Memory Enhancement

Memory Enhancement

Manage citations, verify code references, and track confidence scores for Serena memories. Ensures memories stay accurate by linking them to specific code locations and detecting when those locations change.

Triggers

• "add citation to memory" - Link memory to specific code location
• "verify memory citations" - Check if code references are still valid
• "check memory health" - Generate staleness report across all memories
• "update memory confidence" - Recalculate trust score based on verification

Quick Reference

Decision Tree

Need memory enhancement?
│
├─ Add citation to memory → add-citation command
├─ Verify citations → verify or verify-all command
├─ Check memory health → health command
├─ Traverse memory graph → graph command
└─ Update confidence → update-confidence command

Process

Phase 1: Identify Target Memory

Locate memory file by ID or path:

1. Check default directory - Look in .serena/memories/ for <memory-id>.md
2. Try direct path - If full path provided, use it directly
3. Validate existence - Error if memory file not found (exit code 2)

Verification: Memory file exists and is readable

Phase 2: Add/Verify Citations

Use CLI commands with structured output:

Add Citation

python -m memory_enhancement add-citation <memory-id> --file <path> --line <num> --snippet <text>

Parameters:

• memory-id - Memory identifier or file path
• --file - Relative file path from repository root (required)
• --line - Line number (1-indexed, optional for file-level citations)
• --snippet - Code snippet for fuzzy matching (optional)
• --dry-run - Preview changes without writing (optional)

Exit Codes (ADR-035):

• 0: Success
• 1: Validation failed (stale citation)
• 2: Invalid arguments or file not found
• 3: File I/O error

Verify Citations

# Single memory
python -m memory_enhancement verify <memory-id> [--json]

# All memories
python -m memory_enhancement verify-all [--dir .serena/memories] [--json]

Output Indicators:

• ✅ VALID - All citations point to valid locations
• ❌ STALE - Some citations are invalid
• Confidence score (0.0-1.0)
• Detailed mismatch reasons

Verification: Citations validated against current codebase state

Phase 3: Update Confidence

Recalculate based on verification results:

python -m memory_enhancement update-confidence <memory-id>

Confidence Calculation:

confidence = valid_citations / total_citations

Interpretation:

Verification: Confidence score updated in YAML frontmatter

Phase 4: Report Results

Display summary with actionable recommendations:

List Citations

python -m memory_enhancement list-citations <memory-id> [--json]

Human-readable output:

Citations for memory-001:
Total: 3

✅ src/api.py:42
   Snippet: handleError

❌ src/client.ts:100
   Reason: Line 100 exceeds file length (95 lines)

✅ scripts/test.py

JSON output for programmatic usage:

{
  "citations": [
    {
      "path": "src/api.py",
      "line": 42,
      "snippet": "handleError",
      "valid": true,
      "mismatch_reason": null,
      "verified": "2026-01-24T14:30:00"
    }
  ]
}

Health Report

python -m memory_enhancement health [--format markdown|json] [--include-graph]

Generates comprehensive report with:

• Total memories with citations
• Stale memory count and percentage
• Memories ranked by staleness (worst first)
• Optional: Orphaned memories (disconnected from graph)

Verification: Report generated successfully

Script Reference

Anti-Patterns

Integration with Existing Skills

• reflect - Auto-capture citations from learnings that reference code
• memory - Verify citations during memory search
• curating-memories - Update citations when memories change
• qa - Run verification as part of test strategy

CI Integration (Optional)

Create .github/workflows/memory-validation.yml:

name: Memory Citation Validation

on:
  pull_request:
    paths:
      - '.serena/memories/**'
      - 'src/**'
      - 'scripts/**'

jobs:
  verify:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
      - uses: actions/setup-python@40c6b50cc6aa807e2d020b243100c016221d604c # v5.3.0
        with:
          python-version: '3.12'
      - run: pip install -e .
      - run: python -m memory_enhancement verify-all --json > results.json
        continue-on-error: true
      - run: cat results.json
      - name: Comment on PR
        if: failure()
        uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
        with:
          script: |
            const results = require('./results.json');
            const stale = results.filter(r => !r.valid);
            if (stale.length > 0) {
              await github.rest.issues.createComment({
                issue_number: context.issue.number,
                owner: context.repo.owner,
                repo: context.repo.name,
                body: `⚠️ ${stale.length} stale memory citation(s) detected. Run \`python -m memory_enhancement verify-all\` locally for details.`
              });
            }

Initially set continue-on-error: true (warning only). After adoption, make blocking.

Verification

After using this skill:

• Citations validated against current codebase
• Confidence scores updated in memory frontmatter
• Stale memories identified and reported
• Health report generated (if requested)
• Exit codes follow ADR-035 standard

References

• examples.md - Usage examples and workflows
• confidence-scoring.md - How confidence is calculated
• ADR-007 - Memory-first architecture
• ADR-037 - Reflexion memory schema (if exists)