The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes response data from external targets, which constitutes a potential injection surface if the target is malicious. However, this is handled through structured reporting rather than direct prompt interpolation.
Ingestion points: The scripts/probe_secrets.py script performs GET requests to user-provided URLs and reads the response body and headers to verify findings.
Boundary markers: The script generates structured Finding objects. While it does not wrap evidence in explicit LLM delimiters, the data is categorized and scoped within a security reporting context.
Capability inventory: The skill is granted access to Bash(python3:*) and Bash(curl:*) as defined in SKILL.md.
Sanitization: The script uses regular expressions to fingerprint content (e.g., matching .env keys or private key headers). It does not explicitly sanitize the raw response body or headers before including them in the Finding object's detail and evidence fields.
[COMMAND_EXECUTION]: The skill executes a local Python script (probe_secrets.py) to perform network probes. This execution is limited to the skill's own distributed source code and is the intended primary function.
[DATA_EXFILTRATION]: No evidence of local data exfiltration or credential harvesting was detected. The script is designed to detect data exposure on remote systems and requires an --authorized flag for non-local targets as a safety mechanism.
[EXTERNAL_DOWNLOADS]: The skill performs HTTP GET requests to external targets to check for file existence and content. It does not download and execute remote scripts or binaries.

detecting-exposed-secrets-files