← ClaudeAtlas

ai-adoptionlisted

Measure AI authorship in a git repo from commit signals (trailers, footers, bot identities) and render an executive deck. Local-only, no API calls. Use when the user asks "how much of this codebase is written by AI", wants an AI-adoption report, adoption trend, or human-vs-AI throughput. Methodology ported from StephanSchmidt/loupe (MIT); adds ulk's claude.ai/code trailer and accountability.jsonl ground truth.
izo/Ulk · ★ 1 · AI & Automation · score 71
Install: claude install-skill izo/Ulk
# AI Adoption Answers one question: **how much of this repo is written with AI assistance, and how is that trending?** Runs entirely on the local git history — no SaaS, no API keys, nothing leaves the machine. When invoked via `/ai-adoption`, follow the protocol below. Everything down to the `---` divider is the protocol Claude executes. After the divider is documentation for humans. --- ## What this is (and is not) This is the **local, single-repo** counterpart to `loupe` (StephanSchmidt/loupe, MIT). It reproduces loupe's commit-signal detection without any of its remote, multi-platform ingestion. Scope: - **Reproduces**: AI-authorship detection (trailers, footers, bot identities), weekly aggregation, adoption %, executive deck. - **Does NOT reproduce**: org-wide ingestion across GitHub/GitLab/Azure/Bitbucket + issue trackers (Jira/Linear). For that, wrap the `loupe` binary instead — reimplementing its connectors is the anti-pattern. On an ulk-managed repo this is **more accurate than loupe**: loupe must *infer* AI authorship from commits, whereas ulk's `claude.ai/code` trailer and `.ulk-reports/accountability.jsonl` are *ground truth*. Note that upstream loupe does **not** match the bare `claude.ai/code` URL trailer — this skill adds it, so ulk commits are detected correctly. ## Step 1 — Run the engine ```bash # from the repo root; --since is optional (any git date expression) "$HOME/.claude/skills/ai-adoption/scripts/ai-adoption.sh" --since "6 months ago" ``` The