skilly. Buy ad slot
All skills
Uncategorized / AGENT SKILL

memory-curation

coji831/agentic-devops-solar-ralph
0 installs 2 GitHub stars
0

Use when deciding whether information belongs in repo memory, session memory, the AI ledger, or permanent documentation.

Low community score

memory-curation: This skill has a community score below −5, meaning it has received more downvotes than upvotes. Review its source and assessment before installing. Community votes are not a security review.

BEFORE YOU INSTALL

Understand the trade-offs.

SECURITY REVIEW

Not yet assessed

Review the original instructions and requested permissions before installing.

No security review is available for this catalog entry yet.

SKILL QUALITY

Not yet assessed

How clearly the skill guides your agent, how complete its workflow is, and how you can check the outcome.

No quality assessment is available for this catalog entry yet.

The full skill.

Original instructions from the publisher’s SKILL.md

# Memory Curation

## When to Use

- Classifying new facts discovered during work
- Cleaning up drift between instructions and docs
- Deciding what should persist across sessions
- Writing back a verified fact discovered during a task

## Procedure

1. Put active execution state in `.github/.ai_ledger.md`.
2. Put concise verified operational facts in `.github/instructions/*.instructions.md` (scoped by `applyTo` pattern).
3. Put enduring guidance and rationale in `docs/`.
4. Avoid duplicating the same long-form content across all three layers.

## Write-Back Rule (Ask to Remember)

When you verify a **new fact** during normal work — a real command, a naming pattern, an API contract, a folder path — write it back to the relevant instruction file immediately. Do not wait for the next setup scan.

- Choose the instruction file whose `applyTo` scope covers the fact's domain:
  - Architecture / commands → `architecture.instructions.md`
  - Frontend patterns → `frontend.instructions.md`
  - Backend patterns → `backend.instructions.md`
  - Auth / secrets → `security.instructions.md`
  - Test commands → `verification.instructions.md`
  - Naming / commit rules → `conventions.instructions.md`
  - Workflow steps → `workflow.instructions.md`
- Replace the matching `[FILL IN]` or `[SCAN-INCOMPLETE]` placeholder with the verified value.
- If no placeholder exists, append the fact as a new bullet under the relevant section heading.
- Never overwrite a fact that contradicts your new finding — flag it with `// CONFLICT:` instead.
- Keep entries to one bullet per fact. No prose.

## Output

- Chosen storage location
- Reason for that choice
- Instruction file updated (if write-back occurred)
- Follow-up cleanup needed