Memory and Caching Features¶

Overview¶

The AI Workflow Failure Summary Action now includes memory and caching capabilities to provide context-aware analysis by understanding:

• Past failed runs on the same branch
• Recent commit changes
• Previous AI summaries
• Historical failure patterns

Architecture¶

graph TB
    subgraph "Current Run"
        A[Workflow Fails]
        B[Fetch Current Logs]
    end

    subgraph "Memory Layer"
        C[Load Cache]
        D[Past Summaries]
        E[Branch History]
        F[Commit Changes]
    end

    subgraph "Analysis"
        G[Combine Context]
        H[AI Analysis]
        I[Generate Summary]
    end

    subgraph "Storage"
        J[Save to Cache]
        K[Update History]
    end

    A --> B
    B --> C
    C --> D
    C --> E
    C --> F
    D --> G
    E --> G
    F --> G
    B --> G
    G --> H
    H --> I
    I --> J
    J --> K

Cache Storage Options¶

1. GitHub Actions Cache (Recommended)¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    openai-api-key: ${{ secrets.OPENAI_API_KEY }}
    enable-memory: 'true'
    cache-strategy: 'actions-cache'

Pros: - Built into GitHub Actions - No external dependencies - Automatic cleanup - Cross-run persistence

Cons: - 10GB limit per repository - 7-day retention

2. GitHub Issues/Comments¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    openai-api-key: ${{ secrets.OPENAI_API_KEY }}
    enable-memory: 'true'
    cache-strategy: 'github-issues'
    memory-issue-label: 'ai-memory'

Pros: - Permanent storage - Easy to view and audit - Integrates with GitHub UI

Cons: - Creates issues/comments - Requires issue write permissions - Not as fast as cache

3. Git Notes¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    openai-api-key: ${{ secrets.OPENAI_API_KEY }}
    enable-memory: 'true'
    cache-strategy: 'git-notes'

Pros: - Tied to commits - Part of git history - No extra storage needed

Cons: - Requires push access - Not visible in GitHub UI - Complex to query

Configuration¶

Enable Memory¶

enable-memory: 'true'

Enables all memory features.

Cache Strategy¶

cache-strategy: 'actions-cache' # or 'github-issues' or 'git-notes'

Choose how to store historical data.

Memory Retention¶

memory-retention-days: '30'

How long to keep historical data (default: 30 days).

Max Historical Runs¶

max-historical-runs: '10'

Maximum number of past runs to include in analysis (default: 10).

Include Commit Changes¶

include-commit-changes: 'true'

Include git diff of recent commits in analysis.

Memory Scope¶

memory-scope: 'branch' # or 'repository' or 'workflow'

Scope of memory to consider: - branch: Only this branch's history - repository: All branches - workflow: Only this specific workflow

What Gets Cached¶

1. Failure Summaries¶

{
  "runId": 12345,
  "branch": "feature/fix-tests",
  "commit": "abc123",
  "timestamp": "2024-12-10T18:00:00Z",
  "summary": "Test failed due to...",
  "failedJobs": ["test", "lint"],
  "resolved": false
}

2. Branch History¶

{
  "branch": "feature/fix-tests",
  "lastAnalysis": "2024-12-10T18:00:00Z",
  "totalFailures": 3,
  "commonErrors": [
    "DATABASE_URL not set",
    "Timeout in integration tests"
  ],
  "resolution": "Added DATABASE_URL to workflow"
}

3. Commit Changes¶

{
  "commit": "abc123",
  "files": ["src/api.ts", "test/api.test.ts"],
  "additions": 50,
  "deletions": 20,
  "relevantTo": ["test job"]
}

Enhanced Analysis Prompt¶

With memory enabled, the AI receives additional context:

## Current Failure
[Current logs...]

## Historical Context

### Previous Failures on This Branch
1. 2 days ago: Test failed due to DATABASE_URL not set
   - Resolution: Added to workflow environment

2. 1 day ago: Build timeout
   - Resolution: Increased timeout from 10m to 15m

### Recent Commits
- abc123: "Fix database connection pooling" (2 files changed)
- def456: "Update test fixtures" (5 files changed)

### Patterns Detected
- This is the 3rd failure in the test job this week
- Similar error occurred 2 days ago
- Previous fix did not resolve the issue

## Analysis Request
Considering the history above, analyze the current failure and:
1. Identify if this is a recurring issue
2. Reference previous failures and resolutions
3. Determine if previous fixes were effective
4. Suggest improvements based on patterns

Usage Examples¶

Basic Memory Usage¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    openai-api-key: ${{ secrets.OPENAI_API_KEY }}
    enable-memory: 'true'

Advanced Configuration¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    openai-api-key: ${{ secrets.OPENAI_API_KEY }}
    enable-memory: 'true'
    cache-strategy: 'actions-cache'
    memory-scope: 'branch'
    max-historical-runs: '20'
    memory-retention-days: '60'
    include-commit-changes: 'true'

With Custom Prompts¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    openai-api-key: ${{ secrets.OPENAI_API_KEY }}
    enable-memory: 'true'
    custom-system-prompt: '.github/prompts/system-with-memory.md'

Example prompt:

You are analyzing a workflow failure with access to historical context.

When analyzing:
1. Check if this error has occurred before
2. Review what previous fixes were attempted
3. Determine if patterns suggest a deeper issue
4. Provide more informed recommendations based on history

If this is a recurring issue, explain why previous fixes failed
and suggest a more permanent solution.

API for Memory Access¶

The action exposes memory data as outputs:

- id: analyze
  uses: ianlintner/ai_summary_action@v1
  with:
    enable-memory: 'true'
    # ...

- name: Access Memory Data
  run: |
    echo "Previous failures: ${{ steps.analyze.outputs.historical-failures }}"
    echo "Branch patterns: ${{ steps.analyze.outputs.branch-patterns }}"
    echo "Similar issues: ${{ steps.analyze.outputs.similar-issues }}"

Benefits¶

1. Smarter Analysis¶

With historical context, the AI can: - Identify recurring issues - Suggest better fixes based on what didn't work - Detect patterns across multiple failures - Provide branch-specific insights

2. Reduced Repeated Failures¶

By understanding what was tried before: - Avoid suggesting the same failed fix - Recognize when a different approach is needed - Learn from successful resolutions

3. Better Developer Experience¶

Developers get: - More relevant recommendations - Context about failure history - Insight into patterns - Confidence that issues won't recur

Performance Considerations¶

Cache Size¶

Each analysis typically stores: - Summary: ~2-5 KB - Commit changes: ~1-10 KB - Metadata: ~1 KB

With 10 historical runs: ~50-150 KB total per branch

Lookup Time¶

• Actions Cache: ~1-2 seconds
• GitHub Issues: ~2-5 seconds
• Git Notes: ~1-3 seconds

Memory Usage¶

Minimal impact on action runtime: - Loading cache: +1-3 seconds - Processing history: +0.5-1 second - Saving cache: +1-2 seconds

Total overhead: ~3-6 seconds per run

Privacy and Security¶

What's Stored¶

• Anonymized log summaries (not full logs)
• File paths and commit SHAs
• Error messages (sanitized)
• Job names and conclusions

What's NOT Stored¶

• API keys or secrets
• Full log contents
• Sensitive environment variables
• Private code snippets

Data Retention¶

• Automatic cleanup after retention period
• Manual clear via workflow dispatch
• Respects GitHub's data policies

Clearing Memory¶

Manual Clear¶

- uses: ianlintner/ai_summary_action@v1
  with:
    github-token: ${{ secrets.GITHUB_TOKEN }}
    memory-action: 'clear'
    memory-scope: 'branch'  # or 'repository' or 'workflow'

Automatic Cleanup¶

Memory is automatically cleaned: - After retention period expires - When branch is deleted (branch-scoped) - When cache size limit is reached (LRU eviction)

Troubleshooting¶

Memory Not Loading¶

Check: - enable-memory: 'true' is set - Cache permissions are correct - Cache exists (first run won't have memory)

Out of Date Memory¶

Clear and rebuild:

# Via workflow
gh workflow run clear-memory.yml

# Or disable then re-enable
enable-memory: 'false'  # One run
enable-memory: 'true'   # Next run

Performance Issues¶

Reduce historical data:

max-historical-runs: '5'  # Default is 10
memory-retention-days: '14'  # Default is 30

Future Enhancements¶

Planned features: - Machine learning for pattern detection - Cross-repository insights - Predictive failure analysis - Automatic resolution suggestions based on history - Integration with GitHub Copilot Workspace

Related Documentation¶

• Configuration Reference
• Architecture
• Custom Prompts