The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by generating architectural documentation (codemap.md) that is intended to be read and followed by agents. Specifically, Step 5 of the workflow instructs the agent to reference the root codemap.md in AGENTS.md, which is automatically loaded into the agent's context for every session. This creates a mechanism where malicious instructions embedded in a repository's documentation could influence agent behavior.
Ingestion points: The skill reads codemap.md files (SKILL.md, Step 4) and repository source code (SKILL.md, Step 2).
Boundary markers: Absent. There are no instructions for the agent to use delimiters or security headers when reading or aggregating these files.
Capability inventory: The skill possesses file read/write capabilities, local script execution (cartographer.py), and the ability to delegate tasks to sub-agents.
Sanitization: Absent. The skill does not validate or sanitize the technical summaries generated from the code.
[COMMAND_EXECUTION]: The workflow relies on executing a local Python script (cartographer.py) with command-line arguments (include/exclude patterns) derived from the agent's interpretation of the codebase. While these patterns are used for file filtering, they represent a point where repository-derived data influences the execution of a local script.

cartography