DeepChat Spec-Driven Development Skill
Create durable design specs and implementation plans before substantial DeepChat code, configuration, or architecture changes.
Evidence shows the skill is a documentation process without sensitive data collection or network exfiltration. GitHub issue sync requires explicit approval, demonstrating user confirmation. However, no explicit least-privilege details, rollback, or data-flow transparency, hence deduction.
Instructions are clear and self-consistent, but static review cannot verify execution. Related tests and workflows exist but do not cover the skill's key paths, hence deduction.
Clear scenarios and boundaries, supports Chinese, but no network reachability mention, and trigger depends on developer judgment, hence deduction.
Clear structure with when-to-use, classification, required artifacts, workflow, but lacks versioning, changelog, and known-limitation disclosure, hence deduction.
Core goal (produce spec) clear, but no output examples or verification, static review cannot confirm effectiveness, hence deduction.
CI workflows and tests exist but do not cover the skill itself, and no third-party execution evidence, static review limits confirmation, hence deduction.
- The skill involves no sensitive operations, but users should confirm file paths and GitHub issue sync as expected.
- May be limited in no-network or restricted-network environments; verify beforehand.
- Static review did not verify execution; test in a development environment before use.
What does this skill do, and when should you use it?
This skill guides the creation of specification documents, implementation plans, and task lists before diving into code for substantial changes in the DeepChat project that need shared context or a durable decision record. It classifies the goal into feature, complex bug, or architecture work, and creates folders under docs/features, docs/issues, or docs/architecture respectively. The skill encourages documentation discipline, avoids over-engineering by default, and only syncs GitHub issues when explicitly requested by the developer. It aligns with existing DeepChat patterns such as main-process Presenter boundaries, typed shared/contracts, and Vue 3 Composition API.
The skill inspects current code and docs before execution, decides whether work is substantial enough to warrant SDD, and classifies the goal into one of three documentation folders: docs/features for new capabilities, docs/issues for complex bugs, or docs/architecture for refactors or architecture. For feature and architecture goals, it creates spec.md, plan.md, and tasks.md; for complex bugs, only spec.md. It requires resolving all [NEEDS CLARIFICATION] markers before implementation, breaks work into commit-sized tasks, and optionally links GitHub issues. It updates task checklists as work lands and runs pnpm run format, pnpm run i18n, and pnpm run lint before handoff when relevant.
- A contributor planning a new DeepChat feature, such as adding a new model provider, who wants a shared plan to coordinate with other developers.
- A developer facing a complex bug that is hard to root-cause, who needs to document the investigation and fix for future maintainers.
- A tech lead performing a refactor or dependency boundary adjustment who wants to record architecture decisions before touching code.
- An engineer breaking down a large change into reviewable sub-tasks with clear documentation for each step.
What are this skill's strengths and limitations?
- Enforces writing specs, plans, and tasks before coding, reducing rework.
- Helps prioritize work by categorizing goals into feature, bug, or architecture.
- Avoids unnecessary documentation by default, preventing over-engineering.
- Aligns with DeepChat-specific code patterns (Presenter, contracts, Vue 3).
- Limited to DeepChat projects; less useful for unrelated codebases.
- May feel heavy for simple bugs that require immediate fixing (though the skill advises skipping).
- Written documentation can become stale quickly if not maintained.
- GitHub sync is off by default, requiring explicit developer request.
How do you install this skill?
This skill is part of the DeepChat repository, located at .agents/skills/deepchat-sdd/. To install, clone or download the entire repository and place the .agents/skills/deepchat-sdd folder into your project's supported skills directory.
How do you use this skill?
Before starting a substantial DeepChat code or documentation change, describe your goal to the assistant. The skill will inspect existing patterns, classify the work, and create the necessary spec.md, plan.md, or tasks.md files in the appropriate directory (e.g., docs/features/my-feature/). After implementation, run formatting and linting commands as required by the skill.