Skill Feedback Writer
Evaluate another skill's output to identify missing instructions and generate structured feedback for improvement.
The skill is limited to evaluating other skills' outputs and generating JSON feedback, with no sensitive data handling or external effects. However, it instructs to directly output scores and instructions without explicit user confirmation, lacks least-privilege, confirmation, or data-flow transparency, and relies on unverified output content. Deducted 7.
Skill description is clear, but lacks explicit implementation details, error handling, and edge cases. No tests or reproducible examples, so reliable execution cannot be confirmed. Deducted 5.
Skill targets a specific task (improving pr-comment-evaluator) but does not clearly define scenarios, non-fit cases, or trigger conditions. Chinese support not mentioned. Deducted 8.
Skill provides basic information architecture and output format, but lacks install/dependency notes, version info, maintenance responsibility, and update path. No known-limitation disclosure. Deducted 8.
Skill can complete basic task (produce JSON output), but whether output is directly usable depends on downstream skills, lacking examples or validation. Deducted 9.
No tests or other verification material; only static reading. Cannot independently verify its behavior. Deducted 6.
- Skill may produce inaccurate feedback because it relies on unverified output content.
- No error handling or edge case documentation provided, may lead to unexpected behavior.
- Chinese language support or availability from mainland China networks not mentioned.
What does this skill do, and when should you use it?
This skill acts as an automated quality assessor, determining whether another skill's output, such as that of pr-comment-evaluator, is sufficient for its intended task. It analyzes the diff risk summary and comment evaluation, and produces a JSON report indicating whether the skill performed well or if it missed a key instruction. It assigns a score from 0.0 to 1.0 based on specific rules, prioritizing runtime risk over superficial tone. The skill is designed to be used in a feedback loop to improve a skill library.
Reads the input to the evaluated skill (diff risk summary, comment evaluation) and the output from the evaluated skill, applies predefined logic (e.g., if pr-comment-evaluator only judges tone or fails to compare the comment against concrete runtime risk, then target it for improvement), and generates a JSON object with keys: diff_risk_summary, comment_evaluation, skill_to_improve, score, feedback, missing_instruction. It assigns a score based on missing important requirements, and provides clear feedback for improvement.
- When you need an automated quality check on whether another skill, like pr-comment-evaluator, is performing well.
- When you want to continuously improve a skill library based on evaluations of skill outputs.
- When you need a standardized way to identify missing instructions in a skill.
- When you want a reusable quality gate to ensure skills meet defined criteria.
What are this skill's strengths and limitations?
- Provides clear, structured feedback for skill improvement.
- Gives an objective score (0.0-1.0) based on specific criteria.
- Focuses on key runtime risks, preventing superficial improvements.
- Easy to integrate into automated pipelines.
- The skill's documentation only describes the concept, not the specifics of input/output formats.
- No tests or evidence of validation on specific platforms are provided.
- Assumes the evaluated skill follows specific naming conventions (pr-comment-evaluator).
- Depends on the evaluated skill being able to provide the required inputs, which may not always be available.
How do you install this skill?
Clone the full repository to your local machine or manually create the skill folder. The skill path is: examples/demos/skill_feedback_loop/skills/skill-feedback-writer/SKILL.md. It needs to be placed in a directory that follows the Agent Skills standard (a folder with SKILL.md and optional additional files).
How do you use this skill?
Provide the input that the evaluated skill would have received (typically in JSON format with diff risk summary and comment evaluation), and ask this skill to evaluate it. For example, you could say: "Evaluate this output of pr-comment-evaluator." The skill will return a JSON object with the evaluation. You can use it with any client that supports the Agent Skills framework, such as Claude Code or Codex.
How does this skill compare with similar options?
This skill is related to and responds to other skills in the repository, such as pr-comment-evaluator and diff-risk-explainer, but no other external tools are mentioned.