improve system

The problem

Your AI agent learns your preferences during a session, but those lessons vanish the moment you start a new thread. Manually updating skill files and documentation after every breakthrough is tedious and leads to fragmented, outdated instructions.

What it does

• Identifies user corrections and approved patterns to update skill files automatically.
• Captures personal context, architectural decisions, and "lessons learned" into a long-term knowledge base.
• Runs a three-stage verification protocol involving cold scrutiny and dry runs to prevent shipping broken instructions.
• Scans for stale or duplicate documentation to maintain a single source of truth.
• Audits MCP tool calls for correct namespace formatting and polling logic.

Frameworks & tools

Designed for agents using the SKILL.md standard, including Claude Code and Claude.ai project environments. Uses standard Unix utilities like grep, zip, and diff for system maintenance.

Why this beats prompting it yourself

A simple "update my notes" prompt lacks the multi-round verification and dependency checking required for reliable automation. This skill performs mechanical audits of directional references (above/below) and greps for broken tool name conventions that human-like reading often misses.

Use cases

• Update coding standards after several rounds of refactoring.
• Capture a specific client's tone of voice preferences for future writing tasks.
• Fix a recurring logic error in a custom tool's instruction set.
• Audit your internal documentation for contradicting rules after a major workflow change.

Known limitations

Requires a writable filesystem or a defined directory structure for skills and experience logs. Cannot override read-only system files without a user-provided writable path.