/install ai-transcript-action-extractor
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.
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install ai-transcript-action-extractor - 安装完成后,直接呼叫该 Skill 的名称或使用
/ai-transcript-action-extractor触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
Ai Transcript Action Extractor 是什么?
Convert pasted, user-provided transcripts or chat exports into a structured action register with tasks, owners, dates, decisions, risks, open questions, and... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 104 次。
如何安装 Ai Transcript Action Extractor?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install ai-transcript-action-extractor」即可一键安装,无需额外配置。
Ai Transcript Action Extractor 是免费的吗?
是的,Ai Transcript Action Extractor 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Ai Transcript Action Extractor 支持哪些平台?
Ai Transcript Action Extractor 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Ai Transcript Action Extractor?
由 haidong(@harrylabsj)开发并维护,当前版本 v1.0.1。