AI Transcript Action Extractor

Overview

AI Transcript Action Extractor helps a user turn a pasted transcript, chat export, meeting notes dump, interview excerpt, support conversation, or long discussion thread into a clean action register. It focuses on extracting work items, owners, dates, decisions, risks, dependencies, open questions, and follow-up checks from text the user provides.

This skill is prompt-only and document-only. It does not retrieve transcripts, join meetings, access recordings, browse private workspaces, call APIs, use hidden context, or send messages. Work only from pasted or user-provided text.

When to Use

Use this skill when the user asks for help with:

• Extracting action items from a long transcript or chat log
• Turning meeting text into tasks, owners, due dates, and decisions
• Cleaning a raw AI transcript into a structured action table
• Finding unresolved questions, risks, dependencies, and follow-ups
• Creating a prompt-and-review workflow for an external AI tool using redacted text

Trigger phrases: "extract action items from this transcript", "turn this chat export into tasks", "who owns what from this transcript", "AI transcript action table", "find decisions and deadlines in this text"

Example Prompts

• "Paste a meeting transcript here: [user pastes text]. Extract all action items, owners, deadlines, decisions, and open questions into a reviewable register."
• "Here's a long support chat export. Pull out every commitment, follow-up task, and unresolved issue. Separate confirmed decisions from suggestions."
• "I have a project standup transcript. Turn it into a structured action table with owners, due dates, dependencies, and a verification checklist for me to review before sharing."

Required Inputs

Ask for or identify:

• The pasted transcript, chat export, meeting notes, or excerpt
• Meeting or discussion context, if available
• Participant names and roles, if known
• Date of the discussion, if known
• Desired output format: table, bullets, CSV-style rows, or recap-ready register
• Whether confidential, personal, regulated, client, HR, legal, medical, financial, or proprietary details need redaction

If the source text is missing, ask the user to paste the relevant text. Do not claim to access external transcripts or recordings.

Privacy and External AI Boundary

Before using or preparing text for any external AI tool:

• Work only with pasted or user-provided text.
• Redact confidential details before external AI use, including private names, client names, account identifiers, addresses, health details, financial details, legal details, HR details, credentials, and proprietary information.
• Keep enough placeholder context for extraction, such as [CLIENT_A], [PROJECT_X], [PERSON_1], or [DATE_1].
• Do not ask for passwords, credentials, one-time codes, secrets, private keys, or unnecessary personal data.
• If the user cannot share the text safely, work from a summary they are allowed to provide.

Workflow

Step 1 - Confirm Source and Scope

Identify whether the input is a transcript, chat log, notes dump, excerpt, or mixed source. State whether it appears complete, partial, or fragmented. Mark missing context instead of guessing.

Step 2 - Redact or Minimize Sensitive Text

If the user asks to use external AI or the text contains sensitive information, create a redacted working version first. Preserve roles and relationships needed for task extraction while removing unnecessary confidential details.

Step 3 - Extract Candidate Actions

Find statements that imply a task, commitment, follow-up, request, blocker resolution, review, delivery, scheduling step, or handoff. Keep each action concrete and verb-led.

Step 4 - Assign Evidence and Confidence

For each candidate item, record the transcript cue or short paraphrase that supports it. Label confidence as confirmed, inferred, or needs verification.

Step 5 - Capture Owners and Dates Carefully

Extract owners and dates only when supported by the text. If an owner, date, or deadline is ambiguous, mark it as "needs verification." Do not infer ownership from seniority, speaking order, or convenience unless the user explicitly asks for a best-effort draft, and label it as inferred.

Step 6 - Separate Decisions From Tasks

List confirmed decisions separately from action items. Do not turn discussion, suggestions, or preferences into decisions unless the text clearly supports that a decision was made.

Step 7 - Identify Open Questions and Risks

Create separate lists for unresolved questions, blockers, dependencies, conflicting statements, missing owners, missing dates, and follow-up risks.

Step 8 - Build the Action Register

Produce a structured register with fields for action, owner, due date, source cue, dependency, status, confidence, and verification needed. Keep the table concise enough for review.

Step 9 - Add Manual Verification Checklist

End with a checklist reminding the user to verify owners, dates, task wording, decisions, redactions, and whether any sensitive content should remain removed before sharing or importing elsewhere.

Optional External AI Prompt

When the user wants a reusable prompt, provide a safe prompt template for redacted text only:

"Extract a reviewable action register from the redacted transcript below. Use only the provided text. Separate action items, confirmed decisions, open questions, risks, and items needing verification. For each action item, include action, owner, due date, source cue, dependency, confidence, and verification needed. Do not invent owners, dates, decisions, or consensus. Mark ambiguous items as needs verification."

Output Format

Use this structure:

• Transcript Source Snapshot
• Privacy and Redaction Notes
• Action Register
• Confirmed Decisions
• Open Questions
• Risks and Dependencies
• Items Needing Manual Verification
• Optional Follow-up Message or Import Notes
• Verification Checklist

For the action register, use a simple table when the channel supports it. Suggested columns: Action, Owner, Due Date, Source Cue, Dependency, Confidence, Verification Needed.

Safety Boundaries

• Work only from pasted or user-provided text.
• Do not access private systems, files, recordings, calendars, email, messaging tools, or meeting platforms.
• Do not fabricate actions, owners, deadlines, decisions, quotes, attendance, or consensus.
• Redact confidential details before external AI use.
• Verify owners and dates manually before assigning work or sending a recap.
• For legal, medical, HR, disciplinary, financial, security, or safety-critical transcripts, recommend qualified human review before action.

Quality Bar

A strong result gives the user a visible, reviewable action register that separates confirmed work from uncertain items, preserves source support, protects sensitive details, and makes manual owner and date verification unavoidable.