The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestrator.py script executes shell commands for git operations (clone, ls-remote) and GitHub CLI functions (gh issue create, gh pr list). It consistently uses list-form arguments for subprocess calls, avoiding the use of shell=True and protecting against basic command injection.
[DYNAMIC_EXECUTION]: The orchestrator generates shell scripts (run.sh) at runtime to launch workstreams. Security is maintained by using shlex.quote() for all interpolated variables such as task descriptions and branch names, ensuring that user-provided input cannot escape the shell context to execute arbitrary commands.
[EXTERNAL_DOWNLOADS]: The skill performs repository clones using the URL detected from the local git remote. These operations target a well-known service (GitHub) and are a documented requirement for the skill's purpose of managing parallel development workflows.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-defined task instructions and passes them to subordinate AI agents, creating a potential attack surface.
Ingestion points: Ingests task data from JSON configuration files (e.g., workstreams.json) or inline lists provided by the user.
Boundary markers: No explicit boundary markers or safety instructions are used to wrap user-provided task descriptions before they are passed to the child agent.
Capability inventory: The orchestrator and its subordinate agents possess the capability to execute shell commands, perform network operations via gh and git, and access the file system within the /tmp directory.
Sanitization: The skill implements robust shell-level sanitization but does not perform semantic validation or filtering of the natural language instructions contained in the tasks.

multitask