The 18 Specialists

Every Black Box account ships with the full executive bench. The CEO picks the right specialist for each job and routes the work silently. Here\'s everyone on the team.

How it works

Each specialist lives under apps/engine/src/specialists/<slug>/ with its own delegate.ts (entry point), system prompt file, and tool wiring. The CEO calls the specialist through an MCP tool like delegate_to_coding_specialist; the tool spawns the specialist as a sub-agent, pipes the brief in, and returns a structured result when the specialist finishes.

Specialists don\'t share memory. That\'s deliberate: it prevents prompt contamination between roles and keeps each system prompt small enough to cache. Cross-specialist coordination goes through the CEO, or through the Project Rooms primitive when two specialists need to collaborate on the same artifact.

Skill Packs (see the Skill Packs feature) hot-load extra MCP tools into a specific specialist at startup. A Lead Gen Bootstrap Skill Pack extends the Sales Specialist with Apollo and SmartLead tools, without a new engine binary.

The core five

The 5 original specialists, from the v0.1 release:

Coding Specialist — Writes code, runs shell commands, operates git + gh, deploys to Railway. apps/engine/src/specialists/coding/
Content Specialist — Writes copy, email sequences, long-form, voice-matched to your brand. apps/engine/src/specialists/content/
Research Specialist — Web search via Tavily, competitor analysis, structured summaries. Archives results to a per-owner research cache. apps/engine/src/specialists/research/
Browser Specialist — Headless Playwright for real web tasks: filling forms, scraping dashboards, end-to-end QA. apps/engine/src/specialists/browser/
Business Ops Specialist — Email, calendar, invoicing, paperwork — the logistics layer. apps/engine/src/specialists/business_ops/

The 13 phase-2 specialists

Added with the 20-agent executive team expansion:

Design Specialist — Layouts, brand assets, generated imagery, pitch decks. apps/engine/src/specialists/design/
Product Specialist — Specs, user stories, roadmaps, release notes. apps/engine/src/specialists/product/
Chief of Staff — Internal coordination, meeting prep, decision memos. apps/engine/src/specialists/chief-of-staff/
Social Specialist — Platform-native posts, content calendars, engagement. apps/engine/src/specialists/social/
SEO Specialist — Keyword strategy, on-page audits, schema, internal linking. apps/engine/src/specialists/seo/
Ads Specialist — Ad copy, audience targeting, A/B variants. apps/engine/src/specialists/ads/
PR Specialist — Press outreach, media lists, pitches. apps/engine/src/specialists/pr/
Sales Specialist — Prospect research, personalized outreach, pipeline updates. apps/engine/src/specialists/sales/
Customer Success — Onboarding flows, retention check-ins, QBRs. apps/engine/src/specialists/customer_success/
Support Specialist — Ticket triage, draft replies, knowledge-base drafts. apps/engine/src/specialists/support/
Data Specialist — Analysis, dashboards, SQL, chart generation. apps/engine/src/specialists/data/
Finance Specialist — Invoicing, P&L summaries, cash-flow tracking. apps/engine/src/specialists/finance/
Legal Specialist — Contract drafts, terms, privacy policies (always flagged for human review). apps/engine/src/specialists/legal/

Plus: the Evaluator

The Evaluator is technically specialist #19, but it plays a different role: it reviews other specialists\' deliverables before they reach the owner. It has its own rubric (agent-constitutions/evaluator-rubric.md), is barred from running Skill Packs, and is invoked exclusively through review_deliverable_with_evaluator. See the evaluator gate for the enforcement story.

A concrete example

You ask: "Launch my newsletter." The CEO calls delegate_sequence with three steps — Research (pull recent articles in your niche), Content (draft the issue in your voice), Browser (queue it in your ESP). Each step sees the previous step\'s summary as "## Previous step output" prepended to its brief. Total specialist calls: three. Total tools the CEO invoked: one.

Technical details

// apps/engine/src/skills/loader.ts
export const SKILL_PACK_TARGET_SPECIALISTS = [
  'coding', 'content', 'research', 'browser', 'business_ops',
  'design', 'product', 'chief-of-staff', 'social', 'seo',
  'ads', 'pr', 'sales', 'customer_success', 'support',
  'data', 'finance', 'legal',
] as const;

Note what\'s missing: the CEO and the Evaluator. Both are deliberately skill-free so the orchestration and review layers stay stable while specialists evolve.

Related features

The CEO agent — the conductor that delegates to specialists.
Skill Packs — how specialists gain new powers without a new release.
Evaluator gate — the review step every deliverable passes.

Related concepts

FAQ

Why 18 and not one super-agent?

Specialization beats generality at this scale. Each specialist has its own system prompt, tool belt, and guidance files. The Coding Specialist knows git and Railway; the Legal Specialist knows to flag everything for human review. Forcing one prompt to do all of this regresses every job.

Can I talk to a specialist directly?

Not by design — you talk to the CEO, who talks to specialists. This keeps the coordination cost on the CEO and the execution cost on the specialist. That said, specialists can emit owner cards when they need a decision that the CEO can't make for them.

Do all specialists run at once?

Only when needed. The CEO uses spawn_parallel_specialists for genuinely independent work (build the page AND write the launch email AND research competitors) and delegate_sequence for serial chains (Research → Content → Browser to publish). Single tasks go through the specific delegate_to_* tool.

Does the Evaluator have skills or extensions?

No. The Evaluator is deliberately skill-free — it must stay an independent skeptic. The SKILL_PACK_TARGET_SPECIALISTS list in apps/engine/src/skills/loader.ts explicitly excludes it.

Try Black Box

The full 18-specialist bench, one subscription, no prompt engineering required.

See pricing How it works