The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The AZ Skill Feedback section provides a shell command npx --yes hivenet@latest which downloads and executes code from a remote registry at runtime.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing external tools gh and ghstack which are required for the autonomous workflow.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user input via $ARGUMENTS to generate sub-goal and roadmap files that are subsequently used as authoritative instructions in an autonomous loop.
Ingestion points: User-provided objectives and context passed through $ARGUMENTS.
Boundary markers: The generated files use standard Markdown formatting but lack explicit markers or instructions to ignore potential nested commands in the user content.
Capability inventory: The autonomous loop (driven by the generated files) has the capability to write files, manage GitHub PRs via gh, and execute stack operations via ghstack.
Sanitization: There is no evidence of validation or escaping for user-supplied data before it is written into instruction files.
[COMMAND_EXECUTION]: The skill utilizes system commands like which, wc, cat, and pbcopy to manage the planning environment and validate prompt lengths.

plan-for-mega-goal