| name | agent-loop |
| description | Use when designing AI Agent product capabilities, competition judging evidence, or production agent safety. Creates judgment → action → verification/improvement loops with retrieval, rule checks, approval, logging, and feedback. |
| argument-hint | [agent-capability] |
AI Agent Loop Design
Use when
- Need to show AI Agent, not chatbot
- Need judgment/action/verification loop
- Need reviewer approval/audit/feedback design
Do not use when
- Feature contains no AI judgment or action selection
Inputs to inspect
- Flow
- Feature spec
- Architecture
- Risk rules
- Eval plan
Files to create or update
docs/09_flow.md- Agent sections in
docs/08_feature-spec.md
docs/10_eval-plan.mdrules/agent-rules.md
Operating sequence
- Identify each agent capability.
- Define trigger and inputs.
- Define context gathering/retrieval.
- Define judgment output and confidence.
- Define possible actions and forbidden actions.
- Define verification checks.
- Define approval/escalation.
- Define audit events and feedback.
- Add golden cases.
Agent Capability Card
### Agent Capability: <Name>
| Field | Content |
|---|---|
| Trigger | |
| Inputs | |
| Context Sources | |
| Judgment | |
| Possible Actions | |
| Forbidden Actions | |
| Verification | |
| Approval Needed? | |
| Audit Events | |
| Failure Modes | |
| Eval Cases | |
Judgment Output Contract
An agent judgment must include:
- classification/status/risk/priority,
- evidence used,
- uncertainty or confidence,
- recommended action,
- reason not to act if blocked.
Quality gates
- Generation without verification is incomplete.
- High-risk action without human approval is blocked.
- Every loop has eval cases.
Stop conditions
Stop and route backward when prerequisite artifacts are missing, when the requested action would erase user work, or when a high-risk workflow lacks evidence, approval, audit, or fail-closed handling.
Example invocation
/harness-engineering:agent-loop
Apply this skill to the current repository. Inspect existing artifacts first, update only what is necessary, and report files changed, decisions, assumptions, validation, risks, and next skill.
Shared completion and integrity
End with the completion report in ../../references/common-completion-report.md and follow ../../references/common-integrity-rules.md.
