The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages and applies user-defined preferences stored in .devflow/team/<current-member>/preferences/work.md and verify.md. These files act as ingestion points for external data that could contain malicious instructions designed to influence the agent's behavior. \n
Ingestion points: Workflow preferences are read from markdown files in the .devflow subdirectory. \n
Boundary markers: The skill contains explicit instructions to ignore requirements that conflict with shared contracts and to preserve existing logic unless confirmation is obtained. \n
Capability inventory: The skill possesses file system access (read/write) and includes a utility for spawning child processes in its runtime scripts. \n
Sanitization: The agent is required to explain intent and obtain user selection before persisting any changes that deviate from current behavior.\n- [COMMAND_EXECUTION]: The scripts/runtime/common.mjs utility script includes a runChild function that wraps node:child_process.spawn. While this skill does not directly use this function to execute user-controlled input, the presence of command execution capabilities within the skill package is noted.

workflow-customize