agentleFS
Ethan9123Ethan9123/wx-life-analysis/.claude/skills/task-extract/SKILL.md

task-extract

Skill0 starsChanged 6 months agoApache-2.0View on GitHub

Extract a structured TODO list from a boss/client/family-stakeholder's chat dump + attached files (PDFs, docs, links). Activate when the user says "Jennifer / Alice / 老板 dumped a bunch of stuff on me, what am I supposed to do?", "list the actual tasks from these messages", "what's blocking what", "what did the boss decide vs leave open". Reads people/<task-giver-slug>/chat.md + projects/<project>/*.txt (PDF extracts) + optional topics/<group-slug>/chat.json (group chats), produces a structured task list classified into 6 buckets (Direct work / Decision needed / Data needed / Already-promised / Implicit / Closed) with 3-axis priority scores (stakeholder × urgency × reversibility). Strict quote-don't-paraphrase rule because paraphrasing drift causes wrong-direction work.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 1 is distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When this skill is loaded, you are converting work-chat streams + attached files into a typed TODO list with priorities, dependencies, and …

    line 8
  2. Very specific:

    Full methodology: read docs/task-extract.md.

    line 10
  3. Very specific:

    Run tools/task-extract.ps1 -Person <task-giver-slug> -Project <slug> to get a pre-filtered candidates file.

    line 23
  4. Very specific:

    Don't invent tasks: If a message is ambiguous, list under ## Needs clarification for a follow-up ping, not as a concrete TODO.

    line 70

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Task Extraction Skill

When this skill is loaded, you are converting work-chat streams + attached files into a typed TODO list with priorities, dependencies, and decision points.

Full methodology: read docs/task-extract.md. It defines the 7 speech-act categories, the 6-bucket classification, the 3-axis priority scoring, and the hard rules.

Quick reference

Inputs (in order)

  1. people/<task-giver-slug>/chat.md — last 100-200 messages or since last project milestone
  2. Group chat (optional): topics/<slug>/chat.json — pull with tools/refresh-group.ps1 if not cached
  3. Source materials: projects/<name>/*.pdf → extract with tools/extract-pdf.js first
  4. Prior state: projects/<name>/notes.md — avoid re-listing already-captured tasks

First-pass filter

Run tools/task-extract.ps1 -Person <task-giver-slug> -Project <slug> to get a pre-filtered candidates file. You (the agent) then classify + prioritize. The tool does regex filtering only, not classification.

Speech-act categories that count as tasks

  • Direct ask: "你 [verb]…", "麻烦…", "请…", "Can you…"
  • Implicit ask: "你看一下", "这个怎么处理", "Take a look"
  • Deferred decision: "你来定", "Your call"
  • Information request: "什么时候能…", "Status?"
  • Resource send (file with no comment = implicit "read this")
  • Deadline statement: "X 之前", "by Friday"
  • Escalation: "尽快", "ASAP"

A bare "知道了" / "好" / 👍 closes a thread, not opens one.

6 buckets

Direct work · Decision needed (from them) · Data needed (from third party) · Already-promised (by you) · Implicit / inferred · Closed

3-axis priority (each 1-3, sum 3-9)

Stakeholder weight (3 = boss/family/paying client) × Urgency (3 = named deadline within 7 days) × Reversibility (3 = recoverable, 1 = once shipped you can't take back).

  • P0 = 7-9 (this week)
  • P1 = 5-6 (this month)
  • P2 = 3-4 (whenever)

For INTJ-style bosses (e.g. user's typical setup), calibration note: under-react is the bigger risk than over-react. They don't repeat themselves.

Output format

Produces a diff proposal for projects/<name>/notes.md, not a wholesale rewrite. Each new task includes:

- [ ] **<short imperative>** — bucket: <Direct work / Decision needed / ...>
  - **From**: <stakeholder>, <date> <message-or-file-reference>
  - **Original**: <verbatim or near-verbatim quote, ≤30 chars, in quotes>
  - **Priority**: P<0/1/2> (stakeholder=<n> urgency=<n> reversibility=<n>)
  - **Blocks**: <if blocked, name the dependency>
  - **Status**: pending

Ambiguous messages go in ## Needs clarification, not invented as TODOs.

Hard rules

  1. Quote, don't paraphrase. ≤30-char verbatim quotes. Paraphrasing introduces drift.
  2. Surface your own promises. Outgoing messages where you committed to action ("我下周给你") become Already-promised — easy to forget.
  3. Respect 知道了-closes-thread. Standalone "knows" / "ok" / "👍" from the task-giver CLOSES the prior thread.
  4. Don't invent tasks. If a message is ambiguous, list under ## Needs clarification for a follow-up ping, not as a concrete TODO.
  5. Don't drop dependencies. "Need data from Sophie" must be a tracked task even if Sophie isn't in people/ yet.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.