openclaw - 💡(How to fix) Fix [Feature] Path-specific skill loading (like Claude Code `paths:` frontmatter) [1 participants]

Problem

Currently, OpenClaw loads ALL skills on every request, consuming context tokens regardless of whether the skill is relevant to the current task. This wastes tokens and increases response times.

Proposed Solution

Add a paths: field to SKILL.md frontmatter (like Claude Code's implementation) that allows skills to specify file patterns. Skills with paths: are only loaded when the agent is working with matching files.

Example

---
name: python
description: Python coding standards and best practices
paths:
  - "**/*.py"
  - "**/requirements*.txt"
  - "**/pyproject.toml"
---

# Python coding guidelines...

Behavior

• Skills WITHOUT paths:: Always loaded (backward compatible)
• Skills WITH paths:: Only loaded when agent reads/modifies matching files
• Glob patterns: Standard glob syntax (**/*.py, src/**/*.ts, etc.)

Benefits

Implementation Notes

1. Glob matching: Use micromatch or similar (already used elsewhere in OpenClaw)
2. File access tracking: Track which files agent reads/writes during session
3. Skill eligibility check: At session start AND when new files accessed
4. Hot reload: When skills watcher detects changes, recalculate eligible skills

Prior Art

Claude Code implements this feature:

• Documentation: https://code.claude.com/docs/en/skills#path-specific-rules
• Uses YAML frontmatter paths: field
• Same backward-compatibility approach

Alternative Considered

Using skill description for heuristic matching (less precise, no file context).

Would you like to submit a PR?

Yes, I'm interested in contributing if the feature is approved. I can provide:

1. Implementation for the path matching logic
2. Tests for glob patterns and edge cases
3. Migration guide for existing skills
4. Documentation updates

Community Discussion: This would significantly reduce token usage for users with many skills installed. For example, a user with 20+ skills would only load Python-related skills when editing Python files.