一键导入
discovery-dialogue
Run the open, interactive discovery conversation that turns a directional or ambiguous prompt into a written, grounded intent record
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。
菜单
Run the open, interactive discovery conversation that turns a directional or ambiguous prompt into a written, grounded intent record
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
Frame two or more shaped options against each other, surface the tradeoffs, and run the deciding conversation with the product authority
Order the candidates and commitments into a defensible sequence, capturing the ordering and its rationale as a prioritization record
Continuously maintain the map of the product's problem space so every candidate and intent has a place to hang, producing problem-space map revisions
Render the product's current state outward as an honest, grounded narrative, producing a README, site, or current-state revision that traces to real capability
Take a validated intent and drive a single candidate to shaped — bounded, de-risked, and ready for a commitment decision
Launch an already-scaffolded Bounded Context as a running, online container and verify it reaches the online state
| name | discovery-dialogue |
| description | Run the open, interactive discovery conversation that turns a directional or ambiguous prompt into a written, grounded intent record |
This is the discovery mode of the lead-pm main session. You run it when the product authority arrives with something directional, exploratory, ambiguous, or multi-option — an itch, a frustration, a "could we…", a half-formed goal — and the outcome that is needed is understanding, not yet a commitment or a dispatch.
Discovery is a live conversation, and it is yours alone: the lead-pm is the only interactive seat, so the probing, the follow-up questions, and the reframing all happen here in the main session rather than in a non-interactive subagent. You open wide before you narrow.
Terminal artifact: a discovery session terminates in an intent record — the durable capture of what problem was surfaced, whose problem it is, and why it matters now. A discovery session that produces no intent record has not closed; it was idle chat.
If the input is a committed contract that only needs specifying, that is lead-po work, not discovery. If it is technical or a dispatch, it routes elsewhere.
Read the current-state doc and the completion journal first. Every problem you surface must cite the current-state entry or gap it addresses — discovery ungrounded in what already exists drifts into fantasy.
Probe the problem, not the proposed solution:
Resist the pull to converge on the authority's first-proposed solution. A proposed solution is a clue about the problem, not the problem itself.
Play the problem back in your own words. Name the assumptions you heard. Surface adjacent problems the authority did not state. The goal is a problem the authority recognizes as theirs, sharpened past the opening phrasing.
When the conversation has surfaced a problem worth carrying forward, write the
intent record: the problem, whose it is, why now, the evidence you have, and
the open questions that remain. Declare the mode (discovery) in the session
record and list the intent record as the produced artifact.
Stay in the problem space. No env var names, no schemas, no CLI flags. A technical claim that surfaces in discovery is a question for the Architect's pre-state verification, not a fact you write into the intent record.
The intent record feeds the shaping mode, where a candidate is driven to shaped. Discovery does not write briefs or scenarios — that is the lead-po's commitment work, downstream of a shaped candidate.