The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent and user to clone a repository from https://github.com/EastSword/skill-dfyx_code_security_review.git, which is an external and untrusted source.
[REMOTE_CODE_EXECUTION]: The installation process involves pip install -r requirements.txt, which downloads and installs unverified packages from external registries.
[COMMAND_EXECUTION]: The skill relies on executing various standalone Python scripts provided in the repository (e.g., scripts/code_scan.py, scripts/data_flow_analyzer.py) to perform its primary functions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external source code, which can be used as a vector for indirect prompt injection attacks against the agent.
Ingestion points: External codebase files located at the user-provided /path/to/project path.
Boundary markers: None mentioned; the skill does not explicitly use delimiters or instructions to ignore embedded commands in the audited code.
Capability inventory: The skill has the capability to execute shell commands, write files (reports), and perform multi-phase analysis using subprocesses.
Sanitization: No sanitization or validation of the ingested code content is documented in the skill instructions.
[DYNAMIC_EXECUTION]: The skill documentation describes features for 'POC generation and verification' and 'Writing source then compiling', which involves generating and executing code dynamically at runtime.

dfyx-code-security-review