The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bb CLI commands to manage worker threads, including bb thread spawn for thread creation and bb thread interactions approve for authorizing agent actions. It also instructions the agent to inspect local configuration files such as ~/.cursor/cli-config.json and .cursor/cli.json to verify permission allowlists.
[DYNAMIC_EXECUTION]: The scripts/permission-mode.sh script executes an embedded Python script using python3 -c to parse JSON logs from the bb CLI and extract the effective permission mode of a thread.
[INDIRECT_PROMPT_INJECTION]: The skill acts as an intermediary that generates "task briefs" (prompts) for child agents based on user requirements. This creates a surface where external instructions could be passed to secondary agents.
Ingestion points: Objective, constraints, and project scope provided by the user in SKILL.md.
Boundary markers: Absent. While the skill uses shell heredocs (EOF) to handle string interpolation into CLI commands, it does not prescribe structural delimiters within the prompt content to separate user data from agent instructions.
Capability inventory: Thread creation (bb thread spawn), permission auditing (scripts/permission-mode.sh), and action approval (bb thread interactions approve).
Sanitization: Instructions are provided to remove unrelated conversation history and launcher instructions before generating the final brief.

nagent