The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs file system modifications, including copying a bootstrap pack to a local directory and editing the .cursor/settings.json file to enable the 'pstack' plugin.
[PROMPT_INJECTION]: The skill sets up automations that process untrusted external data from Slack reports and attachments, creating a potential surface for indirect prompt injection.
Ingestion points: Slack message threads, metadata, and media attachments processed by the 'benny-triage' and 'benny-reproduce' automations.
Boundary markers: The setup references a 'trusted triage marker' to gate reproduction steps, though specific delimiters for the external content are not explicitly defined in these instructions.
Capability inventory: The resulting automations are granted permissions to read/write Slack threads, interact with issue trackers, and open pull requests.
Sanitization: Instructions demand 'repro proof' and a 'fail closed' approach for missing capabilities or verification failures.
[DATA_EXPOSURE]: The skill manages identifiers like Slack channel IDs and repository URLs. It explicitly warns against hardcoding secrets in prompts or committed configurations and suggests using secret managers for API tokens.

setup-benny