Use Agent Hooks

[pedrofuentes]

Research the use of hooks to improve the Sentinel behaviour

[pedrofuentes]

Research summary (triage): keeping open as a future optional opt-in example, not core.

What agent hooks are. Hooks (e.g., Claude Code's PreToolUse / PostToolUse) are tool-specific lifecycle handlers that fire before/after a tool call and can validate, log, mutate, or block the action. They would enable stronger local ("Layer 3") enforcement — for example, a PreToolUse hook that blocks a git merge/git push to main unless a Sentinel-APPROVED marker bound to the current HEAD SHA exists.

Why not in the core template. This template is deliberately tool-agnostic — the same markdown is consumed by Claude Code, Cursor, Copilot, Windsurf, Aider, and Gemini. Hook formats and lifecycle events differ per tool (or don't exist at all), so shipping hook config in the core template would (a) couple it to one tool, (b) add new files / adoption complexity, and (c) only benefit a subset of adopters. The portable enforcement layers already shipped are Layer 1 (AGENTS.md instructions) + Layer 2 (branch protection, which blocks merges server-side regardless of tool).

Recommendation. Keep open. If pursued, deliver it as an optional, clearly-labeled opt-in example (e.g., under examples/hooks/: a Claude Code PreToolUse hook enforcing the Sentinel-approval marker), documented as tool-specific and never required for the core flow. This preserves the tool-agnostic guarantee while giving adopters on a supported tool a stronger local gate.