Implement an approved spec or ticket with vertical slices, repository-native verification, and review gates.
Install
npx skillscat add sourabhjainr/auren/implement Install via the SkillsCat registry.
Implement
Implementation is controlled execution, not planning by trial and error.
Inputs
Start from an approved request, spec, or agent-ready ticket. Read repository instructions, acceptance criteria, dependencies, and impacted symbols. Acquire fresh ContextEvidence before changing code.
Execution
- Select the smallest ready work unit.
- Identify a public behavioral seam.
- Use
tddfor the first slice when test infrastructure can observe the behavior. - Implement the minimum change.
- Verify immediately and record evidence.
- Repeat only for the next bounded slice.
- Run the full relevant regression set.
- Run
code-reviewbefore considering the artifact ready.
For repeated repair/research/test passes, use the canonical bounded loop implementation. Its LoopRunReceipt and provenance events are part of the same evidence chain as verification and rollout; do not build a second loop engine here.
Dependencies
Do not start blocked work. Preserve blocking edges in the existing task/decomposition model. Mutating work must respect resource conflicts and the host's permission boundaries.
Completion
A unit is complete only when acceptance checks pass, regression evidence exists, material review findings are resolved, and the artifact can proceed through the existing rollout lifecycle. State remaining uncertainty explicitly.