Imported from joaofssantos/qualroteiro (.claude/skills/make-workspace/SKILL.md). Install upstream with npx skills add joaofssantos/qualroteiro --skill make-workspace. Copyright stays with the author.

/make-workspace

Announce on entry: "Using make-workspace to clone the brain's repos."

Materializes the context's brain repos on the machine. You (the coordinator) do NOT clone by hand — you delegate to the typed CLI, which decides per repo (clone / skip / error), never overwrites anything, and updates state.yaml.

When to use / when NOT

Use it when: onboarding step 2 — .aipe/brain.yaml exists and the repos are not yet cloned into the workspace.

Do NOT use it when: there's no brain yet (run /context-brain first — cloning without the map is meaningless); or you only need to add one new repo (use /aipe-add-repo, which clones incrementally). This skill only clones — it never detects stack, creates worktrees, or edits the brain.

Flow

1. Confirm the workspace. By default it's the current directory (must be an aipe-<context> folder with .aipe/brain.yaml).

2. Check the precondition. The brain must exist. If there is no <workspace>/.aipe/brain.yaml, guide the PE to run /context-brain first — it makes no sense to clone without the map.

3. Run the CLI:

   aipe make-workspace --workspace <workspace>

4. Translate the output to the PE (one line per repo):

   • OK cloned <repo> → cloned now.
   • SKIP <repo> (already present) → was already there, nothing touched.
   • ERROR <repo>: <message> → failed (auth, network, or path occupied by different content). Explain and suggest the fix (e.g. grant access to the repo, move the occupied folder, or fix the URL in the brain via /context-brain).
   • STATE workspace=done|pending → aggregated state.

5. Next step: if workspace=done (all present), tell the PE this step is complete and to open a new session in this workspace to continue with /relationship (a fresh session keeps the coordinator's context clean; the SessionStart hook picks up exactly where onboarding left off). If pending, list what's missing to the PE; re-running is safe and only completes what's missing.

Rules

• Governance (MUST): you are the coordinator — you NEVER edit repo source yourself, because all code work must flow through the dispatch gate in /operate (decompose → dispatch a specialist in a worktree → PR) to keep the audit trail and worktree isolation intact; the non-exceptions there ("simple", "urgent", "one file", "I already know the fix") never apply. Here you only run the aipe CLI.
• Determinism (MUST): never clone or edit brain.yaml/state.yaml by hand — always through the CLI, so clone/skip/error decisions stay idempotent and nothing is overwritten.
• NEVER create worktrees here — that's a separate sub-project; doing it here corrupts the clean-clone assumption the later phases rely on.
• Auth failure is NEVER worked around (no credential-guessing, no URL rewriting): report the exact git message to the PE, because a silent workaround clones the wrong thing or leaks credentials.

Common mistakes

• "Fixing" an auth/access error by editing the URL → report the git message to the PE; the fix is granting access or correcting the brain via /context-brain.
• Re-cloning a repo that's already present → the CLI SKIPs it; trust the idempotent output, don't force.

Self-review gate (before telling the PE this step is done)

• Cloning ran through aipe make-workspace, not manual git clone.
• Every repo shows OK cloned or SKIP — any ERROR was reported verbatim, not worked around.
• STATE workspace=done before pointing the PE to /relationship; on pending, the missing repos were listed.