Skip to content
mediaite-ghostink

ADR-003: Agent Governance and Hooks

Status

Section titled “Status”

Accepted - 2026-04-20

Context

Section titled “Context”

This repository is developed with AI-assisted workflows. Without explicit governance, changes can become non-deterministic, insufficiently validated, or too broad in scope.

Decision

Section titled “Decision”

Adopt governance artifacts as first-class project controls:

  • AGENTS.md for dev-mode authority and operating rules
  • AGENTS.staging.md for staging restrictions and validation gates
  • Prompt artifacts under prompts/core-agent/
  • Operational coordination through docs/TASK.md (internal queue) and HANDOFF.md
  • Hook/eval integration through .claude/ configuration and test execution

Consequences

Section titled “Consequences”
  • Positive:
    • Standardized execution and review behavior across sessions.
    • Lower risk of accidental architectural drift.
    • Better task continuity through documented handoffs.
  • Negative:
    • Documentation overhead increases with project changes.
    • Governance drift can occur if artifacts are not updated alongside code.

Follow-Up

Section titled “Follow-Up”
  • Keep governance files synchronized with implementation and CI checks.
  • Add explicit hook scripts and eval specs as enforcement depth increases.