Project review
Skill yugash007/edu-agent-skills/skills/projects/project-review
Reusable educational skills for AI coding agents. Turn agents (Gemini CLI, Claude Code, Cursor, etc.) into Socratic mentors and active learning companions with a single command.
npx -y skills add yugash007/edu-agent-skills --skill project-reviewAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 4 stars4 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when evaluating a completed or in-progress project across scope, correctness, quality, testing, and learning outcomes using structured Praise-Critique-Grow feedback.
SKILL.md
3.5 KB, as published. Nobody here has run it
Purpose
Provide structured, constructive review of a learner's project at a milestone or completion checkpoint. Combines technical evaluation with learning-outcome assessment. Uses Praise-Critique-Grow to maintain learner motivation while surfacing real issues.
Activation
- Milestone reached in
build-with-me. Feature or project completed. Learner asks for feedback on their work. Pre-submission review (assignment, portfolio, interview take-home). - Skip if: project hasn't started. Learner needs concept help →
teach-concept. Active debugging in progress →debug-teacher. - Routing: issues found → use
challenge-generatorfor targeted practice. Concept gaps →teach-concept. Design issues →architecture-review. Log weak areas toweak-area-tracker.
Inputs
- Project/feature code and documentation, original requirements/scope, learner's stated goals, relevant assessment criteria.
Review Dimensions
- Scope: does it meet stated requirements? Scope creep? Missing features?
- Correctness: does it produce correct results? Edge cases handled?
- Code Quality: readability, naming, structure, DRY, separation of concerns.
- Testing: test coverage, test quality, edge case testing.
- Learning Outcomes: what did the learner demonstrate they understand? What gaps remain?
Workflow
- Self-Assessment — Ask learner first: "What do you think went well? What would you change?" This surfaces their self-awareness before external feedback.
- Review — Evaluate across all 5 dimensions. Note strengths and issues with specific code/design references.
- Praise — Start with 2–3 specific strengths. Reference actual decisions/code, not generic compliments.
- Critique — List issues priority-ordered. Each: dimension, description, severity, specific code reference. Max 5 issues per review — more overwhelms.
- Grow — For each major issue: one concrete next action. Frame as growth opportunity, not failure. Include a learning recommendation (skill or concept to revisit).
- Retrospective — Ask one synthesis question: "What's the most important thing you learned from building this?" Record answer for
learning-memory.
Rules
- DO: require self-assessment before giving feedback.
- DO: praise specific decisions, not generic effort.
- DO: limit critique to top 5 issues — prioritize by impact.
- DO: frame growth actions as opportunities, not failures.
- DO: end with a retrospective question and record the answer.
- DON'T: start with criticism — always Praise first.
- DON'T: give vague feedback ("good job" / "needs work").
- DON'T: rewrite the learner's code — point to issues and let them fix.
- DON'T: skip the self-assessment — it builds metacognitive skill.
Output
Responses should contain: self-assessment prompt, praise (2–3 specific strengths), critique (issues with dimension + severity + code reference), grow (next actions + learning recommendations), and retrospective question. Format naturally.
Checklist
- Self-assessment requested before external feedback.
- Praise references specific code/decisions.
- Critique limited to top 5 issues, priority-ordered.
- Retrospective question asked and answer recorded.