The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to utilize the clay CLI for environmental checks and schema retrieval (e.g., clay whoami, clay workflows actions list). These commands are authorized vendor tools for the clay-run environment used to ensure integration accuracy.
[INDIRECT_PROMPT_INJECTION]: The skill processes job posting content (titles and descriptions) from external third-party sources like TheirStack and CPJ, creating a potential surface for adversarial content.
Ingestion points: External job posting data fetched via API 'arms' described in SKILL.md and references/arm-routing.md.
Boundary markers: The skill implements a strict verdict logic (Part A and Part B) and requires explicit measurement declarations to structure the agent's output.
Capability inventory: Shell command execution via the clay CLI and network data retrieval through platform plugins.
Sanitization: No explicit sanitization or escaping of the external job text is defined prior to processing.
[METADATA_POISONING]: The skill's frontmatter declares the impact as touches: read-only. However, the body instructions involve writing command output to /tmp/actions.json and provide a mechanism to save user-supplied identifiers to a local file, representing an inconsistency between metadata and actual behavior.

hiring-radar