| name | harness-intake |
| description | Use when a user gives a new product, coding, review, debugging, or planning task in this Codex harness. Classifies tier, decides whether subagents/native features are useful, and fixes the initial contract before work starts. |
Harness Intake
Use this skill at the start of non-trivial work.
Steps
- Restate the goal, constraints, and done-when in one short block.
- Classify the task:
trivial: small, low-risk, one obvious path.normal: meaningful behavior or product change.high-risk: auth, payments, security, migrations, infra, broad refactor, legal/compliance, destructive actions, or unclear blast radius.
- Decide whether native features are useful:
- subagents for bounded parallel planning, research, code mapping, review, or disjoint work.
- worktree for broad/risky implementation.
- browser for current docs or UI QA.
- terminal for deterministic checks.
- automation for waits, retros, stale review, or recurring audits.
- For
normal+, define required artifacts:
- goal and non-goals
- acceptance criteria
- validation plan
- rollback or recovery plan
- review owner
- Do not ask the user to run commands unless blocked. Run harness checks yourself when appropriate.
Output Contract
Return:
- tier
- planned approach
- subagents/native features to use or skip
- key risks
- first action
