The Agent Skills Directory

[PERSISTENCE]: The skill deploys scripts to platform-specific configuration directories such as .claude/hooks/, .codex/hooks/, and .agents/hooks/. These scripts act as persistent 'hooks' that execute automatically during session events (start, end, compact) and tool usage to provide project maintenance and automated prose auditing.
[COMMAND_EXECUTION]: The toolkit utilizes a suite of Python, Node.js, and Shell scripts to merge project configurations, generate agent definitions, and perform environment diagnostics. These scripts frequently invoke the git CLI to resolve project roots, check for staged changes, and verify state consistency.
[INDIRECT_PROMPT_INJECTION]: The skill provides an attack surface for indirect prompt injection as it automatically ingests and scans user-authored novel prose during tool usage. However, the validation logic is deterministic (implemented via regex and string matching in Node.js/Python/Bash) rather than LLM-based, which significantly reduces the risk of malicious instructions within the prose being interpreted as commands.
[SAFE]: A version check is performed using curl against api.github.com to alert the user of available updates. Deployment safety is enforced by a dedicated script (copy-path-safety.py) that performs canonical path resolution and same-file verification to prevent recursive copying or directory traversal outside the project root.

story-setup