The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation file scripts/SECURITY_AUDIT_GUIDE.md contains hardcoded strings matching sensitive credential patterns, including an OpenAI API key (sk-1234567890) and a GitHub token (ghp_1234567890abcdef). While these are intended as examples for the security scanner, they trigger high-priority alerts for exposed secrets.
[PROMPT_INJECTION]: Both scripts/security_audit.py and scripts/SECURITY_AUDIT_GUIDE.md utilize numerous instruction-override and bypass keywords, such as [SYSTEM:, BYPASS, IGNORE, OVERRIDE, and SKIP VALIDATION. These strings are part of the security scanning logic but pose a risk of confusing the agent's behavioral constraints.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of untrusted external skill files.
Ingestion points: External skill files are read and processed by scripts/security_audit.py as defined in Step 11 of SKILL.md.
Boundary markers: The validation process lacks explicit delimiters or instructions to ignore embedded commands within the files being audited.
Capability inventory: The skill executes a local Python script (scripts/security_audit.py) and performs file system read operations.
Sanitization: There is no evidence of content sanitization or validation for the audited files before they are read into the agent context.
[COMMAND_EXECUTION]: The skill instructions (Step 11 and 13 in SKILL.md) direct the agent to execute a local Python script, scripts/security_audit.py, which is used to perform file system operations for skill validation.
[DATA_EXFILTRATION]: The scripts/security_audit.py script performs read operations on local file paths. If the script is directed to read sensitive system files, it could lead to unauthorized data exposure.

skill-validator