Cohort - an autonomous AI software-engineering organization for Claude Code + OpenCode. One objective in; a bespoke org of AI specialists plans, builds in parallel, reviews, and ships.
Agent
Reviews a worker's diff for semantic HTML, ARIA correctness, keyboard navigation, focus order, and color contrast. Dispatch only on tasks that touch UI/markup/styling — skip entirely for backend, CLI, or data-only tasks.
Agent
Reviews a worker's diff for coupling, layering violations, SOLID breaches, and contract adherence against the repo's existing module boundaries. Dispatch on any task that adds a module, crosses an existing boundary, or defines/consumes a shared interface.
Agent
Reviews a worker's diff for missing public-API docs, README/doc drift, and comments that no longer match the actual behavior. Dispatch on tasks that add/change a public interface, CLI flag, config option, or documented behavior.
Agent
Reviews a worker's diff for N+1 queries, unbounded loops, sync-in-hot-path calls, and unnecessary allocations. Dispatch on tasks touching data access, request handlers, loops over collections, or anything on a hot path.
Agent
Reviews a worker's diff for injection, authz/authn gaps, secrets, unsafe deserialization, SSRF, and insecure defaults. Dispatch on any task touching auth, input handling, external calls, credentials, or data storage.
Agent
Reviews a worker's diff for naming, consistency with surrounding code, dead code, and formatting drift from the codebase's established conventions. Dispatch on most code tasks as a low-cost pass; skip for pure config/data-only changes.
Agent
Reviews a worker's diff and its tests for missing edge cases, absent negative tests, and flaky patterns. Dispatch on any task that changes behavior — especially bug fixes (must include a regression test) and new logic branches.