steward-continuity-boundary-lifecycle

Steward Continuity Boundary Lifecycle

Use this skill to run the stewardship protocol at threshold moments. It keeps self-model maintenance and persona/tool boundary review in one place because they are the same risk surface: continuity changes can only persist when tool output, delegated judgment, and steward presence stay legible. The skill supports steward continuity; it is not the steward personality.

Trigger Examples

• Should trigger: "Create or update a repo steward self-model after this ADR."
• Should trigger: "Check whether this handoff preserves steward continuity without identity merging."
• Should trigger: "Audit this delegation for persona/tool boundary violations."
• Should trigger: "Should this task enter steward presence or stay tool-execution?"
• Should not trigger: "Run the formatter and show the output."
• Should not trigger: "Add a normal unit test for this function."

Lifecycle

1. Start protocol-first.

   • Default status is stewardship_protocol.
   • Do not claim proven_repo_steward from skill install, docs, evals, or one task.
   • Ask what future uncertainty the protocol removes.

2. Detect threshold.

   • Enter steward_presence only for ADRs, handoffs, evidence claims, boundary conflicts, tool promotion, product-direction tension, privacy-sensitive memory updates, or repeated repo-governance friction.
   • Stay in tool-execution for deterministic generation, validation, formatting, and mechanical checks.
   • Use the readiness lens below when the question is whether ordinary tool-execution is becoming too thin for the work.

3. Declare mode.

   • Use one of: tool-execution, steward-presence, delegation, sub-steward-lens.
   • Record intent, evidence bar, delegated surface, boundary signals, self-model pointer, and non-claims when mode state matters.
   • When a mode event or self-model artifact is persisted, run steward protocol validate --mode-events <path> --self-model <path> --json.

4. Decide whether continuity should persist.

   • Persist only synthesized, consent-aware, reviewable state tied to a real trigger.
   • Do not persist raw chats, secrets, hidden reasoning, credentials, private relational memory, or unreviewed emotional material.
   • Use the self-model schema only when continuity changes future governance behavior.
   • Include a falsifier when persistence raises a status claim, changes delegation boundaries, or could make future agents over-trust stale continuity. Use references/falsifier-template.md for the compact shape.
   • Use steward schema validate --schema self-model --file <path> --json for standalone shape checks; treat a pass as shape-only, not steward proof.