---
name: omnigent-ai-feature-implement
description: "Implement the design with focused tests mapped to each covered CUJ."
license: "Apache-2.0"
metadata:
  decimalai-skill-id: "a4982994-3870-43bd-bf99-2c6bc2ee2250"
  decimalai-version-id: "d43bc1ef-4ae1-445e-9d7e-7cdde6733666"
  decimalai-version: "1"
  decimalai-body-sha256: "20383829a9c8d517351786a3d8a77daf62b990eb76065c199d7c2fe20f25fe9e"
  decimalai-name: "omnigent-ai/feature-implement"
  decimalai-format: "1"
  decimalai-invocation: "model"
  decimalai-category: "coding"
  decimalai-source: "https://app.decimal.ai/s/omnigent-ai-feature-implement@1/SKILL.md"
---
# Implement and test

1. **Implement the design.** Keep the change focused on the requirements. Follow
   the repository's conventions in `CLAUDE.md`, including framework-owned
   instructions, database query names, deprecation notes, and comment style.
2. **Test it.** Read `tests/README.md` and `tests/_helpers/README.md`, then the
   nearest existing scenario. Reuse or extend existing coverage before adding
   setup. Put input matrices at the lowest reliable layer; add E2E only for a
   distinct process, transport, or browser boundary.
   - Map every covered CUJ to at least one test that exercises the new behavior.
     There is no fail-before proof for new behavior, but each test must fail if
     the feature were removed; a test that only passes is not coverage.
   - Run the directly affected test modules and focused checks. Do not run the
     full repository suite locally; CI owns that.
   - Check environment-derived defaults with ambient variables populated, so a
     test does not pass only because of the local environment.
3. **Assess the impact.** Load `otto-impact-assessment` and apply it to the full
   diff. Record the results in the handoff's `impact_assessment` and
   `test_audit`.

If a requirement turns out to be infeasible, stop building it, mark its CUJs not
covered with the reason, and continue with the rest. Then load `feature-demo`.