Grades scaffolded projects by comparing TODO markers against implementations and running smoke tests.
npx skills add https://github.com/unnxt30/skills --skill grade-scaffold-unnxt30
Please help me install this Agent Skill. Skill: grade-scaffold Source: https://github.com/unnxt30/skills/tree/main/grade-scaffold Command: npx skills add https://github.com/unnxt30/skills --skill grade-scaffold-unnxt30
# TODO(<owner>): marker, checks whether the body is still a stub or real code, and grades real code against the stated intent. - Smoke test execution: Detects the project shape (FastAPI app, pytest config, settings module) and runs the appropriate smoke chain with timeouts to verify the code actually runs. - Known bug pattern detection: Flags recurring mistakes like async/sync mismatches, SQLAlchemy ORM/Core mixing, Pydantic v1 leftovers, and module-level singletons. - Use Case: You finished Phase 2 of a scaffolded FastAPI project and type /grade. The Skill reads the spec, cats each scaffolded file, runs the health-check smoke, and returns a grade card with a B+, two must-fix bugs, and a recommended next step. ## Quick Start Type /grade or ask "grade my Phase 2 work" to audit your scaffolded project and receive an honest grade card with prioritized fixes.High-intent search queries and answers about installing and using this skill.
Type /grade or ask to grade a specific phase. The skill finds all owner-tagged TODO markers, compares each implementation against the TODO's intent, runs smoke tests, and produces a grade card with a letter grade and prioritized fix list.
An owner-tagged marker like `# TODO(unnat): implement login` is a scaffold hint telling you what to implement in the body below. Bodies still containing `raise NotImplementedError` count as not done; real code gets graded against the stated intent.
Yes, it works standalone on any project using `# TODO(<owner>):` markers, even without the learning-scaffold skill. However, it produces better results when spec and plan documents exist to define what done means.
It checks for async/sync mismatches like un-awaited `db.scalars()` calls, SQLAlchemy ORM/Core mixing, Pydantic v1 leftovers such as `@validator` and `parse_obj`, missing `app.include_router` mounts, and module-level engine singletons that break testability.
The Read tool can return stale cached content if a file was modified since the last read, silently showing outdated code. Using `cat` bypasses the cache and guarantees the grader evaluates the actual current file contents.
No, it only reports findings and describes the correct pattern. In learning mode the user writes the fix themselves, preserving the educational value of implementing the correction.
insightfloGenerate a structured TASKS.md scaffold with domain-guarded tasks from codebase analysis.
SarthakMishraAnalyze `.todo/` backlog items and prioritize compiler tasks for oxc-react-compiler.
uzumaki-incRun code health checks and compute a weighted 0-10 score.
summarybotngConduct context-driven code reviews focusing on quality, testability, and maintainability.