skills/q00/openclip/oc-orchestrator/Gen Agent Trust Hub

oc-orchestrator

Warn

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: MEDIUMPERSISTENCEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PERSISTENCE]: The skill describes a 'Self-improvement' workflow where the oc-toolsmith subagent authors new tools and saves them to a git-tracked toolbox/ directory. These tools are explicitly designed to persist across runs and are described as being 'reusable by every future run.'
  • [DYNAMIC_EXECUTION]: The skill facilitates the runtime generation and execution of code. When standard tools are insufficient, the oc-toolsmith agent writes new tool scripts which are subsequently executed via the oc toolbox run command.
  • [INDIRECT_PROMPT_INJECTION]: The skill implements a 'steering' system that allows external directives to influence subagent behavior.
  • Ingestion points: The orchestrator reads directives from the open_steering file (accessed via oc status) and project manifests.
  • Boundary markers: No markers or delimiters are specified; instructions mandate folding directives 'into the prompt' verbatim and require workers to 'honor steering over its own default judgment.'
  • Capability inventory: The skill has the capability to spawn subagents via the Agent tool and execute shell commands via the Bash tool.
  • Sanitization: The skill does not mention any sanitization or validation of the input retrieved from the open_steering source.
  • [COMMAND_EXECUTION]: The skill relies on the Bash tool to interact with the oc CLI suite for video ingestion, processing, and management of the dynamic toolbox.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 5, 2026, 10:34 PM
Security Audit — agent-trust-hub — oc-orchestrator