| name | memory-verification |
| description | Use when applying facts from .github/instructions/ to confirm those facts still match the current codebase before acting on them. Prevents errors from stale architecture instructions. |
| user-invocable | false |
Validate that facts from .github/instructions/ still match the codebase before applying them. Prevents errors from outdated instruction content.
When to Use
Load this skill at the start of any task where you:
- Apply a pattern or architecture fact from
.github/instructions/
- Reference a specific file, function, or component from an instruction file
- Use a previously recorded API contract, state shape, or service boundary
Constraints
- Do not skip verification to save time. A 5-second check prevents a 30-minute repair loop.
- Do not assume memory is correct just because it was recently written.
- Do not flag a memory as stale based on superficial differences (e.g., minor rename). Verify the semantic meaning is unchanged.
Approach
For each memory claim being applied:
- Locate the reference — Find the exact file, function, or component the memory describes.
- Verify existence — Confirm the file/function/component still exists at the recorded path.
- Verify semantics — Confirm the described behavior, contract, or shape still matches the current code.
- Check for recent changes — Look for co-located test changes or updated type definitions that suggest a refactor.
- Classify the memory:
- ✅ Valid — Still accurate. Apply it.
- ⚠️ Stale — No longer matches the current code. Do not apply. Record the discrepancy.
- ❓ Uncertain — Partial match or ambiguous. Verify manually before applying.
- Update if stale — Write a correction to the relevant
.github/instructions/*.instructions.md file with current facts before proceeding.
Output Format
For each verified memory:
- Memory claim:
<what the memory states>
- Verification result: ✅ Valid / ⚠️ Stale / ❓ Uncertain
- Discrepancy (if stale):
<what has changed>
- Action taken: Applied / Skipped / Updated