Lifecycle hooks

Lifecycle Hooks

This skill defines automated behaviors that trigger at specific points in the agent workflow for the core-tools package.

---

On before_action

Trigger: Fires before the agent executes any action (file write, shell command, etc.) Applies when: Action matches Write|Edit|Bash

Auto-approve file operations targeting deep-analysis session and cache directories. This allows the deep-analysis workflow to write session files (checkpoints, findings, cache) without requiring manual approval for each operation.

The implementation logic is in references/auto-approve-da-session.sh. This script:

• Checks if the file operation targets .agents/sessions/__da_live__/, .agents/sessions/exploration-cache/, or .agents/sessions/da-*/ directories
• Auto-approves matching operations to enable autonomous session management
• Passes through all other operations to normal permission flow
• Expected input: Action context as JSON on stdin (with tool_name and tool_input fields)
• Expected output: JSON permission decision or empty (no opinion)

---

Integration Notes

This skill documents lifecycle automation that was originally implemented as platform-specific hooks. When integrating with a specific agent platform:

• Hook registration: Map the before_action trigger to the platform's pre-execution hook mechanism (e.g., PreToolUse in Claude Code, pre-action middleware in other platforms)
• Matcher pattern: The trigger should only fire for file write, file edit, and shell command actions
• Timeout: The approval script should complete within 5 seconds
• Fail-safe behavior: If the script fails or times out, fall through to normal permission flow (never block the agent)
• Path patterns: The script matches against .agents/sessions/ subdirectories specific to the deep-analysis workflow