The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies and processes untrusted content from the local project repository which can be manipulated by an attacker to influence the agent's behavior.
Ingestion points: Step 2 in SKILL.md reads content from package.json, requirements.txt, go.mod, Gemfile, README.md, TODOS.md, CLAUDE.md, and source code files (via grep for TODO comments).
Boundary markers: The skill does not implement delimiters or provide instructions to the agent to ignore potentially malicious commands embedded within these files.
Capability inventory: The skill utilizes Bash (ls, find, grep, curl) and python3 to process extracted data and perform network operations.
Sanitization: Content from the ingested files is used directly to construct node labels and descriptions for API requests without validation or sanitization.
[DYNAMIC_EXECUTION]: The skill utilizes python3 -c to execute script templates for parsing JSON and performing geometric calculations for the mind map layout.
Evidence: Steps 1, 2, and 3 in SKILL.md contain embedded Python scripts used to process file contents and generate structured data.
[COMMAND_EXECUTION]: The skill uses shell commands and curl to interact with a local service and explore the file system.
Evidence: Step 4 uses curl to check service availability, fetch existing nodes, and perform POST/PATCH operations on http://localhost:3002. Standard system tools like find, grep, and ls are used to map the project structure.

vibe-map