meeting-transcript-parse

Installable Ships in meeting-to-action — or copy the raw skill below.

Paste into Claude Code, Claude.ai, the desktop app, or the browser extension — ⌘/Ctrl+V.

ExperimentalNo security review

Meeting Transcript Parse

Transform raw meeting transcripts and rough notes into a clean, normalized meeting record. The output is a structured markdown document that serves as the source of truth for every downstream artifact: decision log, action register, follow-up email, and Jira push.

When this skill applies

User pastes a Teams/Zoom auto-caption export, a recording transcript, or a rough notes document
User provides a file path to a .txt, .docx, .md, or .vtt transcript
User asks to "pull actions," "clean up the meeting notes," "write a summary," or "who owned what"
After any client status call, sprint review, steering committee, or internal planning session

Skip when the user already has a clean normalized record — go directly to decision-log-builder, action-item-tracker, or jira-confluence-push.

Workflow

Step 1 — Load and orient

Read all provided files or pasted content. Identify:

Meeting type: Status call / steering committee / sprint review / discovery workshop / internal / other
Date and time (extract from transcript metadata or participant timestamps; ask if absent)
Duration (compute from first/last timestamp if available)
Participants: extract all named speakers; deduplicate by person, not by message
Platform context: detect references to Sitecore, Optimizely, Umbraco, Azure DevOps, Jira, Confluence, or specific project names — flag these for downstream use

Confirm participant list and meeting type with the user before proceeding if either is ambiguous.

Step 2 — Segment the transcript

Break the raw content into logical conversation segments. Each segment corresponds to one agenda item, topic shift, or distinct discussion thread. Label each:

[SEGMENT: <topic label>] [<start timestamp or position>]

If no timestamps exist, infer topic boundaries from conversational signals ("moving on," "next item," "let's talk about," "before we close").

Step 3 — Extract and tag content

Within each segment, tag every substantive contribution:

Tag	Use when
`[DECISION]`	A conclusion was reached and stated (even informally — "ok, we'll go with option B")
`[ACTION]`	A task was assigned or self-assigned with at least an implied owner
`[QUESTION]`	A question was raised that was not answered in this meeting
`[RISK]`	A concern, blocker, or dependency was flagged
`[PARKING LOT]`	Topic was deferred explicitly or implicitly to a future meeting
`[INFO]`	Context, background, or update shared — no action or decision required

Every [DECISION] must capture: what was decided, who decided it (role, not personal name for client notes), and any conditions or caveats.

Every [ACTION] must capture: what task, who owns it (name + role), and target date (explicit or inferred — flag as [DATE TBD] if absent).

Step 4 — Build the normalized record

Produce the structured meeting record using the template in templates.md. Sections in order:

Meeting header — type, date, duration, attendees (name + role + org)
Agenda items covered — bulleted list; mark items not reached as [NOT REACHED]
Discussion summary — one paragraph per segment; factual, past-tense, no editorializing
Decisions — numbered list; each entry: decision text, decision-maker role, conditions
Action items — table: # | Action | Owner | Due Date | Status (Open)
Unresolved questions — numbered list; each entry: question text, who raised it, what's needed to resolve
Parking lot — bulleted list; each entry: topic, why deferred, suggested next meeting slot
Risks and blockers flagged — numbered list; each entry: risk description, raised by role, current status

Step 5 — Self-check

Before returning:

Every [DECISION] has a decision-maker role (not just "the group")
Every [ACTION] has an owner name and a due date (or [DATE TBD])
No participant is identified by personal name in sections that will be shared with clients; roles are used instead (or first name only for internal-only outputs)
Parking-lot items are not silently dropped — they appear explicitly
The action table is sortable by owner (alphabetical by last name or role)

Return the normalized record as a markdown block. Remind the user which downstream skills can consume it: decision-log-builder, action-item-tracker, meeting-followup-draft, jira-confluence-push.

Meeting type adapter notes

Meeting type	Special handling
Status call	Flag any scope change signals — even casual ("we might add...") — as `[RISK]`
Steering committee	Elevate `[DECISION]` entries — these are formal governance decisions; include quorum note if detectable
Sprint review	Link accepted/rejected items to sprint goal if mentioned; flag velocity signals
Discovery workshop	See `discovery-synthesizer` plugin for deeper synthesis; this skill produces the raw record only
Internal	Relax client-anonymization rules; names can appear in action table

Reference files

templates.md — full section templates with placeholder guidance and formatting examples

Technical details

Plugin

meeting-to-action

Author

Dakota Smith

Skill name

meeting-transcript-parse

Maturity

experimental

Security review

No security review

Owners

delivery

Argument hint

[paste transcript or @file-path] [--type status-call|steering-committee|sprint-review|discovery|internal] [--platform sitecore|optimizely|umbraco|azure|agnostic] [--output markdown|json]

Allowed tools

Read
Write
Glob

Source

plugins/meeting-to-action/skills/meeting-transcript-parse/SKILL.md

Edit this skill on GitHub — opens GitHub's web editor. No git knowledge needed.