The Agent Skills Directory

[PROMPT_INJECTION]: No malicious prompt injection patterns, role-play attempts, or instruction overrides were detected. The skill contains 'Iron Laws' and security principles that reinforce rather than bypass safety guidelines (SKILL.md, agent-guardrails-checklist.md).
[DATA_EXFILTRATION]: No evidence of data exfiltration or unauthorized network operations was found. The skill mentions sensitive patterns (e.g., API key prefixes) exclusively in the context of detection rules and search criteria (security-review-checklist.md, claude-config-security.md).
[COMMAND_EXECUTION]: The skill utilizes the Bash tool primarily for executing grep-based security scans. The provided command templates for auditing workspace configurations are standard security practices and do not involve unsafe interpolation of user-controlled input (claude-config-security.md).
[SAFE]: Indirect Prompt Injection Surface:
Ingestion points: The skill ingests untrusted code and configuration files via the Read, Glob, and Grep tools during the scanning phase (SKILL.md).
Boundary markers: The instructions do not define specific delimiters for the ingested content, though evaluation is strictly scoped to identified checklist patterns.
Capability inventory: The skill has access to Bash, Read, Grep, and Glob tools to perform its auditing functions (SKILL.md).
Sanitization: There is no explicit sanitization of the scanned content; however, the agent's role is restricted to reporting findings based on predefined security criteria, which limits the impact of malicious content embedded in the audited files.
[CREDENTIALS_UNSAFE]: No hardcoded credentials were found. Placeholder strings and regex patterns used for secret detection are correctly identified as non-sensitive (security-review-checklist.md).

security-reviewer