agentleFS
prime-radiant-incprime-radiant-inc/greenfield/skills/git-archaeology/SKILL.md

git-archaeology

Skill284 starsChanged 6 months agoApache-2.0View on GitHub

Layer 1 skill for mining git history for behavioral intelligence. Repository overview, commit message mining, PR/MR description extraction, blame-based maintenance heat maps, issue tracker cross-referencing. Loaded by the analyzer agent during Layer 1.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When it was added (the commit date, for version context)

    line 76
  2. Distinctive:

    If the remote is GitLab rather than GitHub:

    line 149
  3. Distinctive:

    Not all repositories have accessible PR history.

    line 172
  4. Distinctive:

    If PR history is not accessible, document the gap and move on.

    line 182
  5. Very specific:

    All claims from git history analysis use source=git-history:

    line 276
  6. Very specific:

    Every behavioral claim gets an inline citation immediately after the claim.

    line 291

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Git Archaeology Methodology

Extract behavioral intelligence from version control history. Commit messages, PR descriptions, blame annotations, and issue references encode design decisions, behavioral changes, and maintenance patterns that no other source captures.

When to Use This Mode

Git archaeology activates when:

  • The target is a git repository (or has a .git directory)
  • The discovery inventory identifies a git repository with meaningful commit history
  • Other modes discover a git repository during analysis

This mode runs independently of all other intelligence sources. It requires only access to the git repository and optionally to a GitHub/GitLab remote. All output is RAW (commit messages and PR descriptions may reference proprietary internals).

Why Git History Matters

Source code tells you what the system does NOW. Git history tells you:

  • What changed and why -- commit messages encode intent that is invisible in the current code
  • What broke and how it was fixed -- fix commits reveal failure modes and edge cases
  • What was deliberately removed -- reverted features and deleted code reveal abandoned behaviors
  • Where maintenance concentrates -- frequently modified files signal complexity and instability
  • What decisions were debated -- PR descriptions capture design trade-offs that never appear in documentation

Phase 1: Repository Overview

Goal: Establish the scope, age, and shape of the project.

# Commit count and date range
echo "Total commits: $(git rev-list --count HEAD)"
echo "First commit: $(git log --reverse --format='%ai' | head -1)"
echo "Latest commit: $(git log -1 --format='%ai')"

# Contributors
git shortlog -sn --no-merges | head -20

# Active branches
git branch -r --sort=-committerdate | head -20

# Tags (releases)
git tag --sort=-version:refname | head -20

# Commit frequency (commits per month, last 12 months)
for i in $(seq 0 11); do
  month=$(date -d "$i months ago" +%Y-%m 2>/dev/null || date -v-${i}m +%Y-%m)
  count=$(git rev-list --count --after="${month}-01" --before="${month}-31" HEAD 2>/dev/null || echo "0")
  echo "$month: $count"
done

# Top-level directory structure at HEAD
git ls-tree --name-only HEAD

Write to workspace/raw/project-history/overview.md.

Phase 2: Commit Message Mining

Goal: Extract behavioral claims from commit messages. Commit messages that describe features, fixes, and breaking changes encode behavioral contracts.

2.1 Feature Commits

# Conventional commits: feat
git log --oneline --grep="^feat" --no-merges | head -100

# Keywords indicating new behavior
git log --oneline --grep="add\|implement\|introduce\|support\|enable" -i --no-merges | head -100

For each feature commit, extract:

  • What behavior was added (the claim)
  • When it was added (the commit date, for version context)
  • The commit SHA (for provenance)

2.2 Fix Commits

# Conventional commits: fix
git log --oneline --grep="^fix" --no-merges | head -100

# Keywords indicating bug fixes
git log --oneline --grep="fix\|bug\|repair\|correct\|resolve\|patch" -i --no-merges | head -100

Fix commits are especially valuable because they reveal:

  • A behavior that was WRONG (the bug)
  • A behavior that is now CORRECT (the fix)
  • Edge cases that the original implementation missed

2.3 Breaking Change Commits

# Conventional commits: breaking
git log --oneline --grep="BREAKING" --no-merges | head -50

# Keywords indicating behavioral changes
git log --oneline --grep="breaking\|deprecat\|remov\|migration\|upgrade" -i --no-merges | head -50

Breaking changes reveal behavioral contracts that were considered important enough to announce their violation.

2.4 Detailed Message Extraction

For high-value commits (features, fixes, breaking changes), read the full commit message:

# Full message for a specific commit
git log -1 --format='%H%n%ai%n%an%n%n%B' <commit-sha>

Extract behavioral claims from the message body. Many commit messages contain:

  • Before/after descriptions -- explicit behavioral contract changes
  • Issue references -- links to detailed context
  • Test descriptions -- what was verified
  • Migration instructions -- how behavior shifted

Write to workspace/raw/project-history/behavioral-claims.md.

Provenance for Commit-Derived Claims

- The `--output csv` flag was added in v2.3
  <!-- cite: source=git-history, ref=abc1234, confidence=inferred, agent=git-archaeologist -->

Phase 3: PR/MR Description Mining

Goal: Extract behavioral intelligence from pull request and merge request descriptions. PR descriptions often contain the richest behavioral context: motivation, design decisions, testing notes, and review discussions.

3.1 GitHub PRs

# List merged PRs (most recent first)
gh pr list --state merged --limit 100 --json number,title,body,mergedAt,labels

# View a specific PR with full body and comments
gh pr view <number> --json title,body,comments,reviews,labels,mergedAt

# Search PRs by keyword
gh pr list --state merged --search "authentication" --limit 20 --json number,title,body

3.2 GitLab MRs

If the remote is GitLab rather than GitHub:

# List merged MRs
glab mr list --state merged --per-page 100

# View a specific MR
glab mr view <number>

3.3 What to Extract from PRs/MRs

For each PR/MR, extract:

  • Motivation -- why was this change made? What problem does it solve?
  • Design decisions -- what alternatives were considered? Why was this approach chosen?
  • Testing notes -- what was tested? How was it verified?
  • Review comments -- what concerns were raised? What was changed in response?
  • Behavioral claims -- explicit statements about what the code does or should do

PR descriptions that contain phrases like "this PR adds", "this fixes", "after this change", "the new behavior is" are direct behavioral claims.

3.4 Availability Check

Not all repositories have accessible PR history. Check first:

# GitHub: check if gh CLI is authenticated and repo is accessible
gh repo view --json name 2>/dev/null && echo "GitHub accessible" || echo "GitHub not accessible"

# Check if remote is GitHub or GitLab
git remote get-url origin

If PR history is not accessible, document the gap and move on. Do not treat inaccessibility as failure.

Phase 4: Blame-Based Maintenance Heat Map

Goal: Identify which parts of the codebase are actively maintained, recently modified, or fossilized. Maintenance patterns reveal where behavioral complexity concentrates.

4.1 Recently Modified Files

# Files modified in the last 90 days, sorted by modification count
git log --since="90 days ago" --name-only --no-merges --format="" | sort | uniq -c | sort -rn | head -50

4.2 Most Frequently Modified Files (All Time)

# Files with the most commits
git log --name-only --no-merges --format="" | sort | uniq -c | sort -rn | head -50

4.3 Fossilized Code

# Files not modified in over a year
git log --diff-filter=M --since="1 year ago" --name-only --no-merges --format="" | sort -u > /tmp/recently-modified.txt
git ls-files | while read f; do
  grep -qx "$f" /tmp/recently-modified.txt || echo "$f"
done | head -100

4.4 Churn Analysis

# Files with the highest churn (lines added + removed) in last 6 months
git log --since="6 months ago" --numstat --no-merges --format="" | \
  awk '{adds[$3]+=$1; dels[$3]+=$2} END {for(f in adds) print adds[f]+dels[f], adds[f], dels[f], f}' | \
  sort -rn | head -30

4.5 Interpreting the Heat Map

PatternSignal
High recent churn, many contributorsActive development, likely unstable behavior
High recent churn, single contributorFocused refactoring or feature work
No modifications in 1+ yearFossilized: either stable or abandoned
Frequent modifications to test filesBehavioral contracts actively evolving
Config files frequently modifiedDeployment or environment complexity

Write to workspace/raw/project-history/maintenance-heatmap.md.

Phase 5: Issue Tracker Cross-Reference

Goal: Extract issue references from commits and fetch the referenced issues for behavioral context.

5.1 Extract Issue References

# Find commits that reference issues (common patterns: #123, GH-123, JIRA-123)
git log --oneline --no-merges --format="%H %s" | grep -oE '#[0-9]+|[A-Z]+-[0-9]+' | sort | uniq -c | sort -rn | head -50

# Full commit messages that reference issues
git log --no-merges --grep='#[0-9]\+' --format="%H %ai %s" | head -50

5.2 Fetch Referenced Issues (GitHub)

# Fetch a specific issue
gh issue view <number> --json title,body,comments,labels,state

# List issues with behavioral labels
gh issue list --label "bug" --state all --limit 50 --json number,title,body,state
gh issue list --label "enhancement" --state all --limit 50 --json number,title,body,state

5.3 What to Extract from Issues

Issues provide behavioral context that commits alone cannot:

  • Bug reports -- describe expected vs. actual behavior (behavioral contract violations)
  • Feature requests -- describe desired behavior (behavioral requirements)
  • Discussion threads -- capture design decisions and trade-offs
  • Reproduction steps -- equivalent to test vectors

5.4 Availability Check

Issue tracker access depends on the hosting platform and repository visibility. If the issue tracker is not accessible, document the gap. Extract what you can from commit message references alone.

Write to workspace/raw/project-history/issue-references.md.

Provenance Rules

Source Type

All claims from git history analysis use source=git-history:

- The retry logic was added after users reported timeout failures on slow connections
  <!-- cite: source=git-history, ref=abc1234, confidence=inferred, agent=git-archaeologist -->

Confidence Levels

  • confirmed -- the commit message or PR description explicitly states the behavioral claim, AND a second source (e.g., a linked issue or test change) corroborates it
  • inferred -- the commit message or PR description directly states or strongly implies the behavioral claim from a single source
  • assumed -- the behavioral claim is derived from commit patterns, file names, or structural reasoning rather than explicit statements

Cite As You Go

Every behavioral claim gets an inline citation immediately after the claim. The ref field should be the commit SHA (short form is acceptable). For PR-derived claims, use ref=PR-<number>.

Output Structure

workspace/raw/project-history/
    overview.md                # Repository overview: age, size, contributors, structure
    behavioral-claims.md       # Behavioral claims extracted from commit messages
    maintenance-heatmap.md     # File modification patterns and churn analysis
    issue-references.md        # Issue cross-references and extracted behavioral context

Rules

  1. Git history is RAW -- commit messages and PR descriptions may contain proprietary information. All output goes to workspace/raw/.
  2. Recency matters -- recent commits are more likely to describe current behavior. Flag the date of every commit-derived claim.
  3. Commit messages are claims, not facts -- a commit that says "fix authentication bug" does not prove the fix is correct. It is an inferred claim until corroborated by another source.
  4. Absence of history is not absence of behavior -- a feature with no commit history may predate the repository or have been imported wholesale.
  5. Do not read source code diffs -- this is git ARCHAEOLOGY, not source analysis. Extract behavioral claims from messages, descriptions, and patterns. Source code analysis is a separate mode.
  6. Document what is inaccessible -- if PR history, issue trackers, or remote repositories are not reachable, document the gap explicitly. Partial results are valuable.
  7. Cite as you go -- every behavioral claim gets an inline <!-- cite: --> comment immediately after the claim. Never defer citation to a later step.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.