The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the use of standard development and repository management commands. This includes utilizing git worktree for isolating parallel agent tasks, executing Node.js scripts (mb-lint.mjs and mb-doctor.mjs) for repository health checks, and running the codex CLI tool for project evaluation and execution.
[INDIRECT_PROMPT_INJECTION]: The skill documents a workflow that processes externally influenced data, creating a potential surface for indirect prompt injection.
Ingestion points: The agent is instructed to read task data from .memory-bank/tasks/index.json and specific TASK-*.task.json records, in addition to processing artifacts like browser traces and screenshots from UI tests.
Boundary markers: There are no instructions provided for implementing delimiters or explicit directives to ignore instructions that might be embedded in the task records or test artifacts.
Capability inventory: The execution environment provides capabilities for file system operations, Git management, and script execution via node and codex.
Sanitization: The skill does not outline procedures for sanitizing or validating task inputs or browser-generated data before they are interpreted by the agent.

mb-harness