Repo Bootstrap

Repo Bootstrap

Reconcile one repository's pack-facing setup without replacing repository-owned choices or starting the workflow that requested setup. Other skills read the installed docs/agents/* contracts directly. They may recommend $repo-bootstrap and stop for a setup gap, but they never invoke or load this package as runtime guidance.

1. Inspect

Resolve the repository root and record the current HEAD, index, worktree, and the contents of any file that may change. Inspect existing setup pointers, repository-owned commands, remotes, and only the setup branches relevant to the request or an observed incompatibility.

When managed setup already exists, run scripts/validate_setup.py as discovery. Compare an applicable target with its current owner; a marker or validator result is structural evidence, not proof that the setup works.

2. Reconcile

Preserve repository instructions, confirmed choices, local additions, staged content, and unrelated work. Ask only about a consequential choice that the repository does not settle.

• Load one tracker seed only when tracker setup applies: GitHub, GitLab, or Local Markdown. Another tracker needs a concrete operation and read-back map.
• Load triage-labels.md only when an active selected workflow needs labels. Reuse or map existing labels before proposing creation.
• Load domain.md only to choose or repair domain routing. $domain-modeling owns domain truth.
• Reconcile parallel support only when the user requested it or $parallel-implement returned that exact blocker. Follow its current lane configuration requirements; do not create worktrees or lane directories.

Prefer the detected tracker. Default domain routing to single-context unless independent models or language require multiple contexts.

3. Propose

Show the exact local edits, external operations, preserved content, unresolved conflicts, and nearest useful proof. With no delta, change nothing and continue to Verify. Otherwise wait for the current user's approval of that exact delta.

4. Apply

Refresh every affected target before writing. If an affected target changed, stop and propose again. Apply only the approved setup-owned edits and create only approved missing labels.

Setup approval does not authorize domain truth, tracker items, worktrees, dependency installation, staging, commits, pushes, or changes to HEAD.

Read back each external effect. If an operation fails or its result is unknown, stop, inspect the affected state, and report what is known to have happened and the safest next action. Do not retry blindly or assume rollback.

5. Verify

Run the validator and inspect the changed files and diff. Read back applicable tracker configuration and labels. Run a repository command only when it can disprove a claim made by the setup change. Confirm that unrelated work, the index, and HEAD were preserved.

Complete when the applicable setup is compatible and every external change was read back, or when the exact blocker and observed partial state have returned to the user. Leave the recommending workflow unstarted.