一键导入
improve
Enhance existing features through story-driven development with full verification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Enhance existing features through story-driven development with full verification.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | improve |
| description | Enhance existing features through story-driven development with full verification. |
| argument-hint | [description or story-id] |
| context | fork |
/improve "add confidence scores to extraction"
/improve E2-S3
Provide either a plain description of the improvement or an existing story ID from specs/stories/. Improvements change observable behavior and require a story with acceptance criteria.
Improvements intentionally change behavior. This distinguishes them from /refactor, which must not change behavior. Every improvement is traceable to a story, has acceptance criteria, and requires tests to be updated or added.
Every improvement must have a story file in specs/stories/ before implementation begins.
If a story ID was provided (e.g., E2-S3): read specs/stories/E2-S3.md and confirm it has acceptance criteria.
If a description was provided: check whether a matching story already exists. If not, create specs/stories/{next-id}.md with:
Do not proceed to implementation until acceptance criteria are written and confirmed.
Read the current codebase to understand what is affected:
Document this assessment before writing any code.
Read specs/design/ for any architecture decisions relevant to the change. Read .claude/skills/architecture/SKILL.md for layering rules.
Confirm the planned implementation stays within the correct layer. If the improvement requires a new type, add it to the types/ layer. If it requires a new DB query, add it to repository/. Do not shortcut layers.
Modify the existing implementation files. Do not create parallel implementations alongside existing ones.
Rules:
_v2 function alongside the original.For each acceptance criterion in the story:
Changing a test to pass rather than fixing the underlying code is not acceptable. The test is the specification. If the test fails because the implementation is wrong, fix the implementation.
Run the full test suite. All tests must pass.
Spawn the code-reviewer agent on the full diff.
Findings:
Maximum 3 retry cycles for BLOCK findings. If BLOCK findings remain after 3 cycles, stop and report.
Add an implementation status section to the story file:
## Implementation Status
Status: COMPLETE
Implemented: {date}
Files changed: {list of files}
Tests added/updated: {list of test files}
AC coverage:
- AC1: covered by test {test name}
- AC2: covered by test {test name}
| Dimension | /improve | /refactor |
|---|---|---|
| Behavior change | Yes — intentional | No — must be zero |
| Requires story | Yes | No |
| Tests | Updated to match new behavior | Must pass unchanged |
| API contracts | May change | Must not change |
If you are not changing observable behavior, use /refactor instead.
| Artefact | Purpose |
|---|---|
specs/stories/{id}.md | Story with AC and implementation status |
| Modified source files | Implementation of the improvement |
| Updated/added tests | Verification of each acceptance criterion |
get_extraction_v2() function alongside get_extraction() creates dead code and confusion. Modify the existing function and update its callers.Autonomous build loop with Karpathy ratcheting, GAN evaluator, and session chaining. Iterates story groups until all features pass or stopping criteria met.
Socratic interview to create a Business Requirements Document. First step in the SDLC pipeline.
Generate system architecture, machine-readable schemas, and UI mockups. Spawns planner + ui-designer concurrently.
Evaluation patterns — sprint contract format, three-layer verification, scoring rubric references.
Standard GitHub issue workflow. Branch, reproduce, fix, test, PR.
Generate production code and tests for a story group using agent teams for parallel execution.