Skills installed into the standard ~/.agents/skills/ directory should be automatically integrated into the prompt and provided to the LLM to also be useable. this should follow the default patterns and structures that skills are built and not rely on any ds4server specific structure and is used to for example use the allium skill in DS4Server to help the weeker model to properly build up plans and requirements.
Skills installed into the standard ~/.agents/skills/ directory should be automatically integrated into the prompt and provided to the LLM to also be useable. this should follow the default patterns and structures that skills are built and not rely on any ds4server specific structure and is used to for example use the allium skill in DS4Server to help the weeker model to properly build up plans and requirements.
hugo
added the question label 2026-08-03 15:12:52 +00:00
hugo
changed title from Is skills in dev brain the correct way? to support skills in ~/.agents/skills/ and integrate them into the prompt2026-08-07 14:32:15 +00:00
Implemented in commit c0681de (Integrate standard agent skills). DS4Server now discovers valid Agent Skills from ~/.agents/skills/*/SKILL.md using the standard YAML name and description metadata, injects a deterministic metadata-only catalog with progressive-disclosure instructions into normal turns, reminders, and Ralph child prompts, and lets the existing read/search/list tools access each resolved skill directory read-only. Linked skill directories are supported; writes and symlink escapes remain blocked. Added an isolated test covering discovery, validation, prompt content, linked skills, resource reads, write rejection, and path escape rejection. Verification: cargo fmt --all -- --check; cargo clippy --all-targets --all-features -- -D warnings; make bundle; cargo test --all-features (182 passed, 14 ignored hardware/browser fixtures).
Implemented in commit c0681de (Integrate standard agent skills). DS4Server now discovers valid Agent Skills from ~/.agents/skills/*/SKILL.md using the standard YAML name and description metadata, injects a deterministic metadata-only catalog with progressive-disclosure instructions into normal turns, reminders, and Ralph child prompts, and lets the existing read/search/list tools access each resolved skill directory read-only. Linked skill directories are supported; writes and symlink escapes remain blocked. Added an isolated test covering discovery, validation, prompt content, linked skills, resource reads, write rejection, and path escape rejection. Verification: cargo fmt --all -- --check; cargo clippy --all-targets --all-features -- -D warnings; make bundle; cargo test --all-features (182 passed, 14 ignored hardware/browser fixtures).
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Skills installed into the standard ~/.agents/skills/ directory should be automatically integrated into the prompt and provided to the LLM to also be useable. this should follow the default patterns and structures that skills are built and not rely on any ds4server specific structure and is used to for example use the allium skill in DS4Server to help the weeker model to properly build up plans and requirements.
Is skills in dev brain the correct way?to support skills in ~/.agents/skills/ and integrate them into the promptImplemented in commit
c0681de(Integrate standard agent skills). DS4Server now discovers valid Agent Skills from ~/.agents/skills/*/SKILL.md using the standard YAML name and description metadata, injects a deterministic metadata-only catalog with progressive-disclosure instructions into normal turns, reminders, and Ralph child prompts, and lets the existing read/search/list tools access each resolved skill directory read-only. Linked skill directories are supported; writes and symlink escapes remain blocked. Added an isolated test covering discovery, validation, prompt content, linked skills, resource reads, write rejection, and path escape rejection. Verification: cargo fmt --all -- --check; cargo clippy --all-targets --all-features -- -D warnings; make bundle; cargo test --all-features (182 passed, 14 ignored hardware/browser fixtures).