Imported from sergio-sisternes-epam/autogenesis (references/modules/validate-skill-import-links/SKILL.md). Install upstream with npx skills add sergio-sisternes-epam/autogenesis --skill validate-skill-import-links. Copyright stays with the author.

validate-skill-import-links

Advisory audit only. Does not mutate the target package.

Given a target skill directory (or list of skills), verify that every skill-to-skill invocation follows the portable multi-harness substrate contract defined in the Autogenesis Atlas page autogenesis/decisions/skill-nesting-invocation-pattern.md.

Arguments

• Required: target (one package or an explicit list of packages).
• Optional: report_format, harness_filter.
• Do not override parent-owned subject, mode, operation, work identity, Atlas, or approval.

Follow <skill_root>/references/modules/workflow-discipline/SKILL.md and its invocation contract. The parent reads this entrypoint and issues a support request with its configured compact card before the procedure. Keep the active operation and inherited context; return the facet result and an honest receipt. Shared assets resolve from skill_root, not cwd.

Procedure

1. Read the Skill chaining rule in the activated Autogenesis root SKILL.md, resolved from resolved.skill_root, as the baseline for external calls.

2. If additional mapping detail is needed, consult the canonical reference page through Atlas in an already available Autogenesis reference store. Obtain its identity from the activated package's mesh, not the target's. Keep the resolved location in a local reference_store_root; never overwrite protected context.atlas_root or mount a store into the target as part of this advisory check. If needed detail is unavailable, report that review limitation rather than invent it. The target need not declare Atlas or copy the reviewer's reference store.

3. Identify the target(s). Accept a path such as /home/workdir/.grok/skills/<name>/ or a list of skill names. arguments.target is required by the invocation contract. Missing or empty input is an invalid request; do not infer the current subject as a fallback.

4. For each target SKILL.md (and any referenced module entrypoints inside it):

   • Distinguish a parent reading a private module from invoking another external skill. Internal reads use declared relative entrypoints; they do not require catalogue loading or an invocation validator.
   • Locate every actual external skill call and check the substrate contract (the three mandatory steps), directly or through a clear shared instruction. A mere resource link is not an external skill call.
   • If the package claims multi-target or lists specific harnesses, also check that the corresponding rows from the per-harness mapping table are present or clearly delegated.
   • With no external skill calls, report substrate_contract n/a and explain why. Do not require unused dependencies or boilerplate.

5. Emit a structured report fragment with exactly this shape:

   FACET: validate-skill-import-links
   target: <path or name>
   substrate_contract: pass | gaps | n/a
   harness_mapping: pass | gaps | n/a
   gaps:
     - <file:line or section> — <missing element>
   suggested_fix: |
     <exact portable text that should be inserted>
   status: pass | needs-work

6. Do not edit the target. If the operator wants the gaps fixed, they must later run an implement path (or edit manually) after reviewing the aggregated REPORT.

7. Optional protocol check: when the target under review implements or claims nesting, the report may recommend re-running the canonical reference fixture skill-test-a → skill-test-b on the current harness.

Non-goals

• Automatic rewriting of the target.
• Design or implement of new skills.
• Changing genesis.
• Full package conformance (that is the deep review-package path).