coordinating-work

Warn

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCEEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill dynamically loads the orca-cli and orchestration skills by name at runtime and derives its command set from external execution guides retrieved from an executable.
  • Evidence: SKILL.md specifies that the agent must "Load the installed orca-cli and orchestration skills by name and retrieve their version-matched guides from the executable selected for this session."
  • Impact: This creates a dependency on code that is not statically defined within the skill, allowing for potential logic shifts if the external guides or host executables are modified.
  • [INDIRECT_PROMPT_INJECTION]: The Coordinator role ingests and relays messages from external task workers and other autonomous coordinators, creating a surface for cross-agent instruction injection.
  • Ingestion points: Identified in SKILL.md ("Other Coordinators") and references/operate.md ("The merge checklist"), where the agent processes messages and reports from sub-agents.
  • Boundary markers: The skill includes logic to treat sub-agent messages as "claims and data" rather than instructions, requiring comparison against a human-approved "record."
  • Capability inventory: The skill can merge pull requests, close tracker issues, manipulate worktrees, and create new workspaces.
  • Sanitization: Instructions mandate quoting original operator words and performing literal comparisons of verification evidence to prevent interpretation bias.
  • [PERSISTENCE]: The skill utilizes a scheduled timer to maintain its operation and recover session state across interruptions.
  • Evidence: SKILL.md references a "timer that can wake the Coordinator's own session," and references/setup.md instructs the agent to "Arm the one check ... at the approved interval."
  • Impact: This functions as a persistence mechanism on the execution host, enabling the agent to resume activity automatically without direct user intervention.
  • [EXTERNAL_DOWNLOADS]: The skill documentation specifies a hard dependency on several external GitHub repositories for its core functionality.
  • Evidence: References to github.com/stablyai/orca, github.com/EveryInc/compound-engineering-plugin, and github.com/pbakaus/impeccable are provided in the compatibility section of the frontmatter.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 19, 2026, 01:21 AM