skill-everything

You are using the skill-everything knowledge system: agent memory in plain Markdown, versioned in Git, that grows by capturing your own past mistakes as committed rules.

Your responsibilities

1. Before every implementation: check the sub-skill directory below,

load the matching skill via skill_resource(...).

1. After every mistake: execute the self-extension workflow.
2. When you learn something new: add it to the appropriate category.

Sub-skill directory

Error capture triggers

Start the self-extension workflow when any of these is met:

• A test fails because of code you wrote.
• The user corrects you ("That was wrong", "Remember this").
• You realise during implementation that your first approach was wrong.
• A deployment problem occurs that your code caused.

Load the workflow:

skill_resource(skill_name="self-extension-workflow")

Important

• Search before write. Before logging a new error, search the existing log for similar entries. Update instead of duplicate.
• Action directives, not descriptions. Always formulate rules as "Always X before Y" or "Never Z without W".
• Stay compact. Each sub-skill stays under 3,000 tokens. If exceeded, split.
• PR-flow is mandatory. Self-extension commits are opened as PRs labelled needs-rule-review. Never push to main.
• Without `skill_resource`? If your agent doesn't have the tool, load files directly via filesystem access from the skills/<name>/ folder.

This file is generated from skills/_index.yml. Edit the index, then run python tools/render_loaders.py.