feat(ci): add AI Contribution Report workflow

[satish-m]

Adds a GitHub Actions workflow that automatically measures AI tool involvement across the repository's full commit history on every merge to main.

What it does

• Triggers on every push to main
• Scans all commit messages for Co-Authored-By: trailers stamped by AI coding tools (Cursor, Claude Code, GitHub Copilot)
• Counts commits attributed to each tool vs. purely human-authored
• Renders a markdown summary table directly in the GitHub Actions job summary (visible in the Actions tab after each merge)

Signal measured

The Co-Authored-By: trailer is written automatically by:

• Cursor → Co-authored-by: Cursor <cursoragent@cursor.com>
• Claude Code → Co-Authored-By: Claude <noreply@anthropic.com>
• GitHub Copilot → Co-authored-by: GitHub Copilot <...>

Commits without any such trailer are counted as human-only. Note that commits may still have had AI assistance without a trailer (e.g. code pasted from a chat interface), so this is a lower-bound estimate.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[satish-m]

I have read the CLA Document and I hereby sign the CLA

[fercor-cisco]

🤖 This review was generated by the Astra agent (claude-opus-4-8). It may contain mistakes.

Verdict: approve — Informational CI reporting workflow; logic is sound for the common case, with a few correctness/hardening improvements worth addressing but none merge-blocking.

Follow-ups

Suggested follow-up work that could be tracked as Shortcut stories:

• .github/workflows/ai-contribution.yaml:8-11: Consider adding an explicit least-privilege permissions: block (e.g. permissions: contents: read) to the job. The step only writes to $GITHUB_STEP_SUMMARY and never uses the token, so restricting the default GITHUB_TOKEN scope is good hardening across CI workflows in this repo.
• .github/workflows/ai-contribution.yaml:3-6: The job re-scans the entire commit history on every push to main and only renders an ephemeral job summary (no artifact/badge/trend). Consider a concurrency group to cancel superseded runs, and optionally persisting the numbers somewhere durable if trend tracking is desired.

[fercor-cisco]

🟡 minor (bug): grep ... | wc -l counts matching lines, whereas total counts commits. Two problems follow:

1. A single commit that carries more than one AI trailer (e.g. both Co-Authored-By: Claude and Co-authored-by: Cursor, which happens on hand-offs between tools) is counted once in total but once in each of cursor/claude/copilot. That inflates ai above the true count of AI-attributed commits and makes human=$((total - ai)) potentially negative, which then renders as a negative count and a negative percentage.
2. grep -i "co-author.*cursor" matches on the whole commit body, so a prose mention (e.g. "discussed co-author roles with the Cursor team") is counted as a trailer.

To count commits rather than trailer lines, and to anchor on the trailer, consider matching per-commit. For example, count commits whose body contains the trailer:

cursor=$(git log --format="%H %(trailers:key=Co-authored-by,valueonly)" | grep -ic "cursor")

or iterate commits and grep each body once, so a commit is attributed at most once per tool.

🤖 Generated by Astra

[fercor-cisco]

🟡 minor (bug): Every Share cell divides by $total. On a repository with zero commits reachable from HEAD, total is 0 and $(( cursor * 100 / total )) aborts the step with a divide-by-zero. Unlikely on main today, but cheap to guard. Consider computing shares only when total > 0, e.g.:

if [ "$total" -eq 0 ]; then total=1; fi

(or skip the percentage column when there are no commits).

🤖 Generated by Astra

[fercor-cisco]

🔵 nit (other): Integer division truncates each share, so the per-row percentages will not sum to 100% (e.g. three rows at 33% each show 99%). If exactness matters for the summary, compute the human share as 100 - ai_share rather than independently, or round. Non-blocking.

🤖 Generated by Astra