The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary repository files to generate architectural summaries and codemaps. This creates a surface where malicious instructions hidden in source code could be incorporated into the generated codemap.md files. These generated maps are subsequently used to guide the behavior of other agents, potentially leading to the execution of attacker-controlled logic.\n
Ingestion points: The scripts/codemap.mjs script reads file paths and contents (via walkFiles and readFileSync) to compute hashes, while Fixer agents (referenced in SKILL.md) read the repository source code to generate descriptive technical documentation.\n
Boundary markers: The skill does not implement delimiters or safety instructions within the generated codemap.md templates to prevent agents from following instructions found within the repository code.\n
Capability inventory: The skill possesses the ability to execute shell commands (running its own Node.js script), write to the filesystem, and modify existing project files like AGENTS.md.\n
Sanitization: There is no evidence of filtering or sanitization of the content extracted from repository files before it is summarized into the project's documentation.\n- [COMMAND_EXECUTION]: The skill workflow involves executing a local Node.js script (scripts/codemap.mjs) via the command line. While the script is internal to the skill and primarily performs filesystem operations (hashing and file writing), the LLM provides arguments that control the scope of recursive repository scanning.\n- [PERSISTENCE]: The skill instructs the agent to modify or create an AGENTS.md file in the repository root. This modification ensures that the generated codemap is automatically injected into the agent's context for all future sessions in that repository, effectively persisting the skill's influence and the results of its code analysis across the project lifecycle.

codemap