| name | ss14-ai-workflow |
| description | Preflight, implementation, and self-review workflow for AI agents working on Shine Project SS14 code. Use when planning or making any repository change, reviewing PRs, hardening instructions, or deciding which .agents skills/rules apply. |
SS14 AI Workflow
Use this skill to keep agent work boring, narrow, and reviewable.
Workflow
- Load
shine-project.
- Read subtree
AGENTS.md files in every edited area.
- Search for existing systems, components, prototypes, localization IDs, tests, and configs.
- Choose the smallest domain skills for the change.
- Implement through existing public system APIs where possible.
- Add new APIs only when they remove real duplication or make future extension safer.
- Validate, then self-review with
.agents/rules/ss14-review-checklist.md.
Bundled References
references/ai-strengths-and-risks.md: where agents help and where they usually create review noise.
references/prompting-patterns.md: how bridge files should route requests into canon without duplicating policy.
references/verification-rules.md: validation and self-review loop for code, resources, docs, and bot config.
Anti-Patterns To Stop
- Writing code before checking current local patterns.
- Copying old WizDen examples without updating to this repo and .NET 10.
- Touching upstream files for convenience.
- Adding hardcoded prototype or localization IDs in systems.
- Making server-only interactive mechanics when shared prediction is feasible.
- Adding a second implementation of an existing mechanic instead of extending it.
CodeRabbit And Review Bots
When editing bot instructions:
- Keep
.agents/rules as the canon.
- Adapters should route to canon, not duplicate long policy.
- Path-specific instructions should be concrete and low-noise.
- Treat assertive review profiles as something to tune after real PRs.