This skill is invoked when a loose idea requires more than one Agent session to turn into a plan. It creates a stateful decision map in a Markdown file and guides the user through working a series of Tickets in order to resolve open questions—questions that may involve prototyping, research, or grilling. The map is domain-agnostic: it can be used to plan engineering work, course content, or anything else to which the same structure applies.
Decision Map
A decision map is a compact Markdown file, one file per planning effort, git-tracked alongside the project. It is the normative artifact—every session loads the entire map as context, so it must remain compact.
Assets created during a Ticket should be linked from the map, not copied into it.
Structure
entries ("Tickets"), each keyed by its short dash-case identifier (slug), which reads like a miniature title (e.g. relational-db、auth-strategy、cache-layer)—concise enough to save tokens, unique within the map.
## relational-db: 关系型还是非关系型数据库?
Blocked by: <slug>, <slug>
Status: open | in-progress | resolved
Type: Research | Prototype | Grilling
Question
Answer
The slug is the canonical ID, used for every Blocked by edge and in-text reference; the title after the colon is optional. When all Tickets in a Ticket's Blocked by list are in resolved status, that Ticket unblocked. Before starting work, a session should first **claim** its Ticket: set Status to in-progress and save the map so that concurrent sessions skip it.
Each Ticket must be sized to fit within a single 100K token Agent session.
Ticket Types
There are three types of Tickets:
- Research: Read documentation, third-party APIs, or local resources (such as knowledge bases). Create a Markdown summary as an asset. Use this type when knowledge outside the current working directory is needed.
- Prototype: Increase the fidelity of the discussion by producing a cheap, rough, concrete, touchable artifact—an outline, a rough first draft, a stub, or UI/logic code generated via the /prototype skill. Create a prototype as an asset. Use this type when the core question is "what should it look like" or "how should it behave."
- Grilling: A conversation with the Agent. Use the /grilling and /domain-modeling skills. Ask one question at a time. This is the default type.
Validation is not a fourth type—it is a red thread running through all three types.
Fog of War
The decision map is beyond the frontierdeliberatelyincomplete. Your task is to explore the frontier and resolve Tickets to push the frontier forward—advancing one node at a time, until the path to the endpoint is clear and no Tickets remain.
Invocation
Two branches. Either way,every session ends with aHandoffend—each session resolves at most one Ticket.
Create the map
The user initiates the invocation with a loose idea.
- Run a
/grilling+/domain-modelingsession to surface unresolved decisions. Ask one question at a time. - Write a new decision map — mostly fog, the frontier has been identified, and items that can be decided simply are resolved inline.
- Handoff. Building the map is a session's worth of work; do not resolve Tickets at the same time.
Work through the map
The user initiates the invocation by passing in the path to an existing map. Ticket slug optional— when not passed, you choose the next decision, not the user.
- Loadthe entire mapas context.
- Choose a Ticket. If the user specified a slug, use it. Otherwise choose the first
openthat is open and has beenunblockedTicket in document order.Claim it: setStatustoin-progressand save before any work begins. - Resolve it, invoking other skills as needed — including any skills listed in the
## Notesblock. When in doubt, use/grillingand/domain-modeling。 - Record the answer in the Ticket body, set
Statustoresolved。 - Add newly discovered Tickets (with the correct
Blocked byedges). If a decision made invalidates other parts of the map, update or delete those nodes. - Handoff.
The user may run unblocked Tickets in parallel, so expect other Agents to edit the map in their own sessions.
Handoff
At the end of each session, clear the context and open one or more new sessions. End with a Next steps block for the user to copy and paste. There are two cases:
There are still unclosed Tickets. List the currently unblocked Tickets, then provide two copy-paste options: a bare command for a single session (you choose the next Ticket), and a fixed command for each unblocked Ticket, for running in parallel. Paste each line into a new window — you can open just one, some, or all of them.
Next steps — 3 Tickets unblocked:
auth-strategy,cache-layer,rate-limits. Clear the context, then open new sessions.Single session — resolve the next unblocked Ticket:
调用 /decision-mapping,映射位于 <path>。Parallel — one window per line, up to 3:
调用 /decision-mapping,映射位于 <path>,ticket auth-strategy。 调用 /decision-mapping,映射位于 <path>,ticket cache-layer。 调用 /decision-mapping,映射位于 <path>,ticket rate-limits。
There are no unclosed Tickets left. The fog of war has been pushed back far enough that the path to the end is clear — the map is complete. (The initial interrogation phase may also have produced no fog at all, in which case there was no map to build from the start.) It is recommended to implement directly, or use /to-prd to schedule a multi-session implementation.
Notes
An optional block for declaring thedomain, the skills each session shouldconsult, and free-form fixed preferences that emerged during planning.