in-session alternative: you — Claude Code — read the transcript and write the summary directly. The output is interchangeable with the provider path: the same summary.json shape, and a summary.md rendered by the kleoth CLI.

Input

A meeting directory written by the Kleoth app or by kleoth transcribe (meeting-yyyy-MM-dd-HHmmss/, containing transcript.md and meta.json). The directory path is the argument to this skill. If none was provided, ask the user which meeting to summarize.

If the meeting has no transcript yet (neither transcript.md nor transcript.json: the app shows it as Untranscribed; it may already have a meta.json, which the app writes when the recording stops), stop and say so: transcribe it first, in the app or with the transcribe-meeting skill.

Steps

  1. Read the transcript. Read <dir>/transcript.md (speaker-tagged Name: text lines). If it is missing but <dir>/transcript.json exists, reconstruct the dialogue from its words[] / transcripts[] and speaker_id. Read <dir>/meta.json for the title, date and participants, and <dir>/speakers.json (if present) for the display names.
  2. Summarize, following the same rules as Kleoth's Summarizer:
    • Write every value (title, tldr, overview, each task, each highlight) in the same language as the transcript. Don't translate it into English or anything else.
    • Keep speaker and owner names exactly as they appear in the transcript.
    • Be precise and factual. Do not invent information. If something is ambiguous, say so.
    • Include an action item only if a concrete task was stated or clearly implied. If the owner is unstated use "unassigned"; if the due date is unstated use null.
  3. Write <dir>/summary.json with the Write tool: valid JSON, snake_case keys exactly as below (what MeetingStore and kleoth render read), every key present. No prose, no markdown fences:
    {
      "title": "a concise, specific 4-8 word meeting title",
      "tldr": "2-4 sentences capturing the essence of the meeting",
      "overview": "a detailed, faithful account of the whole meeting as flowing prose in several paragraphs separated by blank lines (no bullets, no headings): what was discussed and in what order, the context and reasoning, concrete specifics (names, numbers, dates, agreements and who made them), and how things were left",
      "action_items": [{ "owner": "name or \"unassigned\"", "task": "...", "due": "date or null" }],
      "per_speaker_highlights": [{ "speaker": "name", "highlights": ["their most important statements, positions and commitments"] }]
    }
    action_items may be an empty array; the other values may not be blank.
  4. Record who wrote it. In <dir>/meta.json, set "summary_provider": "claude-code" and remove "model", so the meeting doesn't credit an earlier model with this summary. Leave every other field as it is.
  5. Render summary.md with the CLI (no API call, so the formatting matches the provider path exactly):
    swift run kleoth render <dir>          # add --no-transcript to omit the transcript appendix
    If the binary isn't built yet, run swift build first.
  6. Report: tell the user the summary was written in this Claude Code session at no API cost, and print the path to summary.md.

Notes

  • No API key is needed; this uses your Claude Code session.
  • kleoth summarize <dir> --provider claude-code gets a summary from Claude Code without this session: it runs claude -p in isolation and writes the same files.
  • Never read or echo .env or any API key.

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.