| name | skill-creator |
| description | Create, update, review, and validate Codex/OpenAI agent skills in this repository. Use when the user asks to add a new skill, revise an existing SKILL.md, design skill references/scripts/assets, audit trigger descriptions, add smoke tests, package skills under inst/skills, or align skills with UKBAnalytica/RAP privacy and reproducibility standards. |
Skill Creator
Purpose
Use this skill to create and maintain repo-bundled Codex skills. Keep every
skill concise, triggerable, testable, and safe for UK Biobank RAP workflows.
Workflow
- Inspect the existing skill tree before editing:
find inst/skills -maxdepth 3 -type f | sort
- Decide the target:
- New standalone skill:
inst/skills/<skill-name>/SKILL.md.
- New UKBAnalytica workflow skill:
inst/skills/UKBAnalytica_skills/<ukbsci-name>/SKILL.md.
- Existing skill update: patch only the relevant skill and directly referenced resources.
- Write
SKILL.md first. The frontmatter must contain only:
---
name: lower-hyphen-name
description: Clear trigger description with what the skill does and when to use it.
---
-
Keep the body procedural and short. Put detailed examples, disease catalogs,
function lists, or long checklists in one-hop references/ files.
-
Add resources only when they materially improve reuse:
references/: detailed guidance loaded only when needed.
scripts/: deterministic checks or generation utilities.
assets/: templates or media used in outputs.
agents/openai.yaml: UI metadata. Its default_prompt must mention $<skill-name>.
- Validate after edits:
python3 inst/skills/skill-creator/scripts/validate_skill.py inst/skills/<skill-name>
For all repo skills:
python3 inst/skills/skill-creator/scripts/validate_skill.py inst/skills
UKBAnalytica Standards
- Preserve the strict RAP privacy boundary: local agents generate scripts and
interpret aggregate outputs only. They must not read, inspect, summarize, or
process real UK Biobank participant-level RAP data or logs containing row
values.
- Allow script generation, package guidance, simulated-data smoke tests, and
interpretation of aggregate tables/figures.
- Prefer package functions over ad hoc code in examples.
- Remind users to install or load the latest UKBAnalytica before high-level
workflow examples.
- Add minimal smoke tests when a skill depends on package functions whose
interfaces may change.
Reference Files
Read only the needed reference:
references/skill-structure.md: SKILL.md, resource layout, and UI metadata.
references/ukbanalytica-skill-standards.md: UKBAnalytica-specific privacy,
function, and analysis workflow standards.
references/evaluation-and-smoke-tests.md: validation and smoke-test design.
Completion Checklist
SKILL.md has valid frontmatter and no placeholder text.
description contains the trigger conditions.
- Long details are moved into direct
references/ files.
agents/openai.yaml, if present, matches the skill and mentions $skill-name.
- Validation script passes.
- For package-facing skills, a small smoke test exists or the reason for
omitting one is clear.