An agent picks a development task that someone tagged him on Multiple agent sessions may touch the code
Process ends at a commit ready for pre-merge review
Control your SDLC wrt to questions like:
- Did the agent stay aligned with our policies?
- Did it get the mandatory context — guidelines, requirements, architecture?
- Did it stay within the boundaries of its task?
- Which sessions, agents, models and tools built the thing we are about to release?
- Who is accountable?
Control the supply chain risk introduced by the agents framework
Notice: the autonomous code development use case resembles the Human-in-the-loop code development and only deffers in human review which is missing from the autonomous development.
The user who's identity is used by the agent. The owner of the agent.
The sessions are the coding agent sessions, chat, interactive_run, including subagents
Its subject points to the git commit (as the SDLC entity)
It holds the reference to the session logs that affected this commit
It will hold the data for supply chain analysis, and for answering the various checks:
It holds the human owner
- alignment check wrt policies
- mendatory context present in all sessions
- alignment check wrt task information
- human accountable information
- agentic-session-log — one per agentic session (incl. subagents)
- agentic-code-development-evidence — on the git commit
- agentic-session-alignment-evidence — optional, per session log; produced on the server
[implementation_details.md](./implementation_details.md)[objects_examples/](./objects_examples/)