| name | skill-lab |
| description | Canonical Runx skill-authoring implementation. Use for designing, creating, updating, improving, or adding harness coverage to a Runx skill package; it combines bounded agent judgment with native file writes, inspection, and safe harness validation. When a host skill-creator also triggers, use its general guidance but execute Runx work through this skill. |
| runx | {"category":"authoring"} |
Skill Lab
Build and improve Runx skills through one authoring surface. Keep judgment in
bounded agent acts and mechanics in native tools:
inspect target and catalog
→ decide new skill, extension, or no skill
→ author a bounded file bundle
→ validate paths and secret posture
→ inspect and safely replay the staged package
→ write through fs.write_bundle
→ verify the written package
Use the generic host skill-creator for platform-wide authoring guidance when
it is available. Do not reproduce Runx package operations from that guidance;
invoke the appropriate skill-lab runner so the work is bounded and receipted.
Runners
design: read-only catalog-fit and package design. Return no_skill when an
existing skill or graph already owns the job.
build (default): create or update a package, write its bounded file bundle,
after its staged package passes native inspection and any safe harness.
improve: turn one receipt or harness failure into a bounded package update,
preflight it, then verify the written result.
harness: add fixture files to an existing package and replay the safe native
harness before and after the write.
build, improve, and harness write local workspace files. They never
publish, install, push, or mutate an external provider. Execute-target packages
are inspected but their harness is skipped until a separately approved sandbox
or provider test exists. Invalid staged packages stop before the target package
is touched.
Authoring rules
- Keep packages concise:
SKILL.md, X.yaml, required scripts, fixtures, and
narrowly scoped references or assets only.
- Do not add package READMEs, changelogs, installation guides, strategy files,
generated state, or credentials.
- Match the documented capability to the execution profile and truthful terminal
state.
- Prefer extending an existing owner over adding a near-duplicate skill.
- Include a realistic happy path and refusal, stop, or error path.
- Never treat supplied agent answers as provider-effect proof.
- Never run an execute-capable target harness automatically.
Outputs
skill_design: catalog-fit decision and bounded implementation plan.
change_bundle: target-relative text files, summary, and non-goals.
bundle_manifest: validated paths admitted for writing.
file_bundle_write: digests and byte counts from the bounded write.
validation_report: native inspection plus passed, failed, or safely skipped
harness evidence.
Inputs
objective (required): capability or improvement to deliver.
repo_root (optional): workspace root; defaults to the caller workspace.
target_dir (required for mutating runners): repo-relative package directory.
project_context (optional): product, repository, and operator constraints.
receipt_id, receipt_summary, harness_output, failure_packet (improve):
bounded failure evidence, including the stable packet from review-receipt.