The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process external plan files, which presents a surface for indirect prompt injection if those files contain malicious instructions.
Ingestion points: The skill reads work plans from the .omo/plans/ directory and session state from .omo/boulder.json.
Boundary markers: The agent relies on markdown structural elements (checkboxes, headers) to parse instructions; there are no explicit delimiters to ignore instructions embedded within the plan data.
Capability inventory: The orchestrator can spawn subagents with broad implementation permissions and execute shell commands via several CLI tools.
Sanitization: No sanitization or validation of the plan file content is performed before task delegation.
[COMMAND_EXECUTION]: The skill explicitly utilizes shell commands to manage the development environment and perform quality assurance.
Uses git worktree add to create isolated work environments for pull request preparation.
Executes curl -i for endpoint verification during manual QA stages.
Invokes node script/qa/web-terminal-visual-qa.mjs to capture visual terminal artifacts using xterm.js.
[EXTERNAL_DOWNLOADS]: The skill references an external repository for enhanced browser automation capabilities.
Recommends the use of the agent-browser utility from Vercel Labs' official GitHub repository as a fallback when standard browser tools are insufficient.

start-work