check-repo

check-repo — readiness assessment for harny adoption

Interview-style walk through ten readiness dimensions. Goal: produce a scorecard the user can act on and a prep checklist of fixes to apply before the first harny run.

This skill observes the repo and asks the user only what cannot be inferred. Use the tools to look at the repo first; ask only when you need judgment or context.

---

Step 1 — Locate the repo and frame

The cwd of the conversation is the target repo. Confirm by running:

git rev-parse --is-inside-work-tree
git rev-parse --show-toplevel

If the cwd is not a git repo, stop and tell the user. harny needs git.

Greet briefly, no preamble:

Walking your repo through the harny readiness checklist. Ten dimensions, R/Y/G each. I'll observe what I can and ask you when I need judgment. End output is a scorecard you can act on.

---

Step 2 — Optional: validate the validator first

Before scoring, read the "Validate the validator" section in ${CLAUDE_SKILL_DIR}/dimensions.md and ask the user:

Have you already run the install + validator command on main and confirmed it exits 0? If not, we should do that before scoring — it's the single most useful pre-flight step.

If they say no, walk them through it (steps in dimensions.md). If they say yes, proceed.

---

Step 3 — Walk the ten dimensions

Load the full dimensions reference once:

Read ${CLAUDE_SKILL_DIR}/dimensions.md

For each of the 10 dimensions, in order:

1. Read aloud the dimension's "Why it matters" — one short paraphrase, not the full block. The user does not need to hear the doc verbatim.
2. Gather evidence from the repo using whatever tools fit the dimension. Examples:
   • Dim 2 (safe-reset): git status, look for floating files at root, check .gitignore coverage
   • Dim 3 (install): identify lock files, check CI workflow vs declared dependencies
   • Dim 6 (docs): look for CLAUDE.md, ARCHITECTURE.md, module-level docs
   • Dim 8 (branches): git branch -a | wc -l, git branch -r
   • Dim 9 (lint/type): identify configured tools, check whether CI blocks on them
3. Ask only what is needed — the diagnostic question or specific judgment calls. Do not ask the user to answer the checklist items one by one; you should infer most of them.
4. Score Red / Yellow / Green using the calibration in dimensions.md, with a one-sentence rationale that names the specific evidence.
5. Move on. Do not lecture, do not propose fixes inline — fixes go in the prep checklist at the end.

Pace: 30-60 seconds per dimension on average. Heavier dimensions (1, 7, 9) may need a question; lighter ones (8, 10) often score from observation alone.

---

Step 4 — Anti-signals

After the 10 dimensions, scan for anti-signals (in dimensions.md). These are not scored — they are red flags that may make the repo a poor fit for harny regardless of dimension scores:

• "Done" is subjective (UX/design-heavy)
• Legacy without types or tests
• Cross-service changes are the norm
• Build requires hand-holding
• Active human edits in the same files concurrently
• Heavy credential dependence

Note any present.

---

Step 5 — Produce the scorecard

Load the template:

Read ${CLAUDE_SKILL_DIR}/scorecard-template.md

Fill it in. Include:

• One-line rationale per dimension citing the evidence
• Anti-signals list (or "none observed")
• Overall verdict: ready / ready with prep / not yet
• A prep checklist of concrete fixes the user should apply before the first run
• A first-task suggestion: small, surgical, one-module — never a sweeping refactor for the first run

Output the scorecard to the conversation. Do not write it to a file unless the user asks.

---

Step 6 — Offer next steps

Brief offer of follow-ups, only if relevant:

• If ready with prep: list the top 1-3 prep items the user could fix now.
• If not yet: explain the single biggest blocker.
• If ready: suggest a first task and which skill helps next (e.g., /harny:harny for orchestration model).

Do not auto-invoke other skills.

---

What this skill does NOT do

• Does not write or modify any file in the user's repo.
• Does not commit, branch, or push anything.
• Does not invoke harny — that is the orchestrator agent's job.
• Does not auto-fix the items it flags. It produces a checklist; the user (or /harny:harny) drives action.

---

Edge cases

• Repo with no commits. harny needs at least one commit. Tell the user git commit --allow-empty -m "initial" before retrying.
• Repo without CLAUDE.md. Score dimension 6 honestly (likely 🔴 or 🟡), do not lecture.
• Mid-walk interruption. If the user stops you with a question, answer it tersely and resume from the dimension you were on.
• User pushes back on a score. Take the feedback. The user knows their repo better than you do. Re-score with their context.