git-ai-search

name	git-ai-search
description	Search and restore AI conversation context from git history
argument-hint	[search query or workflow]
allowed-tools	["Bash(git-ai:*)","Read","Glob","Grep"]

Git AI Search Skill

Search and restore AI conversation context from git history using git-ai search and git-ai continue.

Overview

Git AI tracks AI-generated code and the conversations that produced it. This skill helps you:

• Search - Find AI prompt sessions by commit, file, pattern, or author
• Continue - Restore AI session context to continue work where someone left off
• Show - View the full transcript of a specific prompt

When to Use Each Command

You want to...	Use...
See what AI context exists for a commit	git-ai search --commit <sha>
See AI context for specific lines	git-ai search --file <path> --lines <range>
Read the full AI conversation	git-ai search --commit <sha> --verbose
Get machine-readable output	git-ai search --json
Find prompts mentioning a topic	git-ai search --pattern "keyword"
Continue someone's AI session	git-ai continue --commit <sha>
Continue working on a code region	git-ai continue --file <path> --lines <range>
Launch Claude with restored context	git-ai continue --commit <sha> --launch
View a specific prompt transcript	git-ai show-prompt <id>

Workflow Patterns

1. Investigate a Commit

Understand what AI assistance was used in a specific commit:

# See a summary of AI sessions in the commit
git-ai search --commit abc1234

# View the full conversations
git-ai search --commit abc1234 --verbose

# Get JSON for programmatic processing
git-ai search --commit abc1234 --json | jq '.prompts | keys[]'

2. Understand a Code Region

Find what AI sessions contributed to specific lines:

# Search by file
git-ai search --file src/main.rs

# Search specific lines
git-ai search --file src/main.rs --lines 100-150

# Multiple line ranges
git-ai search --file src/main.rs --lines 50-75 --lines 200-250

3. Continue Someone's Work

Pick up where a teammate left off:

# See what they were working on
git-ai search --commit abc1234 --verbose

# Restore context and continue in your shell
git-ai continue --commit abc1234

# Or launch directly into Claude Code
git-ai continue --commit abc1234 --launch

# Or copy context to clipboard to paste elsewhere
git-ai continue --commit abc1234 --clipboard

4. Review a Pull Request

Understand AI involvement in a PR:

# Get all commits in the PR
COMMITS=$(gh pr view 123 --json commits -q '.commits[].oid')

# Search each commit
for sha in $COMMITS; do
  echo "=== $sha ==="
  git-ai search --commit $sha
done

# Or search the full range
BASE=$(gh pr view 123 --json baseRefOid -q '.baseRefOid')
HEAD=$(gh pr view 123 --json headRefOid -q '.headRefOid')
git-ai search --commit $BASE..$HEAD

5. Audit AI Involvement in a File

See all AI sessions that touched a file:

# Full file history
git-ai search --file src/auth/login.rs

# Filter by author
git-ai search --file src/auth/login.rs --author "Alice"

# Filter by tool
git-ai search --file src/auth/login.rs --tool claude

6. Search by Topic

Find prompts discussing specific topics:

# Search prompt content
git-ai search --pattern "authentication"
git-ai search --pattern "error handling"

# Combine with filters
git-ai search --pattern "refactor" --tool cursor

7. Pipe to Other Tools

Integrate with scripts and workflows:

# Get prompt IDs only
git-ai search --commit abc1234 --porcelain

# Count matches
git-ai search --file src/main.rs --count

# JSON for processing
git-ai search --commit abc1234 --json | jq '.prompts[] | {tool, model, author: .human_author}'

# Find commits with AI-authored code
git log --oneline | while read sha msg; do
  count=$(git-ai search --commit $sha --count 2>/dev/null || echo "0")
  if [ "$count" != "0" ]; then
    echo "$sha ($count sessions): $msg"
  fi
done

Command Reference

git-ai search

Search for AI prompt sessions by various criteria.

git-ai search [OPTIONS]

Context Source (pick one):

• --commit <rev> - Search a specific commit (or range with sha1..sha2)
• --file <path> - Search by file path
• --lines <range> - Filter to line range (requires --file, repeatable)
• --pattern <text> - Full-text search in prompt content
• --prompt-id <id> - Look up specific prompt by ID

Filters (combinable):

• --author <name> - Filter by human author (substring match)
• --tool <name> - Filter by tool (claude, cursor, etc.)
• --since <date> - Only prompts after date
• --until <date> - Only prompts before date

Output Format (pick one):

• (default) - Human-readable summary
• --json - Full JSON output
• --verbose - Detailed output with full transcripts
• --porcelain - Machine-readable prompt IDs only
• --count - Just the count of matching prompts

git-ai continue

Restore AI session context for continuation.

git-ai continue [OPTIONS]

Context Source (pick one, or none for TUI):

• --commit <rev> - Continue from a commit
• --file <path> - Continue from a file
• --lines <range> - Limit to line range (with --file)
• --prompt-id <id> - Continue specific prompt

Agent Selection:

• --agent <name> - Target agent (claude, cursor; default: claude)
• --tool <name> - Alias for --agent

Output Mode (pick one):

• (default) - Print formatted context to stdout
• --launch - Spawn agent CLI with context
• --clipboard - Copy context to clipboard
• --json - Output as structured JSON

Options:

• --max-messages <n> - Limit messages per prompt (default: 50)

git-ai show-prompt

Display the full transcript of a specific prompt.

git-ai show-prompt <prompt_id> [OPTIONS]

• --json - Output as JSON
• --raw - Show raw internal format

Output Formats

Default Output

Human-readable summary:

Found 2 AI prompt session(s) for commit abc1234

[1] Prompt a1b2c3d4 (claude / claude-sonnet-4)
    Author: Alice
    Files: src/main.rs:10-50

[2] Prompt e5f6g7h8 (cursor / gpt-4)
    Author: Bob
    Files: src/lib.rs:100-150, src/utils.rs:20-30

--json Output

Full structured data for programmatic use:

{
  "source": {
    "sha": "abc1234",
    "author": "Alice",
    "date": "2025-01-15",
    "message": "Add authentication"
  },
  "prompts": [
    {
      "id": "a1b2c3d4",
      "tool": "claude",
      "model": "claude-sonnet-4",
      "author": "Alice",
      "messages": [...]
    }
  ]
}

--verbose Output

Full conversations inline:

=== Session 1: Prompt a1b2c3d4 ===
Tool: claude (claude-sonnet-4)
Author: Alice

**User**:
Help me implement user authentication...

**Assistant**:
I'll help you implement authentication. Here's my approach...

--porcelain Output

One prompt ID per line (for scripting):

a1b2c3d4
e5f6g7h8

--count Output

Just the number:

2

Tips

Combining Filters

Filters use AND semantics - prompts must match ALL specified filters:

# Claude prompts by Alice in the last 7 days
git-ai search --commit HEAD~10..HEAD --tool claude --author "Alice" --since 7d

Using with Git Commands

Combine with git log to explore history:

# Find commits with high AI involvement
git log --oneline --since="1 week ago" | while read sha msg; do
  git-ai search --commit $sha --count 2>/dev/null
done

# Search AI context for files changed in a commit
git diff-tree --no-commit-id --name-only -r abc1234 | while read file; do
  git-ai search --file "$file"
done

Restoring Context for Different Agents

The --agent flag controls context formatting:

# For Claude Code
git-ai continue --commit abc1234 --agent claude --launch

# For other tools, copy to clipboard
git-ai continue --commit abc1234 --clipboard

Inspecting Specific Prompts

If search returns many results, drill down:

# Get the IDs
git-ai search --commit abc1234 --porcelain

# View a specific one
git-ai show-prompt a1b2c3d4

Performance Tips

• Use --commit for fastest results (direct note lookup)
• --file requires blame computation (slower for large files)
• --pattern searches the database (fast substring match)
• Add --count first to check result size before verbose output

Examples

Quick checks

# Any AI in this commit?
git-ai search --commit HEAD --count

# What tools were used?
git-ai search --commit abc1234 --json | jq '.prompts[].tool' | sort -u

# Who used AI in this file?
git-ai search --file src/main.rs --json | jq '.prompts[].author' | sort -u

Detailed exploration

# Full context of the most recent AI session
git-ai search --commit HEAD --verbose | head -100

# Continue the exact session that wrote these lines
git-ai continue --file src/main.rs --lines 50-100 --launch

Integration with CI/CD

# In a PR check, report AI involvement
PROMPT_COUNT=$(git-ai search --commit $PR_HEAD --count 2>/dev/null || echo "0")
if [ "$PROMPT_COUNT" -gt 0 ]; then
  echo "This PR includes $PROMPT_COUNT AI-assisted session(s)"
  git-ai search --commit $PR_HEAD
fi