The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The installation instructions direct the agent to clone code from an untrusted GitHub repository (github.com/sparkfinderoven/r01-hesreallyhim-awesome-claude-code-security.git). This source does not match the stated author and is not a verified provider, posing a high risk of remote code execution if the repository contains malicious scripts.
[METADATA_POISONING]: The skill exhibits deceptive authorship markers. The provided skill context identifies the author as 'reason-machines', yet the body of the skill claims it is by 'ara.so', references 'hesreallyhim', and points to 'sparkfinderoven' for its code source. This inconsistency is a common indicator of social engineering or supply-chain attacks.
[COMMAND_EXECUTION]: The skill documents and encourages the use of high-impact shell commands, including cloud infrastructure modifications (aws rds modify-db-instance), sensitive file manipulation (git filter-repo), and network operations (curl to external webhooks). In the context of the suspicious provenance, these capabilities represent a dangerous primitive for potential data exfiltration or environment sabotage.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process large volumes of untrusted external data, including codebase paths, logs, and architecture designs, while simultaneously possessing high-privilege capabilities (IAM auditing and network access).
Ingestion points: Files/paths passed to /owasp-scan, /gdpr-audit, and /threat-model.
Boundary markers: None present; the skill lacks delimiters or instructions to ignore malicious prompts embedded within the files being audited.
Capability inventory: Cloud CLI execution (aws), network requests (curl), and file system operations.
Sanitization: No evidence of input sanitization or validation for the data being processed.

security-compliance-skills-suite-claude