The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze untrusted tool source code. This creates a potential surface for indirect prompt injection if the audited tool contains instructions designed to deceive the agent. The skill mitigates this risk by providing an adversarial reasoning framework, explicitly instructing the agent to assume the input is malicious and to use a 'mechanical gate' (a static scan) to detect common danger patterns.
Ingestion points: oc --project <P> toolbox show --name <tool> (reading external tool source).
Boundary markers: The instructions explicitly state: 'Assumes the tool is broken or malicious until proven otherwise.'
Capability inventory: The skill uses Bash to execute oc CLI commands.
Sanitization: The process relies on a mechanical gate (oc toolbox promote) that performs static deny-list scans for sensitive operations.
[COMMAND_EXECUTION]: The skill uses the Bash tool to interact with the oc command-line utility. This execution is limited to specific administrative functions (show and promote) related to the tool registry management.

oc-tool-auditor