skill-evolution-manager_skill

Overview

This skill is the central engine for evolving and iterating existing skills based on end-of-session feedback and conversation content. It captures successful patterns, failure lessons, and code or usage constraints to keep the skills library improving over time. It focuses on extracting, persisting, and stitching learned experience into each skill's documentation and metadata.

How this skill works

At the end of a conversation, the skill scans the session to identify satisfiers and pain points, then maps those insights to the specific skills involved. It converts freeform feedback into structured JSON (evolution.json), incrementally merges that data with existing records using helper scripts, and appends human-readable guidance into the skill's documentation. It can also reapply saved experience when skills are updated to preserve accumulated knowledge.

When to use it

• After a session where a tool returned errors, unexpected behavior, or suboptimal results.
• When a user explicitly asks to save feedback or to review what went well and what failed.
• After deploying or updating a skill to reapply previously recorded lessons.
• When multiple skills were used in one conversation and cross-skill lessons emerged.

Best practices

• Always route fixes and preferences through the evolution JSON channel rather than editing docs directly.
• Tag each extracted item with the target skill name and a concise rationale.
• Prefer actionable, reproducible entries (steps to reproduce, environment, exact error text).
• Run the merge and stitch scripts per skill to avoid cross-skill contamination.
• Re-run stitching after skill updates to ensure experience aligns with the new version.

Example use cases

• Capture a user preference like default muted downloads and persist it for the downloader skill.
• Record a platform-specific fix (e.g., Windows path escaping for ffmpeg) so future runs avoid the issue.
• Save a high-value prompt pattern that reliably produces desired output for reuse.
• After a batch skill update, reapply saved constraints so known workarounds aren’t lost.
• When a user reports poor UI wording, extract the complaint and suggest a doc update for clarity.