The Agent Skills Directory

Indirect Prompt Injection Surface: The skill's primary function is to process and analyze untrusted content provided by users (offer letters). There are no explicit boundary markers or delimiters defined in the instructions to isolate this external content from the agent's core instructions, which could allow adversarial text within a document to influence the agent's behavior.
Ingestion points: Untrusted data enters the agent context via the offer letter file or description provided in the argument-hint and instructions.
Boundary markers: Absent; the instructions do not specify the use of tags or delimiters for user-provided text.
Capability inventory: The skill has the capability to write output files to a specific workspace directory (~/.claude/plugins/config/claude-for-legal/employment-legal/matters/).
Sanitization: Absent; there is no mention of escaping or validating the content of the offer letter before processing.
File System Access: The skill interacts with configuration and matter-tracking files located within ~/.claude/plugins/config/. This access is scoped to the plugin's own data directory for the purpose of maintaining jurisdictional rules and matter-specific context.
Consequential Action Guardrails: A safety gate is implemented for non-lawyer users, requiring explicit confirmation of attorney review before the agent provides a final 'Clear to send' recommendation, which helps mitigate the risk of automated legal errors.

hiring-review