New skill proposal
Add a cross-cutting hol-guard-agent-runtime-safety skill for researchers using Scientific Agent Skills from supported local coding-agent harnesses.
The skill would directly install and invoke HOL Guard before an agent starts tool-using scientific workflows:
pipx install hol-guard
hol-guard detect --json
hol-guard install <detected-harness>
hol-guard run <detected-harness>
HOL Guard remains the actual local runtime boundary; the skill would not replace scientific validation, package-specific safeguards, credential controls, or the repository's existing security guidance. Deny/review/error/unavailable Guard states would be fail-closed rather than bypassed by relaunching an unprotected agent.
This seems to fit the repository's existing Agent Platforms & Infrastructure scope and its standard Agent Skills / Agent Plugins distribution. If accepted, the maintained skill could be installed through the same npx skills add K-Dense-AI/scientific-agent-skills and gh skill install surfaces already documented here.
I would keep the implementation small and repository-native: one skills/hol-guard-agent-runtime-safety/SKILL.md starting at metadata version 1.0, with no scripts unless maintainers specifically want executable helpers. That avoids adding unnecessary tests/dependencies while still invoking the real hol-guard CLI.
I maintain HOL Guard / Hashgraph Online. AI assistance was used to prepare this focused proposal. Happy to follow the repository's validation and security-scan requirements in a PR if this scope is welcome.
New skill proposal
Add a cross-cutting
hol-guard-agent-runtime-safetyskill for researchers using Scientific Agent Skills from supported local coding-agent harnesses.The skill would directly install and invoke HOL Guard before an agent starts tool-using scientific workflows:
HOL Guard remains the actual local runtime boundary; the skill would not replace scientific validation, package-specific safeguards, credential controls, or the repository's existing security guidance. Deny/review/error/unavailable Guard states would be fail-closed rather than bypassed by relaunching an unprotected agent.
This seems to fit the repository's existing Agent Platforms & Infrastructure scope and its standard Agent Skills / Agent Plugins distribution. If accepted, the maintained skill could be installed through the same
npx skills add K-Dense-AI/scientific-agent-skillsandgh skill installsurfaces already documented here.I would keep the implementation small and repository-native: one
skills/hol-guard-agent-runtime-safety/SKILL.mdstarting at metadata version1.0, with no scripts unless maintainers specifically want executable helpers. That avoids adding unnecessary tests/dependencies while still invoking the realhol-guardCLI.I maintain HOL Guard / Hashgraph Online. AI assistance was used to prepare this focused proposal. Happy to follow the repository's validation and security-scan requirements in a PR if this scope is welcome.