ai-fluency-stewardshiplisted

# Stewardship — governing the run in motion Delegation, Description, Discernment, and Diligence are mostly *setup* and *review*. Stewardship is the competency that applies **while the agent is running**: being a responsible principal in real time. Its core question is simply **"should it keep going?"** | Competency | When it applies | Core question | |---|---|---| | Delegation | Before the run | How much authority to grant? | | Description | Before the run | What to specify? | | Discernment | After a step/run | Was it good? | | Diligence | Before & after | Disclosed and vouched-for? | | **Stewardship** | **During the run** | **Continue, interrupt, or stop?** | ## The three questions (ask at every checkpoint) 1. **Still in scope?** Is the agent doing what was authorized, or has it expanded scope or crossed a "MUST NOT" from the authority-scope declaration? 2. **Trajectory coherent?** Are the steps logically connected to the goal? Is it making progress, or looping / stuck? 3. **Continue, interrupt, or stop?** - **Continue** — on-scope, coherent, progressing. - **Interrupt** — drifting, an unexpected decision was made, or a checkpoint was reached: pause, review, redirect. - **Stop** — a hard constraint was crossed, an irreversible action is imminent, or the behavior is no longer understandable. ## Tripwires (interrupt regardless of where the run is) - Attempts to modify gate logic (`verify.sh`, check scripts) or the spine (`borromeanrings.toml`). -