> Production API availability: use /api/v1/help?view=guides and the resource operations catalog before following a recipe. Some localhost capabilities are not deployed yet; an absent operation is unavailable.
What this module owns
Agent-proposed research work logs with validation, idempotency and a human review stage before accepted records.
Use the contract endpoint
GET /api/v1/research-logging/contract returns the required fields, supported record types, authentication and review workflow. Build the payload from that live contract rather than inventing an accepted-ledger record.
curl --fail-with-body -X POST "$API_URL/research-logging/intake?dry_run=1" -H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" --data-binary @research-record.jsonOnly submit without dry_run after validation and review of the proposal. The accepted ledger is a different state from an agent intake record. Verify idempotency on a dedicated fixture before enabling automated retries.
First read
Use API_URL with /api/v1, for example http://localhost/api/v1. Set API_KEY privately. Replace any resource placeholder with an ID returned by discovery.
curl --fail-with-body "$API_URL/research-logging/contract" \
-H "Authorization: Bearer $API_KEY"Registered agent routes
The entries below are extracted from the current route and handler source. Scope tokens can represent alternatives or different method branches, not a requirement to grant every listed scope. Body fields are observed inputs, not a complete required-field schema. Follow the resource guide and the deployed validation response for a write.
/api/v1/research-logging/contract
Intended request methods: GET. Confirm allowed_methods and method_evidence in the deployed operations catalog; documented methods do not grant access.
Scope tokens in the implementation: research_logging:read, research_logging:write
Selected rejection messages:
- GET required
/api/v1/research-logging/intake
Intended request methods: GET, POST. Confirm allowed_methods and method_evidence in the deployed operations catalog; documented methods do not grant access.
Scope tokens in the implementation: research_logging:read, research_logging:write
Observed JSON fields: external_id, idempotency_key, payload, program_code, record_type, source_agent, source_reference.
Selected rejection messages:
- GET or POST required
- Idempotency key already exists with different payload.
- external_id already identifies another intake record.
Access boundary
External agents use only the versioned operations documented above and returned by the deployed catalog. Browser actions and internal controllers are not substitutes for missing API operations.
Verification before automation
Read back the owned record after a write. On a dedicated local fixture, test missing input, wrong owner, missing scope and stale revisions where supported. Confirm the request did not change state when rejected. Check the generated projection separately from the source record. Never replay sends, payments or deletion requests merely to test connectivity.
See API validation and remaining gaps for the shared test strategy.
HTTP discovery and safety contract
Read GET /api/v1/help?view=guide&resource=research-logging for this module's machine-readable guidance. Discover supported operations through its discovery_url when present. A null discovery_url means no direct resource API is documented; related endpoints, when provided, have their own ownership and scope requirements.
Use GET /api/v1/me with your key to confirm identity and scopes. Only request methods and fields supported by this deployment. The shared transport boundary rejects unsupported methods and malformed JSON on covered routes, but field schemas and revision checks remain resource-specific. Notes and Carousels expose typed save schemas and payload-only validation; other modules must not be assumed to share that contract.
Create, update, replace, patch, delete, send and publish are distinct capabilities. An operation absent from the returned catalog is unavailable, not permission to guess a URL. Read back after writes; after a timeout inspect the result before retrying. Never automatically replay a payment, message, deployment or deletion.