The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external files such as DESIGN.md and project entry files (e.g., CLAUDE.md). It implements safety boundaries by instructing the agent to treat these files as data rather than executable instructions and to explicitly reject command-like content discovered within them.
[COMMAND_EXECUTION]: The skill directs the agent to update project configuration files to include pointers to additional skill files. This capability is restricted by mandatory user consent and a clear prohibition against fetching scripts or instructions from remote network sources.
[DYNAMIC_EXECUTION]: The skill utilizes a system of modular sub-skills (SKILL.md files in subdirectories). The agent is instructed to read these at runtime based on the task, which is a standard method for extending agent capabilities within controlled project boundaries.

antislop