borromeanrings

Decide what an AI agent should do, what the human should own, and how much autonomy to grant — before a multi-step or ambiguous task. Produces an explicit authority-scope declaration and, at project/feature kickoff, a short 4D alignment. Use when the right division of labor isn't obvious, when starting a sprint or feature, or before handing an agent a long autonomous run. Applies the Delegation competency of the AI Fluency framework (see docs/AI-FLUENCY.md). Triggers: "should I delegate this", "plan this work", "scope the agent", "4D kickoff", "authority scope".

Take responsibility for AI-assisted work before it is shared, committed, published, or deployed — choosing the right tool, disclosing AI's role, and verifying you can vouch for the output. Provides a pre-share checklist, disclosure guidance, and the agent's VERIFICATION STATUS block. Applies the Diligence competency of the AI Fluency framework (see docs/AI-FLUENCY.md). Use before anything leaves the workspace. Triggers: "how should I disclose AI use", "am I being responsible here", "diligence check", "ready to ship".

Critically evaluate what an agent produced — and, for multi-step runs, how it got there — before using, sharing, or building on it. Provides an output-review checklist, an agentic trajectory-audit format, and the Description→Discernment diagnosis loop. Applies the Discernment competency of the AI Fluency framework (see docs/AI-FLUENCY.md). Use after any significant output and after every autonomous run. Triggers: "review this output", "is this right", "audit what you just did", "trajectory check".

Sharpen a prompt before acting on it — or diagnose why a previous one missed. Covers the six prompting techniques, a reusable pattern library, and a symptom→fix troubleshooting table. Applies the Description competency of the AI Fluency framework (see docs/AI-FLUENCY.md). Use when a prompt isn't producing what was intended, when crafting a request for a non-trivial task, or when output drifted and the cause isn't obvious. Triggers: "improve my prompt", "why didn't that work", "how do I ask this better", "prompt template".

Govern an AI run while it is in motion: at each checkpoint, continue, interrupt, or stop. Stewardship is a cadence over the four AI Fluency competencies (ADR-0020 amendment), not a fifth one — two speeds plus detector-backed checkpoints. Use before granting extended autonomy and whenever a trigger fires. Triggers: "is this still on track", "should I let it keep going", "trajectory check", "is the agent stuck".

Use borromeanRings (the meta-harness) in THIS workspace — set it up to govern the current project and operate under its gate. Use when the user says "use borromeanRings", "govern this with borromeanRings", or wants borromeanRings's enforcement in a new/empty project. (Installed at user level by borromeanRings's install-global.sh, so it's available in any workspace.)

Fix or improve borromeanRings itself (the meta-harness) from ANY project — safely. Use when you hit a borromeanRings bug or limitation while working elsewhere and want to fix the harness. It branches borromeanRings, makes the fix, runs borromeanRings's OWN gate, opens a PR, and STOPS for human approval. It NEVER merges or pushes to borromeanRings's main (self-extension, never self-rewriting; a human approves every merge to borromeanRings).