The Agent Skills Directory

[METADATA_POISONING]: The skill documentation falsely asserts that the 'Betterleaks' tool was created by the same author as 'Gitleaks'. This is a deceptive claim intended to exploit the reputation of a trusted tool to encourage the use of an unverified third-party utility.- [COMMAND_EXECUTION]: The audit workflow requires the execution of multiple command-line tools against the local environment, including the untrusted 'betterleaks' utility which the skill suggests running with root-level repository access.- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to download and install 'Betterleaks' from an external, non-standard source (github.com/betterleaks/betterleaks), which introduces supply chain risks.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content from a repository, including agent instruction files like .cursorrules and CLAUDE.md. This creates a surface for indirect prompt injection, where malicious instructions in the audited code could influence the agent's behavior. Evidence: The skill specifically targets agentic risks in Step 4, noting that campaigns like 'TrapDoor' hide injection in these files, yet the skill itself lacks boundary markers or sanitization when processing these untrusted inputs.

security-audit