The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md explicitly direct the agent to execute a local Python script (save_file.py) to save generated files, which represents an active command execution capability managed by the agent.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, including user-provided functional requirements and Dify workflow DSL (YAML), to generate code and determine target file paths.\n
Ingestion points: User requirements and Dify DSL YAML files processed according to instructions in SKILL.md.\n
Boundary markers: No boundary markers or instructions to ignore embedded commands are present in the skill's reference files or instructions.\n
Capability inventory: Arbitrary file system write access via the save_file.py script and dynamic Python code execution via the run_code tool documented in references/common/tools.md.\n
Sanitization: The save_file.py script performs no validation or path sanitization on the --path argument, making it vulnerable to directory traversal attacks via malicious user input.\n- [PERSISTENCE]: The save_file.py script enables writing content to arbitrary filesystem paths. An attacker could provide malicious instructions that trick the agent into writing to sensitive persistence-related locations, such as ~/.bashrc, shell profiles, or cron directories.\n- [DYNAMIC_EXECUTION]: The reference documentation in references/common/tools.md promotes the use of a built-in run_code tool for executing Python code. This poses a risk if the agent generates and executes code that has been influenced by adversarial user prompts.

veadk-skills