Engineering Skills
Engineering skills teach Codex how to perform repeatable engineering work. They are procedural: they describe how to inspect, plan, edit, verify, and document a class of task.
flowchart TD
task["Engineering task"] --> skill["Engineering SKILL.md"]
skill --> references["Required references"]
skill --> scripts["Optional helper scripts"]
skill --> checks["Verification checklist"]
checks --> output["Code, docs, or review output"]
Design Principles
Engineering skills should:
- be task-specific
- load only the references needed for the task
- include deterministic checks where possible
- tell Codex what artifacts to create or update
- include docs parity expectations
- avoid embedding volatile implementation knowledge
Candidate Engineering Skills
| Skill | Purpose | Primary Output |
|---|---|---|
large-change |
Run the large change workflow inside a repository. | Change folder, plan, implementation log |
repository-bootstrap |
Bring a repository into handbook compliance. | AGENTS.md, workflow, docs entry points |
mkdocs-update |
Add or reorganize handbook pages safely. | Docs pages, nav updates, build verification |
docs-parity-review |
Check whether code or process changes require docs updates. | Docs impact report or patches |
testing-verification |
Select and run appropriate verification for a change. | Verification plan and command results |
code-review |
Review changes for bugs, regressions, missing tests, and docs gaps. | Findings-first review |
release-prep |
Prepare release notes and completion checks. | Release checklist and notes |
Skill Anatomy
An engineering skill should usually include:
- trigger conditions
- required inputs
- references to read
- artifact rules
- verification rules
- docs parity rules
- completion criteria
Engineering Skill Lifecycle
stateDiagram-v2
[*] --> Proposed
Proposed --> Drafted
Drafted --> Trial
Trial --> Active
Active --> Revised
Revised --> Active
Active --> Deprecated
Deprecated --> [*]
When Not To Use A Skill
Do not create a skill for:
- one-off decisions
- rapidly changing implementation details
- content that belongs in
AGENTS.md - long-form architecture knowledge
- prompts that are easier to keep as prompt-library entries
Phase 2 Rule
Phase 2 defines skill architecture and candidates. Actual reusable skills may remain in the handbook until anuva-shared-skills is created.