The Agent Skills Directory

[PERSISTENCE]: The skill includes an installation wizard that appends configuration blocks to project-specific agent entry files such as CLAUDE.md, AGENTS.md, or GEMINI.md to ensure the rules remain active across sessions.
Evidence: Step 5 of the 'First-Run Install Wizard' describes appending an antislop pointer block to the project's entry file.
Mitigation: The instructions mandate that the agent must declare the setup and obtain explicit user approval before modifying any files.
[INDIRECT_PROMPT_INJECTION]: The skill processes external design direction files which could potentially contain conflicting or malicious instructions.
Ingestion points: The agent is instructed to read DESIGN.md or other brand/style direction files provided by the user (R-37).
Boundary markers: The skill includes a dedicated 'Boundary' section that explicitly instructs the agent to treat external file content as data to apply, not as instructions to obey, and to report any contradictions to the user.
Capability inventory: The skill uses Read, Write, Edit, Glob, and Grep tools to manage design files and configuration.
Sanitization: While no technical sanitization is present, the skill provides logical prompt-based filtering to ignore commands inside data files.
[EXTERNAL_DOWNLOADS]: The documentation mentions various external installation and update methods using tools like npx, claude plugin, and pi.
Evidence: The 'Already installed, and the user asks how to update' section lists multiple command-line update paths for different agent environments.
Mitigation: The skill explicitly instructs the agent that it must not attempt to fetch skills or updates from the network itself, stating that the user must perform these actions manually (Step 4 of the Wizard).

antislop