agentsclimarketplace

Lifecycle hooks

Skill sequenzia/agent-alchemy/ported/20260310/nested/core-tools/skills/lifecycle-hooks

Behavioral rules and lifecycle event handlers for the core-tools package. Includes auto-approval of file operations targeting deep-analysis session directories.From its SKILL.md

Install
npx -y skills add sequenzia/agent-alchemy --skill lifecycle-hooks

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

3.3 KB, 594 tokens by cl100k_base, as published. Nobody here has run it

Lifecycle Hooks

This skill defines behavioral rules and lifecycle event handlers converted from the core-tools hook configuration. These rules describe when certain operations should be automatically approved or handled without user intervention.


Auto-Approve Deep Analysis Session Operations

Trigger: Before any file write, file edit, or shell command execution.

Matcher: Operations targeting Write, Edit, or shell commands.

Behavior: Automatically approve file operations that target deep-analysis session and cache directories. All other operations pass through to the normal permission flow.

Approved Paths

The following directory patterns are auto-approved for file operations:

PatternDescription
.agents/sessions/__da_live__/*Active deep-analysis session files
.agents/sessions/exploration-cache/*Exploration cache (cached analysis results)
.agents/sessions/da-*/*Archived deep-analysis sessions

Logic

  1. For file write/edit operations: Extract the target file path. If it matches any of the approved path patterns above, approve the operation automatically.
  2. For shell commands: Check if the command string contains any of the approved directory paths. If so, approve automatically.
  3. For all other operations: No opinion — let the normal permission flow handle it.

Implementation Reference

See references/auto-approve-da-session.sh for the reference shell implementation of this auto-approval logic. The script:

  • Reads tool input as JSON from stdin
  • Extracts the tool name and relevant parameters (file path or command)
  • Matches against the approved path patterns
  • Returns an approval decision or exits silently (no opinion)
  • Never exits with a non-zero status code (a non-zero exit would break the permission flow)

Debug Logging

The reference implementation supports optional debug logging:

  • Set environment variable AGENT_ALCHEMY_HOOK_DEBUG=1 to enable
  • Logs are written to the path in AGENT_ALCHEMY_HOOK_LOG (default: /tmp/agent-alchemy-hook.log)

Integration Notes

What this component does: Defines lifecycle event handlers that auto-approve file operations targeting deep-analysis session directories, enabling autonomous execution during analysis workflows without manual permission grants.

Capabilities needed: Pre-action hook/interceptor system that can inspect file operation targets and approve them programmatically. The host platform must support a way to intercept file writes and shell commands before execution.

Adaptation guidance: The auto-approval logic is straightforward path matching. On platforms without a hook system, this can be implemented as: (1) a pre-configured allow-list of directory patterns, (2) a middleware that checks file operation targets, or (3) a configuration setting that grants write access to session directories. The reference shell script in references/ shows the exact matching logic.

Configurable parameters: None (the approved paths are tied to the deep-analysis session directory structure).

What ships with it: 1 file

2.9 KB alongside SKILL.md, 1 of them executable

references/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.