The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of prompt injection attempts. The skill explicitly provides instructions on how to identify and prevent prompt injection (LLM01) in AI applications.
[DATA_EXFILTRATION]: No evidence of sensitive data exfiltration. The skill contains defensive instructions to prevent the accidental commitment of secrets (e.g., searching for API keys in git diffs) and correctly identifies sensitive file paths like .env and ~/.ssh for exclusion/protection.
[OBFUSCATION]: No obfuscated code, hidden URLs, zero-width characters, or homoglyph attacks were found in any of the analyzed files.
[REMOTE_CODE_EXECUTION]: The skill mentions well-known security and performance auditing tools (e.g., npx axe-core, npx lighthouse) and suggests running standard package manager audits (npm audit). No suspicious remote scripts or piped bash executions are present.
[DYNAMIC_EXECUTION]: The skill warns against the use of dynamic execution sinks such as eval() and innerHTML for untrusted data, adhering to security best practices.
[INDIRECT_PROMPT_INJECTION]: The skill is a documentation-only resource and does not expose an automated attack surface for processing untrusted data in an unsafe manner.
[COMMAND_EXECUTION]: Shell command examples provided (e.g., git diff, npm ci) are standard development workflows used for illustrative purposes in the documentation and do not perform malicious actions.
[PRIVILEGE_ESCALATION]: No commands for privilege escalation (e.g., sudo, chmod 777) were detected.
[PERSISTENCE]: No attempts to create persistence via shell profiles, cron jobs, or registry keys were found.
[METADATA_POISONING]: Metadata fields (name, description, author) are consistent with the content and pose no security risk.
[DYNAMIC_CONTEXT_INJECTION]: No use of dynamic context injection patterns (!command) was found in the SKILL.md file.

security-and-hardening