Skills
plan → build → review
Acolyte’s engineering discipline. Each skill encodes a workflow the agent can activate when the task calls for it. Bundled skills are project-agnostic and available in every workspace.
These are specialized for Acolyte from the tool-agnostic set at cniska/skills, which carries the same workflow for any agent.
Multiple skills can be active in one session. The agent can load one or more in a single call, and the active set is shown in the status line.
Skills
| Phase | Skill | Description |
|---|---|---|
| Plan | plan | Design through dialogue, slice vertically, clarify through questions |
| Build | build | Vertical slices — implement, verify, commit, repeat |
| tdd | Red-green-refactor, mock at boundaries | |
| debug | Stop the line, reproduce, fix root cause, guard with test | |
| design | Hard-to-misuse interfaces, contract first, validate at boundaries | |
| simplify | Reduce complexity, Chesterton’s Fence, preserve behavior | |
| git | Atomic commits, clean history, rewrite before pushing | |
| deprecation | Build replacement first, migrate consumers, remove completely | |
| Review | review | Run all review dimensions, severity labels, fix-all policy |
| correctness-review | Logic bugs, edge cases, broken contracts | |
| style-review | Local conventions, naming, control flow, readability | |
| architecture-review | Boundaries, indirection pressure, contract integrity | |
| test-review | Coverage gaps, edge cases, test quality | |
| security-review | Trust boundaries, execution safety, concrete attack paths only | |
| doc-review | Drift detection, terminology, outdated names | |
| Meta | agents-md | Create or update AGENTS.md project rules |
Principles
These show up across multiple skills and form the shared engineering philosophy.
| Principle | In practice | Skills |
|---|---|---|
| Vertical slices | One complete path through the stack at a time | build, plan |
| Contract first | Schema before implementation | design, build |
| SRP | One responsibility per module, one change per commit | architecture-review, build, git |
| YAGNI | Don’t build for hypothetical requirements | architecture-review, design |
| Stop the line | Something breaks — stop, don’t push past it | debug |
| Prove-It pattern | Failing test before fix | debug, tdd |
| Mock at boundaries | Mock external systems, not internal functions | tdd, test-review |
| DAMP over DRY | Descriptive tests over deduplicated tests | tdd |
| Rule of 3 | Extract after three instances, not before | simplify, style-review |
| Chesterton’s Fence | Understand before removing | simplify |
| Hyrum’s Law | All observable behavior becomes a commitment | design, deprecation |
| Code as liability | Less code serving the same purpose is better | deprecation |
| Source over memory | Verify framework behavior in primary docs before implementation | build |
| Save-point pattern | Commit early when exploring uncertain changes | git |
| Evidence threshold | Concrete references, not speculation | review skills |