菜单
Automatic extraction of tasks, decisions, open questions, and blockers at the end of each session. Triggers on "TODO", "to do", or alongside a handover. Scans the entire conversation using a 6-category checklist to ensure nothing is missed. Outputs a human-readable summary + structured JSON.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Produces a structured 5-part handover document at the end of a complex session, ensuring continuity between AI instances. Triggers on "handover", "session end", "wrap up", or after 10+ substantive exchanges. Outputs an operational briefing the next AI instance can parse in 30 seconds — not a conversation summary.
AI Error Post-Mortem. Triggers when the user says "RETEX" or after the AI has failed 3+ times on the same problem. Captures WHY the AI failed and HOW to prevent recurrence. Produces a structured markdown file with 6 mandatory blocks including named anti-patterns and injectable "vaccine" rules. This is not a handover (project state) or an audit (quality check) — it is a diagnostic that produces immunity.
| name | todo-tracker |
| description | Automatic extraction of tasks, decisions, open questions, and blockers at the end of each session. Triggers on "TODO", "to do", or alongside a handover. Scans the entire conversation using a 6-category checklist to ensure nothing is missed. Outputs a human-readable summary + structured JSON. |
Version: 1.0
Part of: never-start-over
Companion skills: handover (project state), retex (error post-mortem)
When the user asks for a TODO:
The AI MUST review these 6 categories:
Note: Trigger phrases below are in English. Adapt to the conversation's language — the patterns matter, not the exact words.
Trigger phrases:
Trigger phrases:
→ Becomes a task IF it requires a follow-up action.
Trigger phrases:
Indicators:
Trigger phrases:
Trigger phrases:
| Tag | Meaning | Example |
|---|---|---|
owner: user | User handles it | "Test the script on mobile" |
owner: ai | AI handles next session | "Code the dashboard" |
owner: external | Depends on a third party | "Wait for IT response" |
owner: both | Collaboration | "Validate specs together" |
| Level | Label | Criteria |
|---|---|---|
| P0 | Critical | Blocking, urgent, deadline approaching |
| P1 | Important | Necessary but not urgent |
| P2 | Normal | Do when possible |
| P3 | Nice-to-have | Optional, improvement |
Priority signals:
{
"project": "PROJECT_NAME",
"generated_at": "2026-04-17T18:30:00",
"session_uri": "conversation-link-or-id",
"summary": "2-3 sentence session summary",
"tasks": [
{
"id": "PROJECT-001",
"text": "Clear task description",
"priority": "P0",
"status": "open",
"owner": "user",
"category": "action",
"tags": ["tag1", "tag2"],
"context": "Why this task exists (1 line)",
"deadline": null,
"created_at": "2026-04-17",
"done_at": null
}
],
"decisions": [
{
"text": "Going with local Python server",
"reason": "Enables dynamic folder scanning"
}
],
"open_questions": [
"How to handle merge conflicts?"
],
"attack_order": ["PROJECT-001", "PROJECT-003", "PROJECT-002"],
"next_session_focus": "Code the server + dashboard"
}
| Category | Description |
|---|---|
action | Concrete task to execute |
decision | Decision requiring follow-up |
question | Question to resolve |
blocker | Blocking problem |
insight | Discovery to retain |
idea | Idea to explore later |
TODO_{{PROJECT}}_{{YYYYMMDD}}_{{HHhMM}}.json
Save the JSON wherever your project stores session data. Suggested locations:
.session/ directory in project roottodos/ folder in your knowledge baseAt the start of a new session, if a previous TODO exists:
"status": "done", "done_at": "date")When the user triggers "TODO":
What was done, key decisions, next steps.
| # | Task | Priority | Owner |
|---|---|---|---|
| 1 | ... | P0 | user |
List if applicable.
Recommended sequence for tackling the tasks.
Code block with JSON ready to save.
| Pattern | Why |
|---|---|
| "We should have..." | Past, not actionable |
| "It would be nice to..." (vague) | Too fuzzy — ask for precision |
| "I did X" | Already done |
| Off-topic conversation | Not project-related |
| Jokes, digressions | Not relevant |
Before generating the JSON:
A TODO extraction is good when it passes these tests:
Completeness test: Read through the conversation one more time after generating the TODO. Did you miss any action item, question, or blocker? If you find even one, the extraction failed.
Actionability test: Could someone execute each task from its description alone? "Fix the bug" fails. "Fix the race condition in WebSocket reconnect logic (server.py line 47)" passes.
Ownership test: Every task has exactly one owner. "Both" is acceptable only for tasks requiring real-time collaboration. If unsure, assign to the user — they can reassign.
Priority test: Do the priorities reflect the conversation's actual urgency signals, not just the AI's guess? Re-read the user's exact words before assigning P0.
Attack order test: Does the sequence respect dependencies? A task that depends on another must come after it. Does it front-load blockers?
At the end of a productive AI session, there are always tasks scattered across the conversation — some explicit ("we need to test this"), some implicit (a decision that requires follow-up), some invisible (a question asked but never answered). Without systematic extraction, these tasks fall through the cracks.
The 6-category checklist ensures nothing is missed, and the attack order transforms a flat list into an actionable plan.