The Agent Skills Directory

[COMMAND_EXECUTION]: In scripts/tmux_run_codex_task.sh, the skill executes the codex exec command with the --dangerously-bypass-approvals-and-sandbox flag. This configuration explicitly disables the security sandboxing and approval mechanisms intended to protect the host system from unauthorized or malicious code.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it parses a user-provided plan.md file to generate instructions for its workers without any validation or sanitization.
Ingestion points: The plan.md file (referenced in SKILL.md) is used as the primary source for task scheduling and logic.
Boundary markers: There are no delimiters or "ignore previous instructions" warnings applied to the data ingested from the plan file.
Capability inventory: The skill possesses high-risk capabilities, including shell command execution, file system modifications, and the ability to run AI-generated code with sandbox bypasses enabled.
Sanitization: No sanitization, escaping, or schema validation is performed on the plan file content before it is transformed into executable prompts.

parallel-task-tmux