The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on local CLI tools bv and br to perform extensive operations on the project's task data stored in the .beads/ directory. These tools can modify project state and generate executable shell scripts via the --emit-script flag.
[PROMPT_INJECTION]: The skill processes and outputs data from task titles and descriptions. This creates a surface for indirect prompt injection where malicious instructions embedded in a task could influence the agent's behavior during triage.
Ingestion points: Data enters the context through commands such as bv --robot-triage and br list --json which read from the local .beads/ directory.
Boundary markers: The tools utilize structured JSON output, providing a degree of separation between metadata and task content, though instructions within the content are not explicitly ignored by the agent.
Capability inventory: The agent has the ability to create, update, and close tasks using br, and can generate scripts using bv.
Sanitization: No explicit sanitization or filtering of task content is described in the documentation or provided recipes.
[COMMAND_EXECUTION]: The tool includes a feature (--emit-script) that generates shell scripts for automating task management. This represents a form of dynamic code generation where the agent might be instructed to execute these scripts at runtime.

beads-bv