Meeting Minutes

Convert a meeting transcript (or any pre-transcribed text) plus a Markdown minute template into a clean, structured meeting minute. No audio processing is involved — the user supplies text.

When to use

• The user pastes a transcript, or points to a .md / .txt transcript file, and wants a structured meeting minute.
• The user has a minutes template (Markdown) they want filled, or wants the skill's default template.
• The user says "整理会议纪要", "summarize the meeting", "fill the minutes template", etc.

Do not use for: transcribing raw audio/video (out of scope — ask the user to provide a transcript first).

Inputs / outputs

The skill ships references/minutes_template.md as the default template.

Workflow

Step 1: Gather inputs

1. Get the transcript — a file path, or pasted text.
2. Get the template — a user template, or use the default: references/minutes_template.md.
3. Confirm the desired output path (default: meeting-minutes.md in CWD).

Step 2: Extract structured data from the transcript

Read the transcript and produce a JSON data object with these keys. Use null or [] when a section is absent — do not invent content.

Extraction rules (opinionated):

• Map Speaker: labels in the transcript → attendees. Unlabeled or "not present" mentions — note under Open Issues or the relevant discussion point; no separate absent field is exposed.
• Pull explicit decisions, approvals, and "we will / let's / agreed to" phrasing → action_items with owner + due when stated, else TBD.
• Summarize each agenda topic into 2–4 bullets; preserve numbers, dates, names, and technical terms verbatim — do not paraphrase metrics.
• Language: Detect the transcript's language and set language to match (e.g. zh for a Chinese transcript). Write all extracted content in that language. For Path A also localize the template's section headers/labels; for Path B pass a localized template — the script localizes its own generated action-items table headers and the (None) / (TBD) labels via language. Empty sections ({{next_meeting}} with no data) are automatically removed from the rendered output.
• If the transcript is partial, mark uncertain fields TBD and note it in the minute (e.g. under Open Issues).

Step 3: Fill the template

Two equivalent paths — pick whichever fits:

A. Direct (recommended): Fill the template in place from the JSON you just built. Replace each {{placeholder}} with the rendered value, and remove any leftover {{...}} tokens. Section headers in the template stay as-is.

B. Scripted (deterministic): Save the JSON to a file and render with the bundled script. This guarantees consistent formatting/tables:

python "<skill_dir>/scripts/fill_minutes.py" <data.json> <template.md> <output.md>

• <template.md> and <output.md> are optional; they default to the bundled template and YYYYMMDD_<MeetingPurpose>_Minutes.md in CWD (named from the date + meeting_purpose/meeting_title field; <MeetingPurpose>_Minutes.md if no date).
• To scaffold a sample JSON, run:

  python "<skill_dir>/scripts/fill_minutes.py" --init-skeleton skeleton.json

The script renders {{attendees}}/{{decisions}}/{{open_issues}} as bullet lists, {{agenda}} as a numbered list, {{discussion}} as ### topic sub-sections, {{action_items}} as a Markdown table, and {{next_meeting}} as a key/value block. When {{next_meeting}} data is absent, the entire ## Next Meeting heading and its content are removed from the output — no placeholder or fallback remains.

Step 4: Verify & present

1. Confirm no {{placeholder}} tokens remain in the .md output.
2. Confirm the .md file was written.
3. Confirm the source transcript and template are untouched.
4. Present the output path and a short summary: # attendees, # decisions, # action items, and any TBD/open items needing follow-up.

If the user also needs a Word (.docx) copy, use the meeting-minutes-export orchestrator skill.

For the full end-to-end workflow (.md → .docx), see meeting-minutes-export/SKILL.md.

File organization rules

• Input transcript / template: Read-only. Never modified.
• Output minute: Saved to CWD. Never overwrites the template.

Bundled resources

• references/minutes_template.md — default Markdown template with {{placeholder}} tokens.
• scripts/fill_minutes.py — deterministic renderer (JSON → templated .md).

Typical user prompts