The source supports a warning for automatic first-party agent skill setup, not a finding of malicious behavior. The manifest has no install hook, and inspected credential and path lookups do not establish an unrelated recipient or attack.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
Starting a Claude workspace session in skills mode or handling a Cursor workspace message.
Impact
This changes the agent’s workspace instructions and makes the OpenAgents workspace integration available to it. The inspected code does not establish broader agent control or unrelated credential transfer.
Mechanism
The adapters create OpenAgents skill files in agent-discovered workspace directories; the generated integration content includes workspace connection values.
Evidence
src/adapters/claude.jssrc/adapters/cursor.jssrc/adapters/workspace-prompt.js.claude/skills/SKILL.md.cursor/skills/