processing-workshop-transcripts

# Processing Workshop Transcripts Turns noisy discussion material into a document that later phases can rely on. It removes greetings, filler and tangents while preserving every actionable and business-critical point. ## Applicability and Precedence If the project already has a transcript or meeting-notes convention, follow it and apply this skill only where that convention is silent. Content preservation always outranks tidiness: when in doubt about whether a passage is business-relevant, **keep it**. ## Explicit Exclusions This skill does not extract epics or stories, judge scope, or resolve contradictions between sources. It produces one cleaned transcript and stops. ## Non-negotiable Rules | Severity | Rule | | --- | --- | | MUST | Preserve every decision, commitment, constraint and open question, including ones stated in passing. | | MUST | Keep the exact wording of requirements, constraints and conflicting viewpoints in a Preserved Context section, with attribution. | | NEVER | Invent metadata, roles or decisions the source does not state. Record "not stated in source" instead. | | NEVER | Infer the contents of a scanned PDF that returns no text. Report it and ask for a text-based version. | | NEVER | Normalise or paraphrase the client's own terminology while cleaning — preserve the term as said, including inflected surface forms, in the transcript's Terminology section. | ## Reference Loading | Reference | Load when | Covers | | --- | --- | --- | | [Cleaned transcript example](./references/cleaned-transcript-example.md) | Before writing the output file — always | The exact section order and table shapes of the deliverable | ## Execution Context Running in the main conversation: ask the user for missing metadata and write the file yourself. Running as the `transcript-cleaner` subagent: you have no interactive and no write tool — return the structured transcript as your final message and collect anything you would have asked under `## Open Questions`. ## Procedure 1. **Identify the format and metadata.** Speaker-labelled, plain notes, timestamped, mixed, or PDF. Read PDFs with the `Read` tool — no separate PDF server; pass an explicit `pages` range beyond 10 pages and work in chunks of ≤20. Capture date, duration (compute it from first and last timestamp), topic and context. 2. **Tag participants.** Name and role where stated or unambiguous from context. List a participant without a role rather than guessing one. 3. **Remove non-business content**: greetings and sign-offs, small talk, filler words and verbal tics, technical difficulties ("you're on mute"), off-topic tangents, and repeated restatements — keeping the clearest version of each point. 4. **Group by discussion topic.** Find the natural topic boundaries, give each a descriptive heading, list key points as bullets with speaker attribution where available, keep chronological order within a topic, and consolidate a topic that resurfaces later in the transcript under one heading. 5. **Extract key decisions** — explicit ("we agreed to…") and implicit (discussion converging without a formal declaration). Record what was decided, who made or endorsed it, and any conditions attached. 6. **Extract action items and open questions.** Action items: what, who, deadline. Open questions: anything raised and unanswered, deferred, or acknowledged as ambiguous. 7. **Preserve critical raw context.** Exact quotes where the original wording matters — requirements in specific business language, stakeholder constraints, conflicting viewpoints, domain terminology defined during the session. 8. **Write the file.** Follow [`cleaned-transcript-example.md`](./references/cleaned-transcript-example.md) and save to `specifications/<workshop-name>/cleaned-transcript.md`. Check before finishing: nothing business-relevant was dropped, topics are logically grouped, decisions, action items and open questions are complete, and the Terminology section captures every client term encountered, in the client's own words. ## Related Skills - [`extracting-epics-and-stories`](../extracting-epics-and-stories/SKILL.md) — consumes the cleaned transcript as its primary input. - [`analyzing-discovery-context`](../analyzing-discovery-context/SKILL.md) — synthesizes it with the other workshop materials.