Describe the feature or problem you'd like to solve
Custom agents ( .github/agents/*.agent.md ) support a tools: frontmatter list, but the documented aliases are limited to execute , read , edit , search , agent , web , todo. there's no way to grant a custom agent access to the skill tool that interactive sessions have. This forces duplicating an installed skill's full instructions (format, rules, examples) verbatim into the agent's markdown body instead of referencing it live, so the two drift out of sync whenever the skill is updated.
Proposed solution
- Add skill as a recognized tool alias in the tools: frontmatter list (alongside execute , read , edit , search , agent , web , todo ), mapping to the same skill tool exposed in interactive sessions.
- When declared, the custom agent's runtime tool set includes skill(name: string) , resolved against the same skill-discovery order already used interactively (project → user → installed plugins), so no new skill-registration mechanism is needed.
- Optionally, allow scoping to specific skills via tools: ["skill/ponytail-review"] (consistent with the existing some-mcp-server/some-tool namespacing pattern for MCP tools), so an agent can be restricted to only the skills it needs rather than all installed skills.
- No change needed to skill file format ( SKILL.md + frontmatter). this is purely a tool-provisioning change on the custom-agent side.
Example prompts or workflows
A PR-review custom agent ( dotnet-pr-reviewer.agent.md ) that wants to apply the ponytail-review skill's methodology by invoking skill(name: "ponytail-review") directly at review time, instead of embedding a static copy of ponytail-review/SKILL.md in its prompt that must be manually kept in sync.
Additional context
No response
Describe the feature or problem you'd like to solve
Custom agents ( .github/agents/*.agent.md ) support a tools: frontmatter list, but the documented aliases are limited to execute , read , edit , search , agent , web , todo. there's no way to grant a custom agent access to the skill tool that interactive sessions have. This forces duplicating an installed skill's full instructions (format, rules, examples) verbatim into the agent's markdown body instead of referencing it live, so the two drift out of sync whenever the skill is updated.
Proposed solution
Example prompts or workflows
A PR-review custom agent ( dotnet-pr-reviewer.agent.md ) that wants to apply the ponytail-review skill's methodology by invoking skill(name: "ponytail-review") directly at review time, instead of embedding a static copy of ponytail-review/SKILL.md in its prompt that must be manually kept in sync.
Additional context
No response