Package Overview¶

bijux-canon-agent exists to coordinate role-based, multi-step behavior without hiding what happened. It owns deterministic orchestration, trace-bearing workflow progression, and agent-facing contracts that sit above reasoning and below runtime authority.

Role Model¶

flowchart LR
    inputs["reasoning artifacts and commands"]
    agent["workflow orchestration"]
    trace["trace-bearing workflow output"]
    downstream["runtime and operators"]

    inputs --> agent --> trace --> downstream

This page should show agent as the workflow layer, not as a place where any hard late-stage problem gets parked. The package matters when it makes coordination visible and replayable rather than merely convenient.

Boundary Verdict¶

If the change decides how roles coordinate, which step runs next, or what trace a workflow must emit, it belongs here. If it decides what a claim means or whether a run counts, it belongs in another package.

What This Package Makes Possible¶

workflow behavior becomes explainable as orchestration rather than as accidental cross-package coupling
trace output stays strong enough for readers to reconstruct what the agent did
reasoning and runtime layers keep their own authority instead of being blurred into orchestration

Tempting Mistakes¶

calling any multi-step behavior an agent feature even when it is really reasoning logic
letting runtime acceptance rules leak into orchestration because they are both late-stage concerns
adding convenience workflow behavior that makes traces harder to defend

First Proof Check¶

packages/bijux-canon-agent/src/bijux_canon_agent for orchestration ownership in code
packages/bijux-canon-agent/tests for determinism and traceability evidence
packages/bijux-canon-agent/apis for tracked agent-facing surfaces

Design Pressure¶

The pressure on agent is to keep orchestration distinct from both reasoning and runtime authority. If traces stop being enough to explain why a workflow moved, the package has absorbed too much hidden policy.