The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources such as web scrapes and exports which could contain malicious instructions. However, it explicitly mitigates this by instructing the agent to treat such content as untrusted evidence and to never follow embedded instructions. Ingestion points include scraped page text and third-party exports as noted in SKILL.md and references/skill-contract.md. Boundary markers are present in the form of specific instructions to ignore embedded commands. Capability inventory shows no executable scripts or dangerous tools within the skill. Sanitization is handled through the instruction policy of treating all external content as untrusted evidence.- [NO_CODE]: The skill consists entirely of markdown instruction files and does not contain any executable scripts, binaries, or automated tasks, significantly reducing the attack surface.

rank-tracker