subagent-conflict-detection
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions require the agent to ingest and analyze the task prompt of a planned subagent to determine its intended file scope.
- Ingestion points: The content of the subagent's task description (specifically sections like ## Task scope and ## Inputs).
- Boundary markers: The instructions do not define strict delimiters or ignore-instructions warnings for the processed text.
- Capability inventory: The agent possesses repository access via git tools (git worktree, git status, etc.).
- Sanitization: There is no specified logic for sanitizing the content extracted from the subagent prompt before it is used for decision-making.
- [COMMAND_EXECUTION]: The skill employs bash pipelines and subshells to query git worktree status.
- Evidence: Execution of pipelines involving git worktree list and subshell commands within different worktree directories.
- [EXTERNAL_DOWNLOADS]: The skill involves network synchronization and references external documentation.
- Evidence: Uses git fetch origin to update branch state and provides links to official guidance on code.claude.com (a well-known service).
Audit Metadata