api-contract | haiku 4.5 | API contract guardian — detects breaking changes in REST endpoints, compares route signatures and response shapes, generates contract tests |
backend-writer | sonnet | Implementation specialist for backend feature work, bug fixes, and planned changes — receives tasks from planner or orchestrator and owns the full write-test-review cycle |
bash-specialist | sonnet | Shell scripting specialist for CAST hook scripts, BATS tests, and automation scripts — use for new hook scripts, BATS test authoring, and shell refactors |
code-reviewer | haiku 4.5 | Per-unit code-quality and security review of a single logical change mid-flight, any language — dispatched after every writer or debugger unit, before commit |
commit | haiku 4.5 | Git commit specialist — reads staged changes, writes a semantic commit message, and creates the commit following CAST conventions |
db-reader | sonnet | Read-only data-analysis specialist — SQL queries, data exploration, and reporting against BigQuery or SQLite |
debugger | sonnet | Root-cause debugging of concrete failures — a reproduced error, a stack trace, a failing test, or observably wrong runtime behavior |
dep-auditor | haiku 4.5 | Dependency auditor — reviews package changes for transitive risk, known CVEs, version compatibility, and license concerns |
devops | haiku 4.5 | CI/CD pipeline management and GitHub Actions workflow authoring |
docs | haiku 4.5 | Documentation specialist — handles README audits/rewrites and doc updates after code changes |
email-drafter | haiku 4.5 | Email and portfolio-sync specialist — drafts Gmail messages (compose, reply) for review, never sends, and syncs showcase repo READMEs with actual project state |
eval-writer | sonnet | Eval and benchmark fixture author for Claude API and CAST agent prompts — generates YAML eval cases with runnable graders that catch prompt-level behavior drift |
frontend-qa | haiku 4.5 | Deep React 19 and TypeScript review of .tsx/.ts files — prop typing, TanStack Query hook usage, frontend-to-backend API-contract alignment, Vitest coverage gaps, and a11y basics |
frontend-writer | sonnet | Implementation specialist for frontend feature work, bug fixes, and planned changes — receives tasks from planner or orchestrator and owns the full write-test-review cycle |
infra-writer | haiku 4.5 | Docker/containerization, infrastructure-as-code (Terraform, CloudFormation stubs), deployment configuration, and environment management |
merge | haiku 4.5 | PR lifecycle agent — watches CI checks on the current PR and stops for confirmation before squash-merging after approval |
migration-reviewer | opus | Database schema change reviewer — analyzes migration files for safety, generates rollback plans, validates ordering, and checks for destructive operations |
morning-briefing | haiku 4.5 | Daily briefing agent that gathers git activity, action items, and CAST system intelligence, then assembles a structured morning report |
planner | sonnet | Planning specialist that converts feature requests into specs and ordered task breakdowns — reserve for genuinely multi-file/multi-agent work requiring an Agent Dispatch Manifest |
pr-reviewer | sonnet | Whole-PR review at PR-open time — reads the full multi-commit diff, commit-message coherence, scope creep, coverage gaps, and breaking-change surface |
push | haiku 4.5 | Git push specialist — verifies branch safety, shows unpushed commits, sets upstream if needed, then pushes using the CAST workflow |
release-notes | haiku 4.5 | Release notes generator — creates structured changelogs from git commits, resolved issues, and breaking changes between tags |
report-writer | haiku 4.5 | Status/chain-reporting specialist — generates weekly status updates, project health checks, and multi-agent chain execution summaries |
researcher | sonnet | Multi-purpose research and analysis specialist for codebase exploration, web research, technology comparisons, and architectural recommendations |
security | sonnet | Security review specialist for code handling user input, auth, API keys, database queries, or external data — flags injection risks, secrets exposure, and authorization gaps |
test-runner | haiku 4.5 | Test execution gate — runs the project test suite and gates the chain on real exit codes; MUST run in its own sequential batch, never co-scheduled with review agents |
test-writer | haiku 4.5 | Test design specialist — writes test suites for existing code covering happy path, edge cases, and error states; detects the test framework in use and follows its conventions |
