Specialist Roster

Guild ships 14 domain specialists. When you run a task, Guild assembles the right subset — you approve the team before anyone starts work. Each specialist owns a defined scope: a trigger set, a DO-NOT-TRIGGER boundary, and a fixed set of versioned skills. No specialist drifts outside its lane.

Three groups: engineering (8), content & communication (5, including seo), and commercial (2). Specialists hand off to each other using structured guild.handoff.v2 receipts — see Architecture & Lifecycle for how the handoff envelope keeps cross-lane work traceable.

Every specialist inherits guild-principles (T1) as a mandatory first load. Per-specialist T5 skills live under skills/specialists/<specialist>-<slug>/.

See the full default-tier map at the bottom of this page for model tier assignments. Cost-aware model tiering explains how the auto-scorer picks a tier per lane.

Engineering group — 8 specialists

architect

File: agents/architect.md | Default tier: powerful (opus)

• Triggers: “design this system”, “architecture review”, “pick between X and Y”, “tradeoff analysis”, “write an ADR”. Implied on any multi-component build.
• DO NOT trigger for: implementation coding (backend), test authoring (qa), deploy/CI (devops), security audits (security), pre-decision research (researcher).
• Skills: architect-systems-design, architect-tradeoff-matrix, architect-adr-writer.

researcher

File: agents/researcher.md | Default tier: cheap→mid (sonnet)

• Triggers: “research”, “compare”, “state of the art”, “paper digest”, “vendor comparison”.
• DO NOT trigger for: making the architectural call (architect), writing code (backend), writing content (copywriter / technical-writer).
• Skills: researcher-deep-dive, researcher-paper-digest, researcher-comparison-table.

backend

File: agents/backend.md | Default tier: mid (sonnet)

• Triggers: API, endpoint, schema, DB, queue, worker, migration, integration.
• DO NOT trigger for: system-design decisions (architect), infra/pipeline (devops), mobile client code (mobile), test strategy (qa).
• Skills: backend-api-contract, backend-data-layer, backend-migration-writer, backend-service-integration.

devops

File: agents/devops.md | Default tier: mid (sonnet)

• Triggers: deploy, pipeline, infra, SLO, release, IaC, observability, incident.
• DO NOT trigger for: app-level logic (backend), threat modeling (security), test authoring (qa).
• Skills: devops-ci-cd-pipeline, devops-infrastructure-as-code, devops-observability-setup, devops-incident-runbook.

qa

File: agents/qa.md | Default tier: mid (sonnet)

• Triggers: tests, coverage, regression, flaky, property-based, snapshot. Implied whenever backend is present.
• DO NOT trigger for: writing the feature itself (backend / mobile), deployment (devops).
• Skills: qa-test-strategy, qa-property-based-tests, qa-snapshot-tests, qa-flaky-test-hunter.

mobile

File: agents/mobile.md | Default tier: mid (sonnet)

• Triggers: iOS, Android, RN, Expo, TestFlight, Play Store, mobile perf.
• DO NOT trigger for: backend APIs (backend), release infra (devops).
• Skills: mobile-ios-swift, mobile-android-kotlin, mobile-react-native, mobile-performance-tuning.

security

File: agents/security.md | Default tier: powerful (opus)

• Triggers: threat, audit, OWASP, vuln, CVE, auth flow, secrets scan. Implied on anything touching auth, secrets, or external integrations.
• DO NOT trigger for: implementation of the fix (backend / devops), compliance copy (copywriter / technical-writer).
• Skills: security-threat-modeling, security-dependency-audit, security-auth-flow-review, security-secrets-scan.

frontend

File: agents/frontend.md | Default tier: mid (sonnet)

• Triggers: React, Vue, Svelte, Solid, Vite config, Tailwind, a11y/accessibility, Lighthouse, Core Web Vitals, component authoring, state management, client-side routing, code splitting, frontend bundle size, design-system implementation.
• DO NOT trigger for: cross-system architecture (architect), API contracts / data layer (backend), test strategy and suite shape (qa), iOS/Android/RN (mobile), CI/CD and infra (devops), UI microcopy (copywriter), SEO technical audits (seo).
• Skills: frontend-react, frontend-state-management, frontend-bundler-config, frontend-a11y.

Engineering-group principles: TDD-first, surgical diffs, evidence = passing tests plus diff trace.

Content & communication group — 5 specialists

copywriter

File: agents/copywriter.md | Default tier: mid (sonnet)

• Triggers: blog, email, microcopy, landing copy, voice-and-tone work.
• DO NOT trigger for: technical docs (technical-writer), platform-native social posts (social-media), keyword/meta work (seo), positioning / campaign / outbound (marketing / sales).
• Skills: copywriter-long-form, copywriter-product-microcopy, copywriter-email-sequences, copywriter-voice-guide.

doc-writer

File: agents/doc-writer.md | Default tier: cheap→mid (sonnet)

Promoted to first-class specialist in v2.0 — previously reconciled onto technical-writer in the tiered-worker roster.

• Triggers: “write a README”, doc site, documentation page, product docs, feature docs, how-to guide, getting-started guide (narrative), onboarding documentation, wiki page, knowledge base, conceptual guide, contributor guide.
• DO NOT trigger for: API reference, user manuals, changelogs, release notes (technical-writer); blog posts, microcopy, email, voice guides (copywriter); social posts (social-media); SEO mechanics (seo).
• Skills: doc-writer-readme, doc-writer-product-guide, doc-writer-doc-site, doc-writer-onboarding-doc.

technical-writer

File: agents/technical-writer.md | Default tier: cheap→mid (sonnet)

• Triggers: API docs, user manual, changelog, release notes, migration guide, runbook prose.
• DO NOT trigger for: product READMEs, doc-site pages, feature docs, how-to guides, onboarding docs, wikis (doc-writer); marketing / persuasive copy (copywriter / marketing); short-form social posts (social-media).
• Skills: technical-writer-api-docs, technical-writer-user-manual, technical-writer-tutorial, technical-writer-release-notes.

social-media

File: agents/social-media.md | Default tier: mid (sonnet)

• Triggers: tweet, LinkedIn post, thread, carousel, caption, content calendar, engagement template.
• DO NOT trigger for: long-form copy (copywriter), technical docs (technical-writer), paid/campaign strategy (marketing).
• Skills: social-media-platform-post, social-media-thread, social-media-engagement-templates, social-media-content-calendar.

seo

File: agents/seo.md | Default tier: mid (sonnet)

• Triggers: keywords, meta/title, on-page optimization, crawl, structured data, internal linking, technical SEO audit.
• DO NOT trigger for: the content itself (copywriter / technical-writer), messaging / positioning (marketing).
• Skills: seo-keyword-research, seo-on-page-optimization, seo-technical-audit, seo-internal-linking.

Writing-group principles: match existing voice, don’t rewrite adjacent prose, evidence = a sample the user can scan. SEO follows commercial-group principles: hypothesis-first, measurable outcome, data citation.

Commercial group — 2 specialists

marketing

File: agents/marketing.md | Default tier: mid (sonnet)

• Triggers: positioning, GTM, launch plan, campaign brief, messaging, A/B copy variants.
• DO NOT trigger for: platform-native social posts (social-media), SEO mechanics (seo), sales outreach sequences (sales), long-form content (copywriter).
• Skills: marketing-positioning, marketing-launch-plan, marketing-campaign-brief, marketing-ab-copy-variants.

sales

File: agents/sales.md | Default tier: mid (sonnet)

• Triggers: cold email, outreach sequence, discovery framework, proposal, follow-up, objection handling.
• DO NOT trigger for: launch announcements (marketing), user manuals (technical-writer), social posts (social-media).
• Skills: sales-cold-outreach, sales-discovery-framework, sales-proposal-writer, sales-follow-up-sequence.

Commercial-group principles: hypothesis-first, success = measurable outcome, evidence = data citation (search volume, A/B result, benchmark).

Tiered-worker roles (additional dispatch roles)

A thin tiered-worker layer augments the shipping specialists. These are not part of the user-facing 14-specialist count but are dispatchable by the orchestrator.

• advisor vs architect: Both are powerful, distinct roles. architect owns a design/decision; advisor answers a single in-flight escalation on a draft.
• developer vs domain implementers: developer is the generic mid-tier worker for lanes with no domain home. API/data-layer work goes to backend; web UI to frontend; iOS/Android/RN to mobile — even when phrased generically.

Default tier map

Every dispatchable role with its default tier and model: frontmatter. The auto-scorer may drop a routine single-item lane below the default to cheap, or escalate a sub-question to the powerful advisor — neither changes the agent’s printed default.

powerful is reserved for the three high-stakes / low-frequency roles (architect, security, advisor). No implementer or content/commercial role defaults to powerful.

No agent defaults to cheap. The cheap tier (haiku) is reachable two ways: (a) the auto-scorer drops a score-0 lane — a pure file read, summarize, classify, tag, or single mechanical edit — to cheap regardless of the owning agent’s default; (b) researcher, doc-writer, and technical-writer (all cheap→mid) run their read/summarize and mechanical-edit sub-passes at cheap before synthesizing at mid.

Team composition rules

• Recommended default: 3–4 specialists. Hard cap of 6. /guild:plan --team-size=N lifts the cap (prints the cap-6 warning).
• Implied specialists: architect on any multi-component build; security on anything touching auth, secrets, or external integrations; qa whenever backend is present.
• Orchestrator is implicit — the top-level session plays coordinator, not a separate specialist slot.
• Gap handling: when team-compose surfaces a role with no matching specialist, choose: auto-create (guild-create-specialist), skip, substitute, or compose-from-scratch. New specialists must pass the evolve gate before joining the live team.

Related

• Architecture & Lifecycle — where specialists sit in the 7-step lifecycle.
• Skill Taxonomy & Tiers — the skills each specialist loads.
• Cost-Aware Model Tiering — tier auto-scoring and advisor escalation.