Participant: codex (AI agent)
Understanding
The autonomy stack exposes multiple truthful but disconnected JSON registries
and projections. data2dsl already compares observations, but agents lack a
deterministic discovery surface describing which sources, entities and URI
references exist and how they connect.
Execution plan
- Normalize explicit JSON sources into a deterministic evidence graph.
- Add bounded query, redaction and limit enforcement.
- Expose the graph through MCP, then verify with heterogeneous
Subactor-shaped fixtures.
Actual changes
- Initialized the bounded ticket and recorded SESSION_EXECUTION_AUTHORIZATION
from the request to execute this work.
- Selected a read-only discovery graph: no implicit disk scan, no credential
resolution and no authority expansion.
- Added deterministic source, entity and reference nodes, typed relationships,
stable hashes, queryable connected subgraphs and fail-closed resource limits.
- Added secret-field redaction and duplicate-source rejection.
- Exposed discovery through the existing MCP definition and dispatcher and
validated it through a real
tools/call request.
- Verified the full suite, scoped lint, governance and a four-source live
Subactor canary.
Blockers
- None inside the recorded intent; proceed without a second confirmation.
- New authority remains required for destructive action, secret access, new
external coordination, material objective expansion and trusted merge.