The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of tickets, specifications, and dependency graphs fetched via the bb-cli tool. This data is passed directly to sub-agents such as /implement and /diagnosing-bugs to perform code modifications and branch management.
Ingestion points: The skill resolves and snapshots project artifacts including "Spec", "tickets", and "tracker" data using bb-cli in the Prepare and Start phases.
Boundary markers: While the instructions reference an "attached full ticket," there are no explicit delimiters or system instructions provided to the sub-agents to ignore potentially malicious prompts embedded within the ticket or specification text.
Capability inventory: The skill possesses significant capabilities through its sub-agents, including local file system modification (implementation and diagnosis), git operations (branching and committing), and network-facing actions (authenticated push and pull request creation).
Sanitization: The workflow does not describe any sanitization or validation steps for the content of the tickets before they are interpreted by the sub-agents, relying on the model's internal safety filters.

orchestrate-implementation