run

kagura-engineer: run

Thin wrapper around the kagura-engineer run CLI verb. It drives one GitHub issue through the full memory-grounded loop to an open PR. It reimplements nothing — the orchestration lives entirely in the CLI; this skill discovers config, gates on doctor, shells out, and surfaces the result.

⚠️ This is a Harness, not an atomic tool. Before launching, make sure the user understands the consequences and has confirmed:

• Repo mutation — creates a run-<issue> git worktree and commits there.
• PR creation — opens a GitHub pull request via gh-issue-driven on success.
• Cost — invokes claude -p multiple times (start / implement / ship).
• HITL — gates can halt (blocked, exit 2) and wait for a human decision.

Announce: "Using the kagura-engineer:run skill — this is a Harness that will create a PR."

No-argument usage

If invoked without an issue number, do NOT guess or shell out — print this and stop:

kagura-engineer:run <issue-number>
  Drive one GitHub issue to a PR via the kagura-engineer harness.
  ⚠ Harness: creates a worktree + commits, opens a PR, spends model budget, gates on HITL.
  Example:  kagura-engineer:run 42
  New here? run kagura-engineer:doctor first to check the environment.

Steps

1. Validate the issue argument. The CLI takes a single positive integer issue number. Reject anything that does not match ^[1-9][0-9]{0,8}$ before interpolating it into the shell — do not pass free-form text through.

2. Discover config. repo.yaml in CWD by default; pass --config <path> otherwise.

3. Precondition — run doctor:

   kagura-engineer doctor --json
   

   If overall == "fail" (exit 1), stop: surface the blocking checks and hand off to kagura-engineer:setup. Do not proceed to run. (ok/warn may proceed.)

4. Confirm with the user (the Harness warning above) unless they have already opted into an unattended run.

5. Run:

   kagura-engineer run <issue> --json
   

   Add --unattended to suppress interactive gates, --no-remember to skip the memory savepoint. Without --json, phase progress streams live.

6. Interpret the status. status == "ok" (exit 0) → PR created (pr_url). "blocked" (exit 2) → a gate halted; show resume_hint and surface the verdict for a human decision. "fail" (exit 1) → hard error; show the failing phase's detail from the phases array (there is no top-level detail key).

7. Surface & hand off. Print the phase table and pr_url. On success, suggest kagura-engineer:review <pr_url> to review the PR. On blocked/fail, surface the resume_hint so the user can retry after addressing the cause.