| name | curator |
| description | Maintain FrogAlert's repo-local skills and operating knowledge. Use when a repeated workflow or lesson may belong in AGENTS.md, MEMORY.md, SKILLS.md, agent-memory, or a new or revised repo-local skill package. |
Curator
Keep the knowledge system small, current, and useful to future agents.
Workflow
- Read
AGENTS.md, MEMORY.md, SKILLS.md, and the candidate lesson.
- Classify it:
- stable project rule or safety invariant →
AGENTS.md;
- durable observation or dated outcome →
agent-memory/ plus MEMORY.md;
- repeatable multi-step procedure → an existing or new skill;
- product requirement/readiness change →
FEATURES.md;
- one-session trivia → do not retain it.
- Prefer updating or consolidating an existing skill over adding another.
- Keep skill bodies imperative, focused, and below 500 lines. Include only
non-obvious procedure and validation gates.
- Keep
SKILLS.md metadata aligned with every live skill package.
- Validate each changed skill with the skill-creator
quick_validate.py.
- Run
./scripts/verify, record meaningful outcomes, and commit the cohesive
knowledge update.
Guardrails
- Never store secrets, private profiles, raw BLE observations, or irrelevant
personal details in this public repository.
- Do not turn every successful command into a skill.
- Remove stale instructions rather than layering contradictory advice.
- Preserve
AGENTS.md as canonical; harness aliases remain symlinks.