flow-next-work

Warn

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill constructs and executes shell commands using variables derived from user input, such as branch names, file paths, and task descriptions (e.g., git checkout -b <branch>, test -f "<path>", flowctl task create --title "Implement <idea>"). There is no evidence of sanitization for shell metacharacters like semicolons, pipes, or ampersands, which could allow for arbitrary command execution if malicious input is provided.
  • [REMOTE_CODE_EXECUTION]: The skill relies on a bundled executable script flowctl located within the plugin's own directory structure (${DROID_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/scripts/flowctl). Executing scripts provided with the skill increases the execution attack surface compared to standard system binaries.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface where it ingests untrusted data from user arguments or external markdown files and passes it into the context of 'worker' subagents.
  • Ingestion points: Phase 1 processes user arguments and local markdown files to determine task content.
  • Boundary markers: No specific delimiters or instructions are used to separate untrusted user data from the instructions provided in the subagent prompt template.
  • Capability inventory: The agent has access to the filesystem, network (via git), and the ability to execute shell commands and spawn additional agents with inherited permissions.
  • Sanitization: There is no evidence of content validation or escaping before interpolation into the subagent prompt template.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 13, 2026, 01:09 AM