The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides several wrapper scripts (submit.sh, submit-dag.py, cleanup.sh) that execute the pueue and pueued binaries. These scripts use safe execution patterns (bash arrays and subprocess lists) to pass user-defined commands to the pueue daemon for queuing.
[INDIRECT_PROMPT_INJECTION]: The skill processes task specifications and commands that may originate from untrusted data (e.g., a YAML DAG spec provided by a user). It mitigates common risks by using yaml.safe_load for parsing and using the -- delimiter to ensure arguments are not misinterpreted by the shell. As this is the core intended purpose of a task queue manager, it is considered a legitimate capability.
[DATA_EXPOSURE]: The documentation in references/remote-daemon.md mentions the shared_secret used by pueue for authentication. It correctly identifies this as a credential and provides manual instructions for the user to manage it, with no automated mechanisms in the skill's scripts to exfiltrate or hardcode such secrets.
[DYNAMIC_EXECUTION]: The submit-dag.py and wait.py scripts use the uv run pattern to manage their execution environment and dependencies (pyyaml) at runtime. This is a standard and verifiable practice for modern Python scripts.

pueue-job-queue