| name | curator |
| description | Maintain vapid.party's repo-local memory and skill structure. Use when Codex learns a durable workflow, repeated pitfall, validation pattern, project convention, or collaborator preference; when adding, revising, consolidating, or pruning MEMORY.md, memory/, SKILLS.md, or skills/; or when adapting recurse.bot-style operating guidance for this repo. |
Curator
Overview
Keep the repo's agent guidance small, accurate, and discoverable. Turn repeated work into concise notes or reusable skills, then validate the result before landing it.
Workflow
- Read
AGENTS.md, MEMORY.md, and SKILLS.md.
- Decide where the learning belongs:
AGENTS.md: repo-wide rule or sharp edge every agent must see.
MEMORY.md: index entry for durable notes or logs.
memory/notes/: reusable project knowledge.
memory/logs/: dated record of completed work and lessons.
SKILLS.md and skills/<name>/SKILL.md: repeated procedure.
- Keep entries concise and remove stale or duplicate guidance.
- Do not record secrets, private keys, bearer tokens, API keys, or raw customer data.
- Validate changed skills with:
python3 /home/pierce/.codex/skills/.system/skill-creator/scripts/quick_validate.py skills/<name>
When Adding A Skill
- Add a skill only when the workflow is likely to repeat.
- Use lowercase hyphen-case for the folder and
name.
- Put trigger conditions in the frontmatter
description.
- Keep the body procedural; omit background that a capable coding agent already knows.
- Update
SKILLS.md in the same change.
When Updating Memory
rg -n "keyword|topic" MEMORY.md memory
- Prefer editing an existing note when the new information refines it.
- Add a dated log for completed structure/workflow passes that future agents may need to understand.