The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides instructions to utilize dynamic code execution surfaces and generates executable scripts at runtime based on plan requirements.\n
The instructions explicitly suggest using the platform's 'eval' surface for multi-call steps instead of standard tool invocations.\n
It orchestrates the creation and execution of specialized scripts, such as Playwright automation and terminal QA tools, by subagents.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from markdown work plans located in the .omo/plans/ directory, which can influence its orchestration logic and the behavior of spawned subagents.\n
Ingestion points: Reads work plan files (.omo/plans/*.md) and session state files (.omo/boulder.json).\n
Capability inventory: The skill can spawn agents with shell access, write product files (via workers), and perform network operations.\n
Boundary markers: There is no explicit use of delimiters or warnings to ignore instructions embedded within the processed plan data.\n
Sanitization: No sanitization or validation of the plan content is performed before it is used to dispatch worker tasks.\n- [COMMAND_EXECUTION]: The skill manages the execution of various system and network commands via its subagent burst mechanism.\n
It manages complex shell operations including git worktree manipulations, curl for API endpoint testing, and terminal management via tmux and xterm.js.

ulw-execute