The Agent Skills Directory

[METADATA_POISONING]: The skill metadata and documentation links point to a domain (jeffallan.github.io) that is currently blacklisted by security scanners. Additionally, the core skill file (SKILL.md) has a negative reputation with file scanners.
[COMMAND_EXECUTION]: The skill requests access to the Bash tool and provides a comprehensive library of commands for system auditing, network exploitation, and infrastructure testing. This includes high-privilege operations such as modifying IAM policies and interacting with cloud provider APIs.
[DATA_EXFILTRATION]: The instructions include specific patterns and tools for reconnaissance (DNS enumeration, subdomain discovery) and credential harvesting (searching for API keys, tokens, and bash history). These capabilities facilitate the collection and potential exfiltration of sensitive information.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data (source code, tool outputs, and infrastructure configurations) while possessing high-privilege tools like Bash.
Ingestion points: Files accessed via Read, Grep, Glob, and Bash commands.
Boundary markers: None present; the skill does not use delimiters or instructions to ignore embedded commands in analyzed content.
Capability inventory: Full shell execution via Bash, file system access, and network operations through common security tools.
Sanitization: No evidence of input validation or escaping before data is processed by the agent or tools.

security-reviewer