ExplanationVersion 0.2.0
LCA Skills host support and qualification
Use explicit structural, runtime, host, and scientific evidence labels instead of a generic supported/unsupported claim.
Qualification vocabulary
| Status | Meaning |
|---|---|
| NOT_TESTED | No representative evidence |
| NOT_INSTALLED | Required component is absent |
| DETECTED_UNQUALIFIED | Component exists, but no successful representative run |
| STRUCTURALLY_QUALIFIED | Package/frontmatter/references/hashes pass static checks |
| RUNTIME_TESTED | A representative local command executed successfully |
| HOST_TESTED | Package was imported and invoked in the named host/version |
| SCIENTIFICALLY_QUALIFIED | Known-case model semantics/results and seeded failures passed with practitioner review |
| FAIL | Tested behavior did not meet acceptance criteria |
Current package evidence boundary
| Host/package | Current evidence boundary |
|---|---|
| ChatGPT upload package | Structurally qualified; scan/install/activation/artifact behavior still requires evidence from the target account/workspace before HOST_TESTED |
| Claude.ai upload package | Structurally qualified; upload/activation/code-execution behavior remains host-specific |
| Claude Code plugin | Structurally qualified package with skills, agents, hook, and MCP config; the actual Claude Code build still needs host testing |
| Claude Code standalone | Structurally qualified; install/invocation must be tested on the target machine |
| Codex package | Structurally qualified; import workflow and invocation syntax must be confirmed in the actual Codex host/version |
| Generic Agent Skills host | Portable package shape; host-specific import/invocation evidence is still required |
A skill package does not grant external-system access
Installing LCA Skills does not automatically grant access to desktop openLCA, local Brightway projects, GREET files, licensed databases, or private files. External-system access requires an explicit connector/app/MCP path and its own permissions and qualification.
Minimum host test
- Record host and version.
- Verify package hash.
- Import/install the package.
- Confirm expected skill discovery.
- Run one positive activation case.
- Run one unrelated negative activation case.
- Run a missing-tool honesty case.
- Test artifact/tool permissions when available.
- Preserve the transcript, permissions, failures, and result.