Agent/session operator¶
Where to start¶
Read the session and agent-state guide, then the repository map.
Where the code lives¶
.codex/: operational state, policy, memory evidence, and session artifactsagents/: packaged orchestration primitivessrc/cognitive_brain/: cognitive contracts and memory/coordination componentssrc/aries_serpent_core/brain/: optional OODA, checkpoint, and resume utilitiesscripts/ci/: session bootstrap, recovery, query, and wrap-up tools
What this role cares about¶
Session continuity, evidence provenance, checkpoint boundaries, multi-lane ownership, WEC preservation, accountability, and safe handling of generated state.
Key technologies¶
JSONL, JSON, YAML, SQLite, DuckDB, OODA/PDA/AfterMath concepts, GitHub MCP, custom agents, and repository CI helpers.
Typical workflow¶
- Load policy, repository state, recent PDA records, and the active PR state.
- Divide independent work into named parallel lanes.
- Keep generated observations separate from canonical source and documentation.
- Record validation evidence and decisions.
- Update accountability and session-close artifacts.
Common gotchas¶
.codex/mixes canonical policy with dated and generated evidence; file labels and owning scripts determine authority.- Agent memory, checkpoints, and git history solve different continuity problems.
- Do not edit generated indexes when their header names a generator.
- Preserve maintainer-controlled WEC selections across progress updates.