What does this skill do?

The Agent Skill Self-Assessment allows agents to take a moment after completing a complex task to evaluate their own performance against a structured five-point rubric: accuracy, completeness, clarity, actionability, and conciseness. This is not a pass-or-fail assessment, but rather a deliberate step in the reflection process that identifies omissions, flags overconfidence, and suggests specific improvements backed by concrete evidence.

After writing a lot of code
Evaluate the output when the code consists of 3 or more files or more than 50 lines.
Completed multi-step workflows
Reflect on the results after implementing → testing → reviewing in a multi-step process.
Long debugging sessions
Self-assess your work after debugging that took 3 or more attempts.
Design or analysis documents
Rate the quality of a design document, architectural decision, or written analysis.

Usage examples

🔍 Direct Assessment
I just added the retry logic to the HTTP client. Test it yourself.
💬 User's question
How did that go? Check your output after implementing the entire registration flow.
🐞 After purification
After those four attempts to fix the authentication bug, evaluate your work.
📄 Design Document
Evaluate the architecture document you just created using the 5-point rubric.

Features

5-axis milling It independently evaluates accuracy, completeness, clarity, actionability, and conciseness.
Required evidence Any score below 5 must specify the exact gap; it is not enough to simply mention it.
Overall score Calculate the arithmetic mean of the 5 axes, rounded to one decimal place, for a quick overview.
Prioritized Improvements Proposes 1 to 3 specific actions, ranked by impact, with a description of the corrected status.
Immediate Correction If the defect can be corrected in less than 30 seconds, he fixes it on the spot; otherwise, he marks it for rework.

Frequently asked questions

After writing code consisting of 3 or more files or 50 or more lines, completing multi-step workflows, debugging sessions involving 3 or more attempts, or producing design or analysis documents.
No. It is a deliberate step in the reflection process that identifies omissions and reveals areas for improvement before the user has to do so.
The skill indicates what it would do differently. If the issue can be resolved in less than 30 seconds, it corrects it immediately; if it requires rework, it explicitly flags it with the cause and a proposed solution.
No. It is an internal reflection skill of the agent that does not require external connections or an MCP.
Agent Self-Assessment — Quality and Reflection with Claude AI

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.