Skip to content

kirillnedoboy/opencare-trust

unversioned · 4bff32e9c5d3Apache-2.0

OpenCare portable trust package: inspect Trust Envelopes and Execution Receipts, verify provenance, policy, and disclosure fields, and summarize source-backed health context. Skill-only; verification is never authorization.

GitHub

opencare-trust

Portable, skill-only Agent Plugins v1 package for the OpenCare trust contract.

Contents

  • plugin.json — strict Agent Plugins 1.0.0 manifest (closed field set; additionalProperties: false).
  • skills/opencare-trust-envelope/ — generic trust skill: inspect Trust Envelopes, verify Execution Receipts, understand provenance / policy / disclosure fields, and distinguish verification from current authorization. Bundles the three trust JSON Schemas and two synthetic fixtures in assets/ and the field-by-field contract reference in references/PROTOCOL.md.
  • skills/opencare-health-agent/ — OpenCare health agent skill: summarizes a supplied source-backed context packet, traces claims to sources, and prepares clinician questions. Never diagnoses or recommends. Bundles context.schema.json, answer.schema.json, and installation/examples docs.
  • README.md — this file.

There is deliberately no mcp.json: the package is skill-only (Agent Plugins 1.0.0 §6.2 permits a missing MCP location).

Why there is no version field

plugin.json intentionally omits version. The plugin package is versioned independently from the SANO application release. No standalone Agent Plugin release version has been minted yet, so the manifest intentionally omits version rather than borrowing the SANO application version. The root project version is unchanged.

Source of truth

Canonical sourcePackaged copy
health skill SKILL.md + schemasskills/opencare-health-agent/
health skill docsskills/opencare-health-agent/references/
trust skill SKILL.md + references/PROTOCOL.mdskills/opencare-trust-envelope/
schemas/agent-trust/*.schema.jsonskills/opencare-trust-envelope/assets/
fixtures/agent-trust/allowed-*.jsonskills/opencare-trust-envelope/assets/

scripts/build_agent_plugin.py is the only writer of this package; editing a skill means editing the canonical file and rerunning the script. A drift test (tests/test_agent_plugin_drift.py) asserts the committed copy stays byte-identical to the canonical sources. The build is deterministic, offline, and creates no symlinks.