git:20261020.a157858 to git:20261020.43b8a7e

4 added, 4 removed. Audit A to A.

---

name: ask-user

description: Ask which installed skill or workflow fits the user's situation.

disable-model-invocation: true

---

# Ask User

Route the user's situation to the smallest suitable workflow. Explain the recommendation briefly, then ask whether they want to invoke it. Do not invoke another user-invoked skill without confirmation.

Before stating what a skill does or recommending a step be skipped, read that skill's `SKILL.md`: the summaries here are for orientation only.

## Main Flow

For an idea that needs to ship:

1. Use `/grill-with-docs` to sharpen an idea in a codebase while updating `GLOSSARY.md` and ADRs. Use `/grill-me` when there is no codebase or durable project documentation is not wanted.

2. If a design question needs a runnable answer, use `/handoff` to branch into a fresh `/prototype` session, then `/handoff` the findings back.

3. For work that fits one session, use `/implement` directly. For multi-session work, use `/to-spec`, then `/to-tickets`, and start a fresh `/implement` session for each unblocked ticket. Use `/implement-spec` instead when the whole ticket graph should run in parallel worktrees and land on one integration branch.

- `/implement` drives `/tdd` at agreed seams and closes with `/code-review`. Use `/tdd` or `/code-review` directly when only that focused workflow is needed. Before creating, updating, or marking a PR ready, `/engineering-pr-review` runs the gate on top of that — `botica lint`, `/code-review`, the checks of the diff — and reports what it ran. `/test-smell-review` applies whenever tests are written or edited; reach for it directly to audit an existing suite for tests that pass without protecting anything.

+ `/implement` drives `/tdd` at agreed seams and closes with `/code-review`. Use `/tdd` or `/code-review` directly when only that focused workflow is needed. Before creating, updating, or marking a PR ready, `/engineering-pr-review` runs the gate on top of that (`botica lint`, `/code-review`, the checks of the diff) and reports what it ran. `/test-smell-review` applies whenever tests are written or edited; reach for it directly to audit an existing suite for tests that pass without protecting anything.

- Keep grilling, specification, and ticketing in one context window when practical, so each step builds on the same thinking. Between independently implementable tickets, clear context — see Phase Boundaries.

+ Keep grilling, specification, and ticketing in one context window when practical, so each step builds on the same thinking. Between independently implementable tickets, clear context; see Phase Boundaries.

## On-Ramps

- Incoming bugs and requests that are not already agent-ready: `/triage`.

- A hard bug, intermittent failure, or performance regression: `/diagnosing-bugs`.

- A huge, foggy effort whose path cannot fit one session: `/wayfinder`, then `/to-spec` when the decision map is clear.

- A completed coding session: `/retro` to identify improvements to the agent's environment and guardrails.

- A PR body to write: the internal `/pr` reads the target repo's `.github/pull_request_template.md`, after the `/engineering-pr-review` house gate.

Do not triage tickets produced by `/to-tickets`; they are already agent-ready. Do not send a resolved wayfinder map directly to `/implement` unless the effort proved genuinely small.

## Codebase Health

- Survey a codebase for deepening opportunities: `/improve-codebase-architecture`.

- Design a chosen module or seam using deep-module vocabulary: `/codebase-design`.

- Refactor an existing repo in tested slices, keeping its local conventions, with botica for the rules: `/engineering-refactor`.

- Resolve unclear or overloaded domain terminology: `/domain-modeling`.

## Phase Boundaries

A phase is a chunk of work inside a session: the grilling, the implementation, the review. Decide what to do with the context only at the boundary between two phases; mid-phase, continue or split the remaining work into subagents. Ask these in order and take the first yes.

1. **Continue** when the next phase needs this one as a primary source, or the remaining context window comfortably fits it. Continue costs nothing and loses nothing, so rule it out first.

2. **Clear** when everything in this session is disposable to what comes next.

3. **Handoff** (`/handoff`) when something has to travel: a different harness, a different directory or worktree, another person, or a side task forked mid-phase. What it buys is portability, so skip it when nothing travels.

4. **Subagent** when the task is scoped tightly enough to run unattended and report back.

5. **Compact** otherwise. This is the default landing spot, not the first reach; pass it an instruction so the summary keeps what the next phase needs.

Every option except Continue replaces the session with a summary of it, which is why Continue is questioned first.

## Standalone

- `/prototype`: answer one state, logic, or UI design question with throwaway code.

- `/research`: delegate primary-source research and capture cited findings in the repo.

- `/grilling`: run the interview primitive directly, with no wrapper flow around it.

- `/teach`: learn a concept over multiple sessions in a stateful workspace.

- - `/writing-for-agents`: guide the writing of documents agents consume — skills, `AGENTS.md`, pointed-at docs.

+ - `/writing-for-agents`: guide the writing of documents agents consume: skills, `AGENTS.md`, pointed-at docs.

- `/to-questionnaire`: turn a decision or interview into a shareable questionnaire.

- - `/wizard`: turn a procedure only a human can perform — provisioning, credentials, CI secrets, clicking through a third-party dashboard, a one-off migration — into an interactive script that walks them through it. Not for steps an agent can perform itself.

+ - `/wizard`: turn a procedure only a human can perform (provisioning, credentials, CI secrets, clicking through a third-party dashboard, a one-off migration) into an interactive script that walks them through it. Not for steps an agent can perform itself.

- `/wait-what`: re-pitch the last message when it did not land. Usable mid-conversation inside any other skill.

- `/journalist`: record or search session notes under `.journals/`.

## Precondition

Use `/setup-matt-pocock-skills` once per repository before engineering flows that depend on its issue tracker, labels, or domain documentation layout.

Credits and sources: [`references/CREDITS.md`](references/CREDITS.md).