| name | whats-next-pocock |
| description | Recommend the next Matt Pocock-style skill from the current repo context. Use when the user asks which skill, flow, or next agent step should run, especially when they are learning how the skills compose. |
What's Next Pocock
Pick the next skill. Do not give the user a catalogue. Give them the next move, what it produces, and what it feeds. This is a router, not an executor: explore, recommend, then stop.
Process
1. Gather context
Read the user's request and the repo-local instructions that are actually present. Do not bind this skill to a specific CLI, TUI, harness, or filename.
- Start with instruction files explicitly supplied by the current run.
- Then check common repo files such as
AGENTS.md and CLAUDE.md.
- Treat files with
## Agent skills or references to docs/agents/* as workflow context.
- If multiple files conflict, mention the conflict. Prefer an explicitly supplied file; otherwise do not pretend the conflict is resolved.
Check the per-repo setup these skills assume:
- A repo-local instruction file describes the agent skills.
docs/agents/issue-tracker.md, docs/agents/triage-labels.md, and docs/agents/domain.md exist.
If setup is missing or inconsistent, recommend setup-matt-pocock-skills first. Do not guess the issue tracker, labels, or domain docs.
2. Read the current skills
- Prefer
~/.agents/skills/*/SKILL.md.
- Use the visible session skill list as a fallback.
- Use current
name and description values when naming skills.
- Notice new Pocock-style skills, but do not invent a new workflow from them.
This check is internal. Do not print a "catalog checked" field.
If a new skill clearly supersedes one of the flows below, recommend the new skill. If it might matter but its place in the workflow is unclear, list it under Possibly relevant.
3. Pick the flow
Prefer the happy path. The goal is to remove the user's guesswork, not enumerate every possible skill.
- Vague idea:
grill-me or grill-with-docs → to-prd → to-issues → tdd
- Feature already discussed:
to-prd → to-issues → tdd
- PRD, plan, or spec already exists:
to-issues → tdd
- Issue creation, labels, or AFK-agent prep:
triage
- Bug, regression, exception, failing test, or performance problem:
diagnose
- Unfamiliar code area:
zoom-out
- Architecture friction, shallow modules, or testability problems:
improve-codebase-architecture
- Interface design request: use
design-an-interface if it exists; otherwise use grill-with-docs.
- General design grilling outside repo docs:
grill-me.
If the user already named a skill, check whether it fits. Suggest a different skill only when the mismatch is clear.
4. Answer
Next: `<skill-name>`
Why: <one or two sentences grounded in the repo/request>
Repo evidence: <what you saw>
What it produces: <artifact, decision, issue, test loop, or understanding>
What feeds the next step: <how this output is used next>
Possibly relevant: <new/related skills, or "None">
Open questions for the human: <questions, or "None">
Suggested prompt: <short prompt the user can send next>
If several steps are needed, show the sequence and mark the immediate next skill clearly.
Expand uncommon acronyms inline the first time if they matter to the recommendation.
Rules
- Do not mutate files.
- Do not auto-handoff into the next skill.
- Do not duplicate target skill instructions.
- Do not answer with a generic menu.
- Keep the recommendation short and grounded in the repo/request evidence.