Rhizamonitor
live

Roadmap

Runs and journal

landedintention4 leaves, 4 with evidence

Intent

Why this exists

Runs and journal: genesis run start, status and resume run a workflow's agent, wait and command steps with a local journal and no daemon.

Leaves

4 leaves, 4 with evidence

  • has evidence

    A workflow is a TOML file of named steps, each of kind agent, wait or command, with outcomes naming the next step; an unknown kind or a missing target is refused.

    AcceptanceA test loads a valid file and rejects both bad ones with a message.
  • has evidence

    An agent step launches its agent through the adapter on the run's change set and ends with the session's outcome.

    AcceptanceA testscript with a fake adapter runs an agent step to its outcome, and the fake Claude Code receives the step's prompt, its placeholders filled, as its positional prompt; a step with no prompt or an unknown placeholder is refused.
  • has evidence

    A step's `max_visits` caps how often a run enters it, and going over fails the run with the step named.

    AcceptanceA testscript loops a step past its cap and sees the run fail.
  • has evidence

    Each run keeps a local journal of its steps, outcomes and visits outside the model's log.

    AcceptanceAfter a run the journal holds every step and `.genesis/log/` is unchanged.

Relations

Serves and depends on

Serves
Genesis runs the work: milestone 3 is built with Genesis orchestrating, not an orchestrator session.
Depends on
—

Runs

0 runs

No runs

Nothing has run on this change set.

History

Lifecycle

  • Proposed
  • Accepted
  • Landed