mine-history

When to use this skill

Mine this repository's git history for institutional knowledge the team accumulated while building it. Produce codebase directives (Type 1) that prevent AI assistants from repeating mistakes, and raw incident clusters (Type 2 candidates) for practice synthesis.

---

Step 1 — Verify and orient

Confirm we're inside a git repository:

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

If this fails, stop and tell the developer this skill requires a git repository.

Get the repo name and rough size so you can calibrate the analysis window:

git rev-list --count HEAD
git log --format="%ad" --date=short | tail -1
git log --oneline -5

Tell the developer:

• Repo name and total commit count
• Date of oldest commit
• What you're about to do: classify commits, read diffs for hot subsystems, synthesize directives, ask them what git missed

---

Step 2 — Select analysis window and read commits

Goal: land in the 150–400 commit range. This is enough history to find patterns without too much noise to classify. Adjust the window iteratively until you're in range — don't just apply a fixed time cutoff.

Starting point by repo size:

• < 200 total commits → start with all commits
• 200–2000 total commits → start with last 6 months

• 2000 total commits → start with last 3 months

Then count and adjust:

git log --oneline --after="6 months ago" | wc -l

• If count > 400: halve the window (6 months → 3 months → 6 weeks) and recount
• If count < 100: double the window (3 months → 6 months → 1 year → all) and recount
• Stop when you're in the 150–400 range, or when you've hit the full repo history

If the full repo history is still under 150 commits, use all of it and note that the repo is young or low-velocity.

Run the final window:

git log --format="%H %ad %s" --date=short --after="{final_window}"
# or without --after for full history

Tell the developer the window you landed on (e.g. "last 10 weeks, 287 commits") and offer to extend it after completing the analysis.