Summary
The CLI supports single-file sub-agent validation and publishing via:
skillserver lint subagent <path>
skillserver publish-subagent <path> --version <version>
This works for one-off publishing, but it leaves a gap compared to the existing skills workflow:
skillserver lint ./skills
skillserver publish-all ./skills
For repositories that maintain multiple sub-agent definitions, CI/CD currently has to implement its own file discovery, version lookup, lint loop, publish loop, dry-run behavior, and error aggregation.
Requested capability
Add first-class batch support for sub-agent definitions, similar to the existing skills workflow.
Possible command shape:
skillserver lint subagents ./agents
skillserver publish-subagents ./agents
skillserver publish-subagents ./agents --dry-run
skillserver publish-subagents ./agents --force
Alternative command names are fine. The important capability is batch validation and batch publishing over a directory of sub-agent markdown files.
Versioning question
publish-subagent currently requires --version because local sub-agent files do not require version metadata. For batch publishing, the CLI needs a deterministic way to resolve each file's publication version.
Possible approaches:
- Support an optional frontmatter field such as
metadata.version or version for publish-time use.
- Support a sidecar manifest mapping sub-agent names to versions.
- Support a directory-level default
--version for all files, if that is sufficient for release-style workflows.
Desired CI behavior
A CI-friendly batch command should:
- Return non-zero on validation or publish failures.
- Support
--dry-run without requiring writes.
- Report all invalid files where practical rather than failing at the first file.
- Detect duplicate sub-agent names in the same input set.
- Preserve idempotent behavior for already-published versions, matching the skills publishing experience.
Why this matters
Sub-agents are now first-class native resources in SkillServer, but the CLI ergonomics are still single-file oriented. Batch commands would let source repositories manage sub-agents with the same operational model already available for skills.
Summary
The CLI supports single-file sub-agent validation and publishing via:
skillserver lint subagent <path>skillserver publish-subagent <path> --version <version>This works for one-off publishing, but it leaves a gap compared to the existing skills workflow:
skillserver lint ./skillsskillserver publish-all ./skillsFor repositories that maintain multiple sub-agent definitions, CI/CD currently has to implement its own file discovery, version lookup, lint loop, publish loop, dry-run behavior, and error aggregation.
Requested capability
Add first-class batch support for sub-agent definitions, similar to the existing skills workflow.
Possible command shape:
Alternative command names are fine. The important capability is batch validation and batch publishing over a directory of sub-agent markdown files.
Versioning question
publish-subagentcurrently requires--versionbecause local sub-agent files do not require version metadata. For batch publishing, the CLI needs a deterministic way to resolve each file's publication version.Possible approaches:
metadata.versionorversionfor publish-time use.--versionfor all files, if that is sufficient for release-style workflows.Desired CI behavior
A CI-friendly batch command should:
--dry-runwithout requiring writes.Why this matters
Sub-agents are now first-class native resources in SkillServer, but the CLI ergonomics are still single-file oriented. Batch commands would let source repositories manage sub-agents with the same operational model already available for skills.