How does it help with long-term AI alignment?

By recording user corrections and preferences as PPUs, the skill ensures that future sessions inherit these insights, reducing the need for the user to repeat instructions.

Where is the feedback data stored?

Feedback is written to structured markdown files within the 'workspace/{process}/{slug}/feedback/' directory for the orchestrator to process.

What types of feedback does the skill capture?

It captures four main types: Persistent Preference Updates (PPU), Output Quality Issues (OQI), Stability Gates (GATE) to reject harmful changes, and Stability Anchors (STA) to preserve successful behaviors.

Does this skill increase token costs during coding?

No, the skill is designed to have zero cost during productive work; it only activates at the very end of a session after the primary tasks are completed.

What is the Self-Evaluation skill for Claude Code?

It is a specialized skill that allows Claude agents to generate structured markdown feedback about their performance, preferences, and errors during the shutdown phase.

Agent Self-Evaluation & Feedback

Name: Agent Self-Evaluation & Feedback
Author: ZoranSpirkovski

byZoranSpirkovski

0•

生产力与工作流

Generates structured improvement signals at agent shutdown to drive continuous learning and persistent behavioral alignment.

The Self-Evaluation skill enables Claude agents to autonomously reflect on their performance and record structured feedback signals after completing a task. By activating only during the shutdown phase, it provides a zero-cost mechanism for continuous improvement, capturing Persistent Preference Updates (PPU), Output Quality Issues (OQI), and Stability Gates (GATE). This skill ensures that every user correction and session outcome is documented in a standardized format, allowing the PAS framework to optimize agent behavior over time without interrupting active development workflows.

主要功能

01Zero-cost execution during active productive work

020 GitHub stars

03Post-session reflection and structured signal generation

04Automatic routing of framework-level feedback to GitHub issues

05Stability Gates and Anchors to prevent behavioral regressions

06Persistent Preference Updates (PPU) for long-term alignment

使用场景

01Recording user preferences to eliminate repetitive corrections across sessions

02Documenting factual errors or inefficiencies to improve future output quality

03Establishing stability anchors for complex tasks that require consistent behavior

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add zoranspirkovski/pas self-evaluation

For use in Claude.ai and ChatGPT

主要功能

01Zero-cost execution during active productive work

020 GitHub stars

03Post-session reflection and structured signal generation

04Automatic routing of framework-level feedback to GitHub issues

05Stability Gates and Anchors to prevent behavioral regressions

06Persistent Preference Updates (PPU) for long-term alignment

使用场景

01Recording user preferences to eliminate repetitive corrections across sessions

02Documenting factual errors or inefficiencies to improve future output quality

03Establishing stability anchors for complex tasks that require consistent behavior

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add zoranspirkovski/pas self-evaluation

For use in Claude.ai and ChatGPT