The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow where user-defined requirements and feature specifications are collected and then interpolated into the system prompts of subordinate Worker and Validator agents.
Ingestion points: User input provided during the Intake phase (Phase 1) is stored in local YAML and JSON files.
Boundary markers: Templates in the references/ directory use markdown headers to isolate user-provided data from core agent instructions, though they lack explicit "ignore embedded commands" delimiters.
Capability inventory: The system can spawn sub-agents using the platform's Agent tool, perform git branch operations, and execute shell commands through the mission.sh library.
Sanitization: The skill relies on structured templates and the orchestrator's oversight rather than programmatic string sanitization of input.
[COMMAND_EXECUTION]: The mission.sh library automates repository management and state tracking. It uses standard shell utilities such as mkdir for directory scaffolding, grep and sed for YAML state updates, and git for branch management and merging feature work into the base branch.
[EXTERNAL_DOWNLOADS]: The documentation suggests the installation of yq for YAML processing and agent-browser (a tool from Vercel) for automated UI validation. These are established, well-known utilities within the developer ecosystem and are treated as safe dependencies.

missions