The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from Slack threads and attachments to drive its triage and bug-fixing workflows. This content influences classification, ticket creation, and automated code generation.
Ingestion points: Slack threads and attachments are read in references/benny/skills/triage-issue-reports/instructions.md and references/benny/skills/reproduce-and-fix-issues/instructions.md.
Boundary markers: The skill implements strict coordination rules, immutable source thread coordinates, and parent preflights to ensure it only replies to the correct context.
Capability inventory: The automation can read/write to Slack, interact with issue trackers (e.g., Linear), drive application UI via a control adapter, and author code changes in the repository.
Sanitization: It uses a 'trusted triage marker' system where the fix workflow only proceeds if a specific marker (e.g., [benny:bug]) is posted by a configured trusted identity.
[COMMAND_EXECUTION]: The skill executes actions through a 'control adapter' to interact with the application UI and performs file-system operations to implement bug fixes.
Evidence: references/benny/skills/reproduce-and-fix-issues/instructions.md describes driving the target app through real UI actions (clicks, typing, etc.) and authoring bounded root-cause fixes.

cursor-setup-benny