Installed skills fail to execute due to missing progressive loading and lost context paths

[felix307253927]

Description

Installed skills are not executing correctly. The root cause appears to be the lack of a progressive loading mechanism. Currently, the system attempts to eagerly load all enabled skills defined in the skill_allowlist simultaneously. Furthermore, the execution context seems to lose the specific file paths required to locate and run these skills.

Expected Behavior

1. Progressive Loading:

• When the LLM determines that a specific skill is required to answer a user query, it should trigger a tool call to a dedicated method, e.g., load_skill(skill_name).
• This load_skill method will then:
  • Resolve the correct file path for the requested skill.
  • Load the full instructions, code, or context associated with that skill.
  • Inject this detailed content into the current conversation context for immediate execu

2. Path Preservation: Ensure that the absolute or relative file paths for each skill are correctly maintained and accessible within the execution context.
3. Successful Execution: Skills should execute reliably when invoked, without requiring a full preload of all allowed skills.

Steps to Reproduce

1. Install multiple custom skills and add them to the skill_allowlist.
2. Restart the application/service.
3. Observe the startup logs (likely showing attempts to load all skills).
4. Attempt to invoke a command that requires one of the installed skills.
5. Result: The skill fails to execute.

OpenFang Version

0.4.4

Operating System

Windows

Logs / Screenshots

No response

[felix307253927]

Can you add a rustfmt.toml file? I modified the build_skill_summary prompt and can now run skills normally, but cargo fmt formats some code differently from the repository's style.

[jaberjaber23]

Investigated. Installed skills from ClawHub work correctly — load_skill() sets real filesystem paths. Bundled skills use PromptOnly runtime (no code execution needed), so the placeholder path doesn't affect them. The progressive/lazy loading described is a feature request, not a bug fix. The current eager loading at boot is intentional for performance. If you have a specific skill that fails to execute, please share the skill name and error message.