Hussh Consent MCP Skill

Purpose and Trigger

• Primary scope: hushh-consent-mcp-ops
• Trigger on direct connector tasks against the Hussh Consent MCP surface:
  • search_user_scopes
  • prepare_campaign_context (Hussh ADK compatibility)
  • request_consent
  • check_consent_status
  • get_encrypted_scoped_export
• Use this when the question is about what the connector returns right now, not when changing backend consent logic.
• Avoid overlap with backend, mcp-developer-surface, and security-audit.

Coverage and Ownership

• Role: spoke
• Owner family: backend

Owned repo surfaces:

1. packages/hushh-mcp
2. consent-protocol/docs/mcp-setup.md
3. .codex/skills/hushh-consent-mcp

Non-owned surfaces:

1. backend
2. security-audit
3. docs-governance

Do Use

1. Verifying whether a specific user resolves to any requestable scopes.
2. Comparing prod vs UAT connector behavior for the same identifier.
3. Debugging connector startup/config mismatches across Codex, Claude Desktop, or other MCP hosts.
4. Verifying whether a missing result is a connector/config issue versus an actual empty discovery response.

Do Not Use

1. Changing consent scope semantics or IAM policy.
2. Broad backend runtime work outside the MCP connector.
3. UI-only consent surface work.

Read First

1. packages/hushh-mcp/README.md
2. consent-protocol/docs/mcp-setup.md
3. docs/reference/operations/coding-agent-mcp.md

Workflow

1. Verify the local host config first if the issue smells like a connector startup failure.
2. When scope search returns no matches, check at least two environments when available before concluding it is an information issue.
3. Distinguish clearly between:
   • connector failed to start
   • connector connected but returned an empty scope-search result
   • environment mismatch between local, UAT, and production
4. Default scope search should stay compact and paginated. Deep leaf-path expansion is debug-only.
5. When auditing PKM parity, compare connector output to the backend metadata route and manifest-backed domain truth for the same user before concluding discovery is wrong.
6. Report only the safe structured connector result before inferring causes; never surface identifiers or tokens.
7. If the connector returns empty in multiple environments, treat that as a valid empty response until proven otherwise.

Handoff Rules

1. If the request is still broad or ambiguous, route it back to backend.
2. If the task becomes package/docs evolution, hand off to mcp-developer-surface.
3. If the task becomes consent-policy or trust-boundary analysis, hand off to security-audit.
4. If the task becomes backend API contract work, hand off to backend-api-contracts.

Required Checks

python3 .codex/skills/codex-skill-authoring/scripts/skill_lint.py
./bin/hushh codex audit
cd packages/hushh-mcp && npm run print-config