The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill captures arbitrary user input and stores it for later processing in automated workflows.
Ingestion points: User-supplied suggestions are recorded directly into the .bkit/btw-suggestions.json file via the /btw {suggestion} command.
Boundary markers: There are no instructions to use delimiters, XML tags, or safety markers when storing or retrieving the suggestion text.
Capability inventory: The skill utilizes the Bash and Write tools for file management and triggers a separate skill-create workflow.
Sanitization: The skill lacks validation or escaping for the suggestion field. When suggestions are used as a 'seed' for the skill-create pipeline or processed during /btw analyze, a malicious suggestion containing prompt injection patterns could attempt to influence the agent's behavior or the generated output.

btw