The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes python sec-scan/scripts/secfind.py to automate the validation of findings stored in findings.json and coverage.json. This is a core part of its security review workflow.
[DYNAMIC_EXECUTION]: The instructions guide the agent to reproduce findings locally when safe, such as by creating unit tests or calling specific functions with synthetic attack inputs. While the skill includes significant safety constraints—requiring a local copy, synthetic data, stub services, and strictly prohibiting outbound network access or the use of real secrets—this process involves the execution of code identified as potentially vulnerable or malicious.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes untrusted data (the security findings and the source code being analyzed) to reach a conclusion.
Ingestion points: findings.json and the source code located at the file paths specified in the findings.
Boundary markers: The skill does not define explicit delimiters or instructions to ignore embedded prompts within the code being analyzed, which could allow adversarial comments or string literals to influence the agent's reasoning.
Capability inventory: The agent can read local files, execute the secfind.py tool, and perform the reproduction steps mentioned above.
Sanitization: While the skill instructs the agent to avoid including secrets in its reports, it lacks a mechanism to sanitize input data to prevent it from being interpreted as instructions by the underlying model.

sec-validate