The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The update command defined in SKILL.md fetches a shell script from a remote repository via https://raw.githubusercontent.com/odysseyalive/claude-enforcer/main/install using curl.
[REMOTE_CODE_EXECUTION]: The skill performs remote code execution in the update command by piping the content fetched from the remote URL directly into a shell interpreter using bash -c "$(curl -fsSL ...)".
[COMMAND_EXECUTION]: The skill dynamically generates and executes shell scripts used as 'enforcement hooks' (detailed in references/procedures/hooks.md). These scripts are stored within the skill's local directory and are configured to execute during specific tool invocations via the agent's Bash tool.
[COMMAND_EXECUTION]: The skill utilizes the TaskCreate, TaskUpdate, and TaskGet tools to orchestrate complex operations, including the spawning of specialized subagents for validation and research.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it ingests and processes untrusted data from other skill files and CLAUDE.md (documented in references/procedures/audit.md).
Ingestion points: Reads content from .claude/skills/*/SKILL.md and project root CLAUDE.md.
Boundary markers: The skill does not implement explicit delimiters to separate untrusted skill content from its own instructions.
Capability inventory: The skill has the capability to write files (Write, Edit), execute shell commands (Bash), and create sub-tasks (TaskCreate).
Sanitization: The procedures focus on verbatim preservation of directives, which may allow malicious instructions in processed files to influence the agent's behavior during audits or optimizations.

skill-builder