The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to download and execute a shell script (fdk-kit.sh) from the author's GitHub repository using shell process substitution (bash <(curl ...)). This method allows for immediate execution of remote code to initialize the development environment.
[EXTERNAL_DOWNLOADS]: The skill fetches development tools and the 'overstack' kit components from the author's remote infrastructure on GitHub (Rheinmir/setup) to set up a local sandbox.
[COMMAND_EXECUTION]: The skill uses various shell commands and Python scripts (e.g., medic.py, new-skill.py, sync-skill.sh, gh pr) for environment auditing, skill registration, and repository synchronization. It also generates interactive HTML documentation (fdk-problem-tree.html) containing embedded JavaScript and SVG rendering logic.
[PROMPT_INJECTION]: The skill processes local project files such as rule-registry and policy.yaml to manage framework rules. This creates an indirect prompt injection surface as the agent ingests data from these external files into its context without defined boundary markers or explicit sanitization logic. Mandatory Evidence Chain:
Ingestion points: Reads local configuration files (rule-registry, policy.yaml) and framework documentation.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present for the ingested data.
Capability inventory: Includes shell execution (bash), script execution (python3), file writing (HTML/JS generation), and repository management (gh).
Sanitization: No specific validation or escaping mechanisms are described for the processing of external project data.

fdk