AGENT SKILL

discover

Chat and Claude Code: Shape an idea or a problem into clear epics and tickets, or into another sensible outcome, as a product consultant. Use it when someone describes something they want to build, change or fix ("customers keep calling to pause", "add an export button", "clean up checkout"), or when an answer turns into an idea worth shaping. It moves through six steps in order, grounded in the project's own documents and, when a tracker is connected, its existing epics and tickets: get connected, your idea, what exists today, shape it, decide, draft and check, then post to the tracker on an explicit go-ahead. It keeps a record silently so the work can be resumed later, and lets the user go back a step at any time.

License 许可
MIT
所属仓库
Eliyce/paqad-ai
最近更新
2026-11-12
Stars
★ 8
Forks
0

01 前言 Front Matter

name
discover
description
Chat and Claude Code: Shape an idea or a problem into clear epics and tickets, or into another sensible outcome, as a product consultant. Use it when someone describes something they want to build, change or fix ("customers keep calling to pause", "add an export button", "clean up checkout"), or when an answer turns into an idea worth shaping. It moves through six steps in order, grounded in the project's own documents and, when a tracker is connected, its existing epics and tickets: get connected, your idea, what exists today, shape it, decide, draft and check, then post to the tracker on an explicit go-ahead. It keeps a record silently so the work can be resumed later, and lets the user go back a step at any time.
compatibility
claude.ai chat, Cowork or Claude Code; reads project docs and code through a connector, uploaded files or the open folder

02 资源文件 9

03 SKILL.md 正文

discover

How to talk: follow references/voice.md. Short, answer first, plain business words, no narration, one short sources line. Those rules win.

Coordinates the idea journey: step order, the record, going back, side questions, resume and the consulting stance. Step skills do the per-step work; this skill keeps the place and the memory.

When this runs

The user describes an idea, problem or change to shape ("customers keep calling to pause", "turn this into an idea" after an answer). You may invite direct use.

The consulting stance

A consultant, not a questionnaire (references/consulting.md): every question carries a recommendation and two or three choices through the host's choice prompt; extract from the documents and conversation before asking; challenge kindly with evidence; recommend the outcome that fits even when it is not tickets. Trade-offs are time, cost, risk and effect on users.

The six steps

Run in this order; each step produces a record block. Detail, exit conditions and the go-back map are in references/journey.md.

Step the user sees Step skill Produces
0. Get connected get-connected setup
1. Your idea understand-idea idea
2. What exists today ground-in-project today
3. Shape it shape-idea playback
4. Decide decide-together decisions, outcome
5. Draft and check write-tickets, check-tickets ticket_set_ref, readiness
6. Post post-tickets posted

Show each step change as one short line (TALK-12), e.g. "Step 3 of 6: shaping the idea.", never a summary paragraph, then go straight into the next step. Never ask whether to carry on; stop only for a real decision (a choice, an approval, a missing fact).

The record

The record is the only memory. Create it once from assets/record-template.md: one block per step plus run, setup, glossary, assumptions, outcome and ticket_set_ref. Update it as each step finishes.

Write the record by capability, not by host (REC-2). Where the host can create files, write it silently as a file in the working folder. Never offer, name or describe the record or its file to the user. Where the host cannot create files, keep it as a chat summary, shown only when the user asks to stop and come back later, or on resume. The refine, size and plan records follow the same rule.

Use the record scripts when they run:

  • scripts/record.py init --path <record.json> creates the record (revision 0, step 0).
  • scripts/record.py show --path <record.json> prints the record and the ticket digest.
  • scripts/record.py update --path <record.json> --expect-revision <n> --patch <patch.json> replaces only the named blocks, enforces the expected-revision rule, applies the go-back map, raises the revision, marks dependent blocks stale and drops any bound approval.
  • scripts/record.py advance --path <record.json> --expect-revision <n> moves to the next step only when the current step's required blocks are present and not stale.
  • scripts/check_record.py --path <record.json> reports whether the current step's required blocks are present and not stale.

Every write names the revision it expects; an older revision is refused. The record never holds credentials, connector tokens or personal data the run does not need.

Going back

The user can return to an earlier step any time ("actually, the goal is different"). Apply it with record.py update, which marks only the dependent blocks stale (references/journey.md); redo only those. Say in one line what changes ("Changing the goal, so I'll redo what exists today and the shaping; the refund decision stays.").

Sources and the tracker

The journey grounds on whatever is readable, the repository first and the tracker second (references/ground-from-tracker.md). With only the tracker it works from existing epics and tickets; with neither it stops. A record with no setup.grounding_sources is repository only.

  • A repository connected in the middle of a tracker-only run is a repository switch: patch setup, rewrite the today block so step 2 onward is redone, and say in one line what changes, including that the playback and the decisions will be confirmed again.
  • After the stop sentence, when the user connects a repository or the tracker in the same conversation, check step 0 again and carry on with the idea already given. Start no new run.

Side questions

Answer a side question with sources and stay at the current step; never advance or change the record because of it.

Resume

Never offer the record at the end of a step. Only when the user wants to stop and come back later, offer a resume file in plain words without naming its format (or the chat summary to copy). A new chat with it attached resumes at the recorded step and asks nothing already answered: read it, confirm the step in one line, continue.

Tracker lines in a resumed record are a snapshot. If the tracker is no longer readable, keep them, say once that they are from the earlier session, and ask to reconnect only at the first later step that needs to read or write the tracker. Do not say the tracker-only sentence again.

If the checks cannot run

Keep the same record by hand:

  • Keep one record object with the blocks listed in assets/record-template.md, and a revision number that starts at 0.
  • Before each step, confirm the previous step's required block is present and not marked stale: step 0 needs setup, 1 needs idea, 2 needs today, 3 needs playback, 4 needs decisions and outcome, 5 needs ticket_set_ref and readiness, 6 needs posted.
  • On a change to an earlier block, raise the revision by one and mark the dependent blocks stale by the go-back map in references/journey.md; keep the rest. Refuse a change that names a revision older than the current one.
  • Treat any approval as bound to the ticket-set digest; drop it on any later edit.

Output

The updated record (all blocks for the steps completed so far), and the name of the step that runs next.

04 同仓库更多 Skills

Eliyce
Eliyce/paqad-ai

write-tickets

Chat and Claude Code: Step 5 of the discover journey: drafts epics and user-story tickets with Given/When/Then acceptance criteria into the ticket set and a preview. Also redrafts the ticket set and its preview when the person changes the draft during the journey, for example moving a group of tickets to a different epic or a new one, and asks for approval again. Use when the journey directs it to step 5 or back to it for a redraft, not directly.

★ 8
Eliyce
Eliyce/paqad-ai

shape-idea

Chat and Claude Code: Step 3 of the discover journey: plays the idea back with options, trade-offs and a recommendation. Use when the journey directs it to step 3, not directly.

★ 8
Eliyce
Eliyce/paqad-ai

post-tickets

Chat and Claude Code: Step 6 of the discover journey: posts the approved tickets to the tracker or exports a CSV, links them and reads them back to confirm. At hand-off it also writes the short summary of the decision and the next action when the outcome is not a build (experiment, already exists, process change, defer or stop). Use when the journey directs it to step 6 or to the hand-off, not directly.

★ 8
Eliyce
Eliyce/paqad-ai

ground-in-project

Chat and Claude Code: Step 2 of the discover journey: records what exists today for the idea from the project's documents and code, with sources. Use when the journey directs it to step 2, not directly.

★ 8
Eliyce
Eliyce/paqad-ai

get-connected

Chat and Claude Code: Step 0 of the discover journey: confirms the repository and issue tracker are reachable, finds the evidence level and saves the connection setup. Use when the journey directs it to step 0, not directly.

★ 8
Eliyce
Eliyce/paqad-ai

decide-together

Chat and Claude Code: Step 4 of the discover journey: settles each open choice with the user, weighing the options, and records every decision and assumption. Use when the journey directs it to step 4, not directly.

★ 8