The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's 'Autoresearch Mode' constructs and executes a shell command (omc autoresearch) by interpolating user-supplied strings for the mission and evaluator fields. This creates a command injection vulnerability if the input contains shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from two main sources to influence its behavior:
Ingestion points: User input via {{ARGUMENTS}} and AskUserQuestion, and codebase context retrieved by the explore sub-agent.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the ingested codebase context.
Capability inventory: The skill can write files (Write), execute shell commands (omc autoresearch), and trigger other high-privilege execution skills (omc-plan, autopilot, ralph).
Sanitization: There is no evidence of escaping or validation of external content before it is interpolated into prompts for the scoring model or command lines.
[DYNAMIC_EXECUTION]: The skill uses the Skill() tool to dynamically transition to other agent skills such as omc-plan, autopilot, or team based on runtime decisions and user selection.

deep-interview