The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is designed to transmit local execution data (the failing command and error excerpts) to an external service named 'Jev' via the omh_jev_ask tool. The instructions include specific mitigations: requiring the agent to inform the user what data leaves the machine, obtaining explicit consent before the first transmission, and instructions to remove credential-like content from the payload rather than redacting it silently.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local environment, specifically system error logs and command output, which are then processed and sent to an external analysis tool. This creates a potential surface for indirect prompt injection if an attacker-controlled file or process generates a malicious error message designed to influence the agent's triage behavior.
Ingestion points: Error excerpts and command history are collected into the state object within SKILL.md (Procedure step 2) and references/full-contract.md (Required inputs).
Boundary markers: The data is encapsulated within a structured JSON state object before being passed to the tool.
Capability inventory: The skill uses the omh_jev_ask tool to perform external analysis and subsequently hands off to build-failure-triage or agent-debug based on the result.
Sanitization: The skill instructions in references/full-contract.md (Recovery Notes) explicitly direct the agent to check for and remove 'credential_like_content' from the state object before transmission.

omh-jev-failure-triage