Capabilities / B · Embodiment (tools)
B.2 Tool registration Active
Given a tool, the agent writes its registry entry with metadata, version, permissions and input/output schema. Graded by schema validation and consistency tests against the MedGauge tool registry schema.
Evidence: none. No benchmark card lists this capability yet. No level is issued: methodology 0.1.0 sets no level thresholds.
- Domain
- B · Embodiment (tools)
- Evidence form
- program
- Veto constraints
- —
- Depends on
- B.1 Tool creation
- Required by
- B.6
- Internal task families
- —
- External anchors
- SkillsBench
- Version
- 0.1.0 · 0.1.0
Definition in detail
A tool registry is only useful if each entry describes what the tool actually does and what it may touch. Wrong permissions or schemas make harness composition and audit unreliable.
Inputs a task gives
- A tool's source and documentation
- The registry schema
- Existing registry entries, including earlier versions of the same tool
- The permission categories the registry recognises
Artifact required
A registry entry with id, name, version, description, permissions (for example file read, file write, network), input_schema, output_schema, dependencies and provenance.
How it is graded
A program validates the entry against the registry schema and runs consistency tests: the declared schemas must accept what the tool accepts and describe what it returns on probe inputs; declared permissions must cover what the tool uses; the version must follow the versioning rule relative to earlier entries; the id must be unique.
In scope
- Metadata, versioning and provenance fields
- Input and output schemas that match observed tool behaviour
- Permission declarations that match what the tool does
Out of scope
- Writing or fixing the tool itself - graded under B.1 and B.5
- Choosing and calling tools during a task - graded under B.3
- The harness card that lists registered components - graded under B.6
Typical failures
Expected failure patterns; not measured rates.
- Copying documentation text that no longer matches the tool's behaviour
- Omitting a permission the tool uses, most often network or file write
- Version numbers that do not reflect an interface change
- Schemas that accept inputs the tool rejects
Standards: std-mcp-2026-07-28 · std-agent-skills
Benchmarks
No benchmark card lists this capability.
Grading references
No grading reference is mapped to this capability.