meeting-transcript-to-summary

Use when (1) user pastes meeting transcript and needs structured summary with action items and decision owners.

Use when (1) user pastes meeting transcript and needs structured summary with action items and decision owners.

Core Position

This skill solves the specific engineering problem of: user pastes meeting transcript and needs structured summary with action items and decision owners

This skill is NOT:

• A general-purpose capability that activates on anything
• A replacement for manual human judgment
• A tool that stores state or remembers across sessions

This skill IS activated ONLY when the trigger conditions are explicitly met.

Modes

/meeting-transcript-to-summary

Default mode. Performs the core task end-to-end.

When to use: User provides input matching the trigger conditions above.

Execution Steps

1. Receive transcript — User pastes meeting transcript text (from Whisper, Otter, Fireflies, or manual notes)

   • If the input looks like a different format (code, logs, article), state: "This skill extracts structured summaries from meeting transcripts. Please provide a conversation transcript."

2. Identify speakers and segments — Parse the transcript structure:

   • Detect speaker labels or turn-taking patterns
   • Identify distinct topics or discussion sections
   • Note timestamps if present

3. Extract key content — Identify the substantive parts:

   • Key decisions made (explicit statements of what was decided)
   • Action items with assignees (who will do what by when)
   • Open questions or unresolved issues
   • Important context or constraints referenced

4. Format structured summary — Organize into the standard format:

   • Summary: 2-3 sentence overview of the meeting topic and outcome
   • Decisions: numbered list of decisions made with rationale
   • Action Items: who → does what → by when (if stated)
   • Open Items: questions left unresolved or follow-ups needed

5. Deliver and validate — Return the formatted summary:

   • If any section is empty, note "No [decisions/action items] identified"
   • If speakers aren't labeled, note "Speaker names not detected; labeled as Speaker 1, 2..."
   • Ask if the user wants to adjust granularity or add missing items

Mandatory Rules

Do not

• Do not make up facts or claim actions were taken that were not
• Do not hardcode API keys — use os.getenv("API_KEY") instead
• Do not store sensitive user data beyond the current session
• Do not exceed token budget without warning the user first
• Do not activate for off-topic requests — return a brief decline message

Do

• Validate all inputs before acting
• Handle errors gracefully with actionable error messages
• Log actions taken for auditability
• State explicitly when you are uncertain or data is insufficient

Quality Bar

A good output:

• Solves exactly the problem described in the trigger conditions
• Provides actionable result in the expected format within 3 turns
• Handles error cases with specific guidance, not generic "try again"
• States assumptions explicitly when input is ambiguous

A bad output:

• Solves a different problem than the one triggered
• Provides a generic "I can't help with that" without explaining why
• Crashes, hangs, or returns malformed output on valid input
• Activates for off-topic requests (false positive)

Good vs. Bad Examples

Scenario	Bad Output	Good Output
Trigger matched	"I can help with that." + no action	Correct transformation delivered in structured format
Invalid input	Crash or wrong result	"Missing required field: [X]. Please provide [Y]."
Ambiguous input	Guesses and might be wrong	States assumption and asks for confirmation
Off-topic request	Attempts to help anyway	"This skill activates when [trigger]. Please restate your request."

References

• references/ — Detailed templates, schemas, and edge-case rules for this skill