Git AI Archaeology

git-ai-archaeology

Produces a complete analysis of AI config evolution in a git repository. Finds when each AI configuration file was created, how AI-config commit velocity evolved month by month, which PRs structured the evolution, and identifies maturity phases.

Output: a single file {output_dir}/{slug}-git-archaeology.md

Expected Input

/git-ai-archaeology repo_path=/path/to/repo [output=./talks/slug] [slug=talk-name] [since=2025-01-01]

• repo_path: absolute path to the target git repo (required)
• output: output directory (default: ./talks)
• slug: output filename (default: repo folder name)
• since: analysis start date (default: first repo commit)

Workflow

1. Verify the repo: ensure the path exists and is a git repo
2. Global metrics: total commits, releases, contributors, time period
3. Section 1 — First commits: find creation date for key AI-config paths
4. Section 2 — Monthly distribution: commits filtered by AI-config keywords
5. Section 3 — Major PRs: extract and categorize significant AI-config commits
6. Section 4 — CHANGELOG: if CHANGELOG.md exists, extract releases with AI mentions
7. Section 5 — Phases: synthesize evolution phases
8. Save the output file

Step 1: Verification and Global Metrics

# Verify it's a git repo
git -C {repo_path} rev-parse --git-dir

# Global metrics
git -C {repo_path} log --oneline | wc -l                                    # total commits
git -C {repo_path} tag --sort=version:refname | wc -l                       # total releases
git -C {repo_path} shortlog -sn --no-merges | wc -l                         # contributors
git -C {repo_path} log --pretty=format:"%ad" --date=short | tail -1         # first commit
git -C {repo_path} log --pretty=format:"%ad" --date=short | head -1         # last commit
git -C {repo_path} log --merges --oneline | wc -l                           # merged PRs

Step 2: Section 1 — First Commits per AI-Config Path

For each path, find the origin commit with --diff-filter=A:

# Paths to analyze — adapt based on what exists in the repo
PATHS=(
  "CLAUDE.md"
  ".claude"
  ".claude/commands"
  ".claude/agents"
  ".claude/hooks"
  ".claude/skills"
  ".claude/rules"
  ".agents"
  ".cursor"
  "doc/knowledge-base.md"
  "doc/guides/ai-instructions"
  "doc/guides/ai-review"
)

for path in "${PATHS[@]}"; do
  git -C {repo_path} log --diff-filter=A --follow \
    --format="%ad | %H | %s" --date=short \
    -- "$path" | tail -1
done

Build the Section 1 table from results. Skip paths with no output (don't exist in this repo).

Also build the ASCII timeline:

{date} ─── {path} ─── {message}

Sorted chronologically.

Step 3: Section 2 — Monthly Distribution of AI-Config Commits

Filter commits by AI-config-related keywords:

# All commits with AI-config keywords
git -C {repo_path} log --format="%H %s" | \
  grep -iE "(claude|feat.ai|docs.ai|tech.ai|mcp|skill|hook|agent|llm|prompt)" \
  > /tmp/ai_commits_filtered.txt

# Count AI-config commits per month
git -C {repo_path} log --format="%ad %H" --date=format:"%Y-%m" | \
  while read month hash; do
    if grep -q "$hash" /tmp/ai_commits_filtered.txt; then
      echo "$month"
    fi
  done | sort | uniq -c

More direct alternative:

git -C {repo_path} log --format="%ad %s" --date=format:"%Y-%m" | \
  grep -iE " (feat|fix|docs|tech|chore|refactor)\(ai\)|claude|mcp.*server|\.claude/|skill|hook.*security|guardrail" | \
  awk '{print $1}' | sort | uniq -c

Compute per month:

• AI-config commit count
• % of monthly total (cross-reference with all-category monthly total)
• Context (if notable period)

Build ASCII distribution chart (horizontal or vertical bars).

Step 4: Section 3 — Major PRs and Commits

3.1 — feat(ai): / docs(ai): / tech(ai): commits

git -C {repo_path} log --format="%ad | %H | %s" --date=short | \
  grep -iE "\(ai\)|\(mcp\)|\[ai\]"

3.2 — MCP Server integrations

git -C {repo_path} log --format="%ad | %H | %s" --date=short | \
  grep -iE "mcp|serena|grepai|perplexity|sonar|postgres.*mcp|cursor.*mcp"

3.3 — Skills, commands, hooks, agents

git -C {repo_path} log --format="%ad | %H | %s" --date=short | \
  grep -iE "feat\(skill|feat\(hook|feat\(agent|feat\(command|feat\(dx\)|feat\(ci\)" | \
  grep -v "^$"

3.4 — Code review automation

git -C {repo_path} log --format="%ad | %H | %s" --date=short | \
  grep -iE "review|code-review|pr.*auto|ci.*review"

Step 5: Section 4 — CHANGELOG Analysis (if available)

# Check if CHANGELOG.md exists
ls {repo_path}/CHANGELOG.md

# Extract releases with AI mentions
grep -n "## \[" {repo_path}/CHANGELOG.md | head -30

Read the CHANGELOG and build a table:

Only list releases with AI-config content (CLAUDE.md, MCP, agents, skills, hooks, guardrails, prompts, etc.).

Step 6: Section 5 — Evolution Phases

Analyze collected data and identify maturity phases. Typical pattern:

Adapt phases to what the data actually reveals.

Identify the main inflection point: the month where AI-config commit volume spiked.

Compute the "recent vs historical" ratio (e.g., "81% of AI-config commits in the last 2 months").

Output Format: {slug}-git-archaeology.md

# Git Archaeology — AI Config Evolution: {slug}

**Source**: Git history of repo `{repo_path}` ({total_commits}+ commits, {total_releases}+ releases)
**Method**: `git log --diff-filter=A` for first commits, filtered monthly distribution, major PRs
**Last updated**: {date}

---

## Section 1: First Commit per Key Path

| Path | Creation Date | Commit Message | Hash |
|------|--------------|----------------|------|
{rows}

### Creation Timeline

\```
{ascii_timeline}
\```

---

## Section 2: Monthly Distribution of AI-Config Commits

| Month | AI-Config Commits | % of Total | Context |
|-------|-------------------|-----------|---------|
{rows}

### Visualization

\```
{ascii_chart}
\```

**Inflection**: {insight on the commit spike}

---

## Section 3: Major PRs and Commits Related to AI Tooling

### 3.1 PRs `feat(ai):` / `tech(ai):` / `docs(ai):`

| Date | Hash | Message | Impact |
|------|------|---------|--------|
{rows}

### 3.2 MCP Server Integrations (chronological)

| Date | MCP Server | Hash / PR | Role |
|------|------------|-----------|------|
{rows}

### 3.3 Skills, Commands, Hooks, Agents

| Date | Hash | Message | Category |
|------|------|---------|----------|
{rows}

### 3.4 Code Review Automation

| Date | Hash | Message |
|------|------|---------|
{rows}

---

## Section 4: CHANGELOG AI Mentions by Release

{section if CHANGELOG available, otherwise "Not applicable"}

---

## Section 5: Evolution Phases

### Evidence-Based Timeline

| Milestone | Exact Git Date | Git Evidence |
|-----------|----------------|-------------|
{rows}

### {N} Evolution Phases

#### Phase 1: {Label} ({period}) — {n} commits
{description}

#### Phase 2: {Label} ({period}) — {n} commits
{description}

#### Phase 3: {Label} ({period}) — {n} commits
{description}

#### Phase 4: {Label} ({period}) — {n} commits
{description}

### Key Insight

{Summary paragraph: main inflection point, recent/historical ratio, what the data reveals about the project's AI maturity.}

---
*Generated by git-ai-archaeology — {date}*
*Repo: {repo_path} | {total_commits} commits | {total_releases} releases*

Important Rules

• Read-only: no git commands that modify repo state
• Verify before asserting: a date not found in git = note "unverified"
• Adapt paths: Section 1 paths must be filtered to what actually exists in this repo
• Extensible keywords: if the repo uses different conventions (e.g., feat[ai] vs feat(ai)), adapt grep patterns
• Section 4 optional: if no CHANGELOG.md or no AI mentions, note "Not applicable" and skip to Section 5
• Adaptive phases: 4 phases is a common pattern, not a rule — 2 phases or 6 phases are equally valid

Anti-Patterns

• Inventing data not found in git
• Rounding numbers without flagging it
• Analyzing paths that don't exist in this repo
• Confusing a rename commit with a creation
• Omitting "flat" months (0 AI-config commits also tells a story)

Validation Checklist

• Repo verified and readable
• Section 1: only paths that exist in this repo
• Section 2: distribution covers the full repo period
• Section 3: commits sorted chronologically, hash included
• Section 4: cleanly skipped if no CHANGELOG
• Section 5: phases based on data, not the template
• Output file saved