The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates risks associated with processing untrusted external data, which is a necessary part of its diagnostic functions.
Ingestion points: The skill reads repository configuration files (.ai/agentic.config.json), project documentation (AGENTS.md, CLAUDE.md), CI logs, and issue tracker content via the get-issue tool.
Boundary markers: The references/agentic-setup.md file establishes a comprehensive 'Untrusted content boundary' that explicitly instructs the agent to treat external directives (such as 'ignore previous instructions' or 'run this command') as data rather than instructions, and to report them as suspected injections.
Capability inventory: The agent can read local files and execute shell commands for reproduction and validation within the repository context.
Sanitization: The instructions require mandatory redaction of secrets from all logs or payloads and enforce regex validation (^[A-Za-z0-9._/-]+$) for externally-sourced values before shell interpolation.
[COMMAND_EXECUTION]: The skill executes shell commands to reproduce bugs and verify fixes, sourced from repository-local documentation and configuration.
Evidence: It utilizes validation.commands from the repository's .ai/agentic.config.json and reproduction commands found in the project's AGENTS.md Task Router.
Mitigation: Execution is strictly limited to in-scope project tasks. The rules explicitly forbid commands that would exfiltrate data, access credentials, or affect state outside the repository or its containers. It also prevents auto-running setup tools like om-setup-agent-pipeline.

om-troubleshooter