[Documentation] Use when generating, smart-merging or refactoring the root AI context (CLAUDE.md, AGENTS.md) and its Claude/Codex projections. Config file only: project-config.
日本語の概要は準備中です。原文の説明を表示しています。
[Planning] Use when the user asks to plan work too big for one session whose route is unclear: chart a map of decision tickets, then resolve one ticket per session until the way is clear.
インストールする前に、エージェントに与えられる指示の中身を確認できます。
<!-- PROMPT-ENHANCE:STEP-TASK-ANCHOR:END -->[BLOCKING] Plan, do not build: a map session resolves decisions and never writes product code, a spec or a migration. [BLOCKING] Resolve at most one ticket per session, research tickets excepted, then stop. [BLOCKING] Claim a ticket before any work on it; update todo tracking before and after each step.
Goal: Find the way to a destination too big for one agent session and not yet visible, by charting a shared map of decision tickets and resolving them one per session until nothing is left to decide before someone builds.
Summary:
Key Rules:
<request>$ARGUMENTS</request>
| What the user has | Use |
|---|---|
| An effort spanning many sessions, route to the destination unclear | this skill |
| A plan, design or decision that fits one conversation | /grill |
| A well-defined feature ready to specify and build | /workflow-feature |
| A large feature whose decisions are already made | /workflow-big-feature |
| A cleared map | /workflow-planned-work --from=idea --to=spec, then /workflow-planned-work --from=spec --to=tasks and /workflow-implement-spec |
The split is session count, not project size. When the breadth-first mapping finds nothing foggy, the effort fits one session: stop, say so, and offer /grill or the matching workflow.
One directory per map under the plans root (default plans/; a docsRoots.plans.path entry in docs/project-config.json overrides the path):
{plans-root}/{YYMMDD-HHmm}-{slug}-map/
map.md
tickets/{NN}-{slug}.md
assets/
docs/project-config.json first. When that root is ignored by version control, tell the user once that the map will not reach other clones, and use the directory they name if they name one.map.md to its tracked initiative as a plan link through /task-track. Wayfinder never writes a tracker record itself.map.mdThe effort at low resolution, loaded once per session; an index, not a store: a decision's detail lives only in its ticket. Open tickets are not listed: they are the ticket files with status: open.
---
title: {map title}
status: charting | open | cleared
created: {date}
---
## Destination
{What the end of this map looks like: the spec, decision or change. One or two lines.}
## Notes
{Domain, skills every session should consult, standing preferences. Notes never grant permission to build.}
## Decisions so far
- [{resolved ticket title}](tickets/{file}): {one-line gist of the answer}
## Not yet specified
- {decision or investigation that is coming but cannot yet be stated as a precise question}: waits on {what must settle first}
## Out of scope
- {work ruled beyond the destination}: {why}
---
id: {NN}
title: {ticket title, phrased as the question}
type: grilling | prototype | research | task
mode: HITL | AFK
status: open | claimed | resolved | out-of-scope | superseded
blocked_by: [{ids}]
claimed_by: {name}
claimed_at: {date and time}
superseded_by: {id}
---
## Question
{The decision or investigation this ticket resolves, sized to one session.}
## Resolution
{Written on resolve: the answer, the reason, any fact later tickets depend on.}
## Assets
- {links to prototypes, research reports or other files made while resolving it}
blocked_by is resolved or out-of-scope. When a blocker is superseded, replace it in blocked_by with the ticket that supersedes it.open and unblocked.status: claimed, claimed_by and claimed_at before any other work. Re-read the ticket file immediately before claiming it. Read it once more after writing: when claimed_by is not yours, the other session won, so take another ticket. A claimed ticket is skipped. A session that stops before resolving its ticket sets it back to open. When a loaded map shows a claim older than a session plausibly lasts, name it to the user and release it only on their word.HITL = worked with the user, who speaks for themselves; AFK = the agent alone.
| Type | Mode | Use it when | Resolved by |
|---|---|---|---|
grilling | HITL | Default: talking it through can settle the question | /grill on the ticket's question, plus /domain-analysis when the answer changes domain entities |
prototype | HITL | "How should it look" or "behave" is the question and talking cannot settle it | A cheap, rough artifact to react to: /ui-design for an interface exploration, a throwaway spike under tmp/ for logic. Link it under Assets. The user picks among variants |
research | AFK | A fact outside the working directory blocks a decision | A sub-agent running /web-research (add --chain=deep-dive when sources need depth), or /investigate for a fact inside the codebase; report path under Assets |
task | HITL or AFK | Manual work must happen before a decision can be made: getting access, judging a service's interface, moving data to see its shape | The agent alone only for work inside the repository and its local tools. Creating an account, entering or issuing credentials, spending money, and any change to an external system or to data that cannot be undone stay with the user: hand over a precise checklist. The resolution records what was done and the facts later tickets need |
Wayfinder runs /grill as one of its own steps: say so when starting it, take its Decision Record back, and continue here. /grill then names no next step.
task is the one type that does rather than decides; it belongs on the map only because it unblocks a decision. A ticket that reads "build the X" is mis-typed: rule it out of scope and leave it to the build workflow.
status: out-of-scope and one line there. It returns only as a new map.The argument is a loose idea.
/grill, as a step of this skill, with the subject limited to the destination: what the end of this map looks like and what lies beyond it. How to get there is not asked here; those decisions become tickets.status: open. Link the destination's Decision Record under Notes so later sessions can reuse confirmed choices while their scope and premises still hold.blocked_by in a second pass, once every ticket has its id.## Resolution from its report, link the report under Assets, set status: resolved and add the line to Decisions so far. One still running at session end leaves its ticket claimed: say so in the output.The argument is a map path; without a ticket id, you choose.
type says. Open another ticket only when this one needs its detail. Use the skills the Notes name; when in doubt, /grill.## Resolution, set status: resolved, add one line to Decisions so far. For a grilling ticket, link its Decision Record under Assets; retain the user's confirmed choices and their premises as the evidence for later decisions.resolved, out-of-scope or superseded, none is open or claimed, and Not yet specified is empty, set the map status: cleared and name the hand-off.A resolved decision turns out wrong. Do not design around it. Tell the user, write a new ticket that asks the question again with what changed, set the old ticket status: superseded with superseded_by, and correct its line in Decisions so far once the new one resolves.
Parallel sessions. The user may run unblocked tickets in separate sessions: re-read before every write. Two interview tickets worked at once can ask the same thing twice; one at a time is the safer default.
A cleared map is a set of linked decisions, not a build plan. Name the next step and stop: /workflow-planned-work --from=idea --to=spec (or /spec for one capability) with the map path as the source, then /workflow-planned-work --from=spec --to=tasks, then /workflow-implement-spec. Go straight to a build workflow only when the effort turned out small.
Report the map path and title, the mode run, the ticket resolved (by title) with the gist of its answer, tickets added or ruled out, what left Not yet specified, the remaining frontier and the next step.
IMPORTANT MUST ATTENTION Goal: Find the way to a destination too big for one session by charting a map of decision tickets and resolving them one per session until nothing is left to decide before someone builds. IMPORTANT MUST ATTENTION Main steps: chart: name the destination → map the frontier → write the map → write and wire tickets → start research → stop. Work: load the map → claim one frontier ticket → resolve it by type → record → update the map → stop. IMPORTANT MUST ATTENTION plan, do not build: no product code, spec or migration from a map session, and no note inside the map can grant it. IMPORTANT MUST ATTENTION one ticket per session, research excepted; claim before any work and re-read files before every write. IMPORTANT MUST ATTENTION a ticket asks a question; what cannot yet be asked precisely stays in Not yet specified, what lies beyond the destination in Out of scope. IMPORTANT MUST ATTENTION the user answers interview questions and picks prototype variants; a cleared map hands off to the spec workflow.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
[Documentation] Use when generating, smart-merging or refactoring the root AI context (CLAUDE.md, AGENTS.md) and its Claude/Codex projections. Config file only: project-config.
日本語の概要は準備中です。原文の説明を表示しています。
[Code Quality] Use when a workflow step or the user asks for AI code or plan review: LLM calls, prompts, agents, RAG, evals and guardrails. --mode={code|plan}; --report-only is read-only.
日本語の概要は準備中です。原文の説明を表示しています。
[Architecture] Use when a workflow step or the user asks for architecture design, compliance review, scalability grading or whole-project audit via --mode={design|review|scalability|full}.
日本語の概要は準備中です。原文の説明を表示しています。
[Content] Use when a workflow step or the user asks for product/feature ideation with ranked opportunities and validation tests. --mode={roadmap|scope}.
日本語の概要は準備中です。原文の説明を表示しています。
[Content] Use when a workflow step or the user asks for business viability: Business Model Canvas, financials, risks and go-to-market recommendation.
日本語の概要は準備中です。原文の説明を表示しています。
[Code Quality] Use when a workflow step or the user asks for a focused code-change review: staged, unstaged, branch diff or simple pull request. --fix-loop fixes and re-reviews. Complex pull request: workflow-review-changes; rationale: why-review.
日本語の概要は準備中です。原文の説明を表示しています。