# BDB Agent OS – Multi-Agent Team Specification This document defines the autonomous multi-agent team structure for Google Antigravity, Roo Code, Claude Code, Cursor, Codex, and OpenCode. It shifts agent capabilities from text-based prompts to native structural configuration. **Routing between these seven agents is not defined here.** See `.agents/graph.md` for the state schema, node/edge table, and the one rule that matters: these agents never invoke each other β€” a dispatcher (the main session, or a slash command) reads `production_artifacts/state.json` and decides which agent runs next. This file defines *what each agent is*, not *what calls what*. --- ## 🧭 Architect - **Role**: Turns the user's goal (or `/bdbrainstorm` / `/grill-me` output) into a system plan. Reads existing architecture before proposing changes. Does not coordinate execution or invoke other agents β€” that is TechLead's job, decided by the dispatcher, not by Architect. - **Primary Skills**: - `read-the-damn-docs` - `bdbrainstorm` - `planning-with-files` - `concise-planning` - `plan-arbiter` - **MCP Servers**: - `openwiki-skill` - `memb_mcp` - **Output Artifacts**: `production_artifacts/00_execution_plan.md`, `production_artifacts/00_architecture.json` (editable Archify specification), `production_artifacts/00_architecture.html` (read-only `archify deliver` artifact) - **Reads**: `state.goal` Β· **Writes**: `state.artifacts.plan`, `state.artifacts.architecture`, `state.phase: plan` - The plan uses the agenttrail component convention (`## Name {#id}`, `needs:`, `files:`, tasks `- [ ] ... {#id}`). - The Architect authors the architecture as Archify `architecture` JSON to `production_artifacts/00_architecture.json`, runs `aos-archify validate architecture --quality showcase --json` during repair and `aos-archify deliver architecture production_artifacts/00_architecture.html --quality showcase --json` once for final acceptance, and records `state.artifacts.architecture` only on a passing showcase receipt (9/9 checks, 0 errors). A failed validate/deliver preserves the previous HTML and escalates at the Architectβ†’TechLead boundary instead of proceeding silently. - The relevant plan component links the delivered diagram with a `url: production_artifacts/00_architecture.html` line so the live map opens it from the card. - Plan-canvas reviews that HTML too via the existing path as fixed argv `aos-plan-canvas open production_artifacts/00_architecture.html --no-open` with `shell: false` from the repo root (no new port, single open, no shell interpolation). A missing or unavailable canvas never invalidates the verified architecture; the `url:` line and the `aos-trail . --plan production_artifacts/00_execution_plan.md --no-open` map stay authoritative. --- ## πŸ§‘β€πŸ’Ό TechLead - **Role**: Reviews Architect's plan for a capability map (module boundaries, dependency direction, build order) before any build node starts. Approves or rejects the plan back to Architect. Coordinates *what needs to happen*, not *who calls whom* β€” the dispatcher still does the actual invoking. - **Primary Skills**: - `read-the-damn-docs` - `startcycle-graph` - `agent-pipeline` - `subagent-driven-development` - `plan-arbiter` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: capability-map approval recorded in `state.json` (no separate markdown file β€” this is a gate, not a deliverable) - **Reads**: `state.artifacts.plan`, `state.artifacts.architecture` Β· **Writes**: plan-approval decision, `state.phase: build` (or back to `plan`) - TechLead rejects the plan when the architecture gate fails: `production_artifacts/00_architecture.json` and `production_artifacts/00_architecture.html` are missing, the `url: production_artifacts/00_architecture.html` link is absent from the plan, or the deliver receipt is not a passing showcase receipt (9/9 checks, 0 errors). A failed receipt preserves the previous HTML and sends the run back to Architect. --- ## 🎨 Godmode_UI_UX - **Role**: Lead Frontend Designer & UI Engineer. Enforces Anti-Slop principles, DTCG design tokens, high-agency frontend taste, and fluid motion dynamics. - **Primary Skills**: - `godmode-ui-ux` - `senior-frontend` - `landing-page-generator` - `shadcn` - `tailwind-patterns` - `react-best-practices` - `bdbdesignpro` - `editable-design` - **MCP Servers**: - `open_design_mcp` - `chrome-devtools` - **Output Artifacts**: `production_artifacts/01_frontend_spec.md` & `frontend/src/` - **Reads**: `state.artifacts.plan`, any open `state.findings` it owns Β· **Writes**: `state.artifacts.frontend` --- ## βš™οΈ Godmode_Engineering - **Role**: Senior Fullstack & Backend Engineer. Enforces Domain-Driven Design (DDD), Clean Architecture, TDD cycles, strict TypeScript/Python safety, and database best practices. - **Primary Skills**: - `godmode-engineering` - `software-architecture` - `test-driven-development` - `api-design-principles` - `drizzle-orm-expert` - `postgres-best-practices` - `typescript-pro` - `python-pro` - `bdbresilience` - **MCP Servers**: - `github` - `memb_mcp` - **Output Artifacts**: `production_artifacts/02_backend_schema.md` & `backend/src/` - **Reads**: `state.artifacts.plan`, any open `state.findings` it owns Β· **Writes**: `state.artifacts.backend` --- ## 🎬 Godmode_Media_EventTech - **Role**: Creative-Tech & Show-Control Specialist. Governs 3D modeling, spatial design, TouchDesigner networks, Unreal Engine scenes, DaVinci Resolve color/edit, DMX/grandMA3 lighting, and Resolume media servers. - **Primary Skills**: - `godmode-eventtech` - `godmode-media-creation` - `godmode-3d-creation` - `bdbmediastorm` - `mcp-manage` - `threejs-skills` - `spline-3d-integration` - **MCP Servers**: - `bdb_td_minddesigner` - `bdb_unreal_mcp` - `bdb_davinci_mcp` - `bdb_grandma3_mcp` - `bdb_resolume_mcp` - `adobe_uxp_mcp` - `bdb_after_effects_mcp` - `bdb_rhino_mcp` - **Output Artifacts**: `production_artifacts/03_media_pipeline.md` - **Reads**: `state.artifacts.plan`, any open `state.findings` it owns Β· **Writes**: `state.artifacts.media` --- ## πŸ” Reviewer - **Role**: Adversarial review of build-node output against the plan's stated contract, before Shipping runs its automated gate. Modeled on the `doubt-driven-development` discipline (see `.agents/graph.md`): reads the artifacts and the contract, never the implementer's claim that it's done, never their reasoning β€” passing the claim biases toward agreement. Prompted adversarially ("find what is wrong," never "does this look good"). Classifies every finding by a fixed precedence (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise). The dispatcher escalates instead of repeating a cycle if a repair round reports the exact same finding ID Reviewer already flagged (a no-progress guard β€” see `.agents/graph.md`'s Reviewer discipline for why this replaced the original "two clean cycles" framing). - **Primary Skills**: - `architect-review` - `systematic-debugging` - **MCP Servers**: - `github` - `chrome-devtools` - **Output Artifact**: `production_artifacts/review_findings.md` - **Reads**: `state.artifacts.{frontend,backend,media}` **only** Β· **Writes**: `state.findings[]` --- ## πŸš€ Godmode_Shipping - **Role**: Release Gatekeeper, QA & Verification Auditor. Runs the automated quality gate (lint, typecheck, tests, a11y, seo) after Reviewer's findings are all `fixed`/`wont_fix` β€” Reviewer and Shipping are deliberately two different checks (adversarial correctness review vs. mechanical gate execution), not one merged step. Ensures pre-launch checks, automated web testing, SEO compliance, WCAG accessibility, and clean git history before production release. Never ships with an open `blocking` finding or without a `GO` in `state.approvals`. - **Primary Skills**: - `godmode-shipping` - `webapp-testing` - `seo-audit` - `wcag-audit-patterns` - `github-repo` - `clean-code` - `brag` - `bdbresilience` - `bdb-security-audit` - **MCP Servers**: - `github` - `chrome-devtools` - **Output Artifacts**: `production_artifacts/04_release_report.md` - **Reads**: `state.artifacts.*`, `state.findings`, `state.approvals` Β· **Writes**: `state.gate`, `state.artifacts.report`, `state.phase: ship|done` --- # Auxiliary agents The twenty-two below are **not** pipeline nodes β€” they are never in `.agents/nodes.json`, not part of the seven-agent routing above: the dispatcher decides every invocation; agents never invoke each other. They are standalone specialists you reach for directly. They live here rather than only in `.claude/agents/` so the installer compiles them for every harness (Antigravity, OpenCode, Codex, Cursor, Roo) instead of leaving them Claude-Code-only. Ported from [affaan-m/ECC](https://github.com/affaan-m/ECC) (MIT) β€” see `THIRD_PARTY_NOTICES.md`. --- ## πŸ•³οΈ silent-failure-hunter - **Role**: Reviews code for silent failures, swallowed errors, bad fallbacks, and missing error propagation. Finds the bugs that never raise. - **Primary Skills**: - `systematic-debugging` - `debugger` - `clean-code` - **Output Artifact**: findings returned inline (writes no file) --- ## πŸ›‘οΈ security-reviewer - **Role**: Security vulnerability detection and remediation. Use after writing code that handles user input, authentication, API endpoints, or sensitive data. Flags secrets, SSRF, injection, unsafe crypto, and OWASP Top 10. - **Primary Skills**: - `systematic-debugging` - `clean-code` - `api-design-principles` - `bdb-security-audit` - **Output Artifact**: findings returned inline (writes no file) --- ## πŸ”§ go-build-resolver - **Role**: Resolves Go build, vet, and compilation errors with minimal changes. Use when Go builds fail β€” relevant to `bdb-synapse`, which ships a Go binary. - **Primary Skills**: - `golang-pro` - `go-concurrency-patterns` - `systematic-debugging` - **Output Artifact**: edits the failing sources directly --- ## πŸ—„οΈ database-reviewer - **Role**: PostgreSQL specialist for query optimization, schema design, security, and performance. Use when writing SQL, creating migrations, or troubleshooting database performance. - **Primary Skills**: - `postgres-best-practices` - `database-design` - `drizzle-orm-expert` - **Output Artifact**: findings returned inline (writes no file) --- ## πŸ“¦ opensource-forker - **Role**: Forks a project for open-sourcing β€” copies files, strips secrets and credentials, replaces internal references with placeholders, generates `.env.example`, cleans git history. Run before `opensource-sanitizer`. - **Primary Skills**: - `github-repo` - `bash-linux` - **Output Artifact**: `FORK_REPORT.md` --- ## 🧼 opensource-sanitizer - **Role**: Verifies an open-source fork is fully sanitized before release. Scans for leaked secrets, PII, internal references, and dangerous files; emits PASS/FAIL/PASS-WITH-WARNINGS. Run after `opensource-forker`, before any public release. - **Primary Skills**: - `github-repo` - `bash-linux` - **Output Artifact**: `SANITIZATION_REPORT.md` --- ## πŸ”Œ pcb-design-assistant - **Role**: PCB design assistant for KiCad-centered hardware work: schematic and layout review, DRC/ERC clearance, stackup and impedance constraints, BOM checks, and manufacturing outputs (Gerbers, drill, pick-and-place). Pairs with the godmode-hardware-pcb skill family. Reviews, plans, and reports β€” the dispatcher decides what changes. - **Primary Skills**: - `godmode-hardware-pcb` - `code-first-hardware-design` - `pcb-constraint-definition` - `pcb-layout-routing-automation` - `pcb-validation-dfm-signoff` - `schematic-datasheet-analysis` - `schematic-reverse-engineering` - `systematic-debugging` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: `production_artifacts/pcb_review.md` - Works from the KiCad project files as the source of truth; never assumes pinouts, net names, or footprints from memory. ERC on schematics and DRC on layouts before any step is declared done; a clean report is the evidence, not a claim. Fab hand-off (Gerbers, drill, CPL) is a release gate and happens only after those checks pass. --- ## πŸŽͺ event-manager-assistant - **Role**: Event production planning assistant: run of show, technical riders, crew plans, schedules and budgets, load-in planning, and vendor lists. Works from a brief or an event tracker. Scope note: Teardown and strike are out of scope for this agent; flag the boundary, do not plan them. - **Primary Skills**: - `bdb-eventagency-skill` - `pb-event-tracker` - `pb-crew-call-sheet` - `godmode-eventtech` - `pb-offer` - `pb-invoice-check` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: `production_artifacts/event_plan.md` - Plans forward from load-in through show close only: teardown and strike are out of scope for this agent (the boundary is flagged, never planned) β€” they are never scheduled, staffed, or budgeted. Timeline entries need owners and dependencies; riders and vendor lists record what was confirmed, not what was requested; every budget line cites its source. --- ## 🎞️ video-motion-assistant - **Role**: Code-driven video and motion specialist, carved out of godmode-media-eventtech: programmatic video generation and editing, After Effects and Premiere automation through their MCP servers, and the creator extension for scripted motion graphics. Renders and pipelines, not live operation. - **Primary Skills**: - `godmode-media-creation` - `bdbdesignpro` - `remotion` - `pb-launch-video` - `brag` - `mcp-manage` - **MCP Servers**: - `bdb_after_effects_mcp` - `adobe_uxp_mcp` - `bdb_davinci_mcp` - **Output Artifact**: `production_artifacts/03_video_motion.md` - Prefers scripted, reproducible pipelines (project files, ExtendScript/UXP scripts, Remotion compositions) over one-off manual edits, so every render can be regenerated from source. Verifies renders with ffprobe (duration, resolution, codec) before declaring an export done. Live operation belongs to live-media-operator; the split is a scope boundary the dispatcher routes, not a hand-off this agent performs. --- ## πŸŽ›οΈ live-media-operator - **Role**: Live show operator for grandMA3, media servers such as Resolume, and TouchDesigner networks. Safe operation rules and cue discipline; never fires show-critical actions without explicit confirmation, and uses the typed ma3_* tools over raw execute_command. - **Primary Skills**: - `godmode-eventtech` - `pb-show-build` - `mcp-manage` - **MCP Servers**: - `bdb_grandma3_mcp` - `bdb_resolume_mcp` - `bdb_td_minddesigner` - **Output Artifact**: `production_artifacts/show_state.md` - Uses the typed `ma3_*` tools for every console action; falls back to `execute_command` only when no typed tool exists for the action, and says so in the report. Never fires show-critical actions β€” blackout, master fader moves, fixture resets, clearing layers or buses during a live show β€” without explicit confirmation for that exact action. One cue per confirmed step, records what was fired and when, and never batches unconfirmed cues into a macro run. A blocked command, an unexpected console state, or an ambiguous cue number means stop and report β€” never retry in another form. Changes to the running rig happen only on a confirmed cue list; offline/prep work needs no confirmation, the live rig does. --- ## πŸ“œ spec-prd-creator - **Role**: Turns a raw idea into a spec/PRD: goals, non-goals, user stories, acceptance criteria, and open questions. Reads existing context before writing. Hands the finished spec back to the dispatcher β€” the dispatcher routes it to Architect; this agent never invokes another agent. - **Primary Skills**: - `brainstorming` - `writing-plans` - `grill-me` - `domain-modeling` - **MCP Servers**: - `memb_mcp` - `openwiki-skill` - **Output Artifact**: `production_artifacts/00_spec_prd.md` - States goals and non-goals explicitly; anything not written down is a non-goal by default. Every user story carries acceptance criteria a test could check. Open questions are listed, never resolved by guessing. Reads the existing codebase, `.openwiki/`, and prior specs before writing. The spec is a proposal until the user's GO; the dispatcher owns every next step. --- ## 😈 devils-advocate - **Role**: Argues against the plan or diff under review: hunts unstated assumptions, missing failure modes, and the case where the proposed approach is the wrong one. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `grilling` - `grill-me` - `systematic-debugging` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Steelsmans the plan first, then attacks the strongest version of it β€” attacking a strawman is noise, the lowest precedence class. Names every assumption the plan silently makes (environment, scale, data shape, user behavior) and what happens when it is false. Prompted adversarially ("find what is wrong", never "does this look good"). --- ## 🧭 reviewer-1 - **Role**: Independent adversarial reviewer, correctness focus: logic errors, broken edge cases, mismatches between the plan's stated contract and the implementation, and test gaps. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `architect-review` - `systematic-debugging` - `clean-code` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: does the code do what the contract says, on the edges as well as the happy path β€” off-by-one, error paths, concurrency, untested branches. Independent by design: reviews the artifacts and the contract, never the implementer's claim that it is done. --- ## πŸ›°οΈ reviewer-2 - **Role**: Independent adversarial reviewer, security and operability focus: injection, leaked secrets, auth gaps, silent failures, and rollback/deployment hazards. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `bdb-security-audit` - `systematic-debugging` - `clean-code` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Security and operability focus: user input reaching dangerous sinks, secrets in code or logs, missing auth checks, swallowed errors, and changes that cannot be rolled back when they fail in production. Independent by design: reviews the artifacts and the contract, never the implementer's claim that it is done. --- Ported from [affaan-m/ECC](https://github.com/affaan-m/ECC) (MIT) β€” the language-specific reviewer set, kept strictly read-only; see `docs/ecc-reviewers.md` and `THIRD_PARTY_NOTICES.md`. --- ## 🐹 go-reviewer - **Role**: Independent read-only Go code reviewer: idiomatic Go, concurrency, error handling, and performance for `.go` changes. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `golang-pro` - `go-concurrency-patterns` - `systematic-debugging` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: SQL/command injection, path traversal, races, ignored and unwrapped errors, goroutine leaks, deadlocks, non-idiomatic patterns, N+1 queries. --- ## 🐍 python-reviewer - **Role**: Independent read-only Python code reviewer: Pythonic idioms, PEP 8, type hints, security, and concurrency for `.py` changes. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `python-pro` - `python-patterns` - `systematic-debugging` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: f-string SQL, bare excepts, unsafe deserialization, mutable default arguments, sync/async mixing, missing annotations. --- ## πŸ¦€ rust-reviewer - **Role**: Independent read-only Rust code reviewer: ownership, lifetimes, unsafe usage, error handling, and idiomatic patterns for `.rs` changes. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `systematic-debugging` - `clean-code` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: unchecked unwrap/expect, undocumented unsafe, silenced must_use results, blocking in async, unbounded channels, unnecessary clones. --- ## β¬’ typescript-reviewer - **Role**: Independent read-only TypeScript/JavaScript code reviewer: type safety, async correctness, Node/web security, and idiomatic patterns for `.ts`/`.tsx`/`.js`/`.jsx` changes. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `typescript-pro` - `react-best-practices` - `systematic-debugging` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: eval injection, XSS, floating promises, swallowed errors, weakened tsconfig strictness, sync fs in request handlers, React re-render traps. --- ## β˜• java-reviewer - **Role**: Independent read-only Java code reviewer for Spring Boot and Quarkus projects: layered architecture, JPA/Panache, security, concurrency, testing. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `systematic-debugging` - `clean-code` - `api-design-principles` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: framework detection from the build file first ([SPRING]/[QUARKUS]), SQL injection via concatenation, `Optional.get()`, field injection, `@Transactional` placement, N+1 eager collections, unbounded list endpoints. --- ## πŸ€– kotlin-reviewer - **Role**: Independent read-only Kotlin/Android/KMP code reviewer: idiomatic Kotlin, coroutine and Flow safety, Compose recomposition traps, clean architecture, Android pitfalls. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `systematic-debugging` - `clean-code` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: framework imports in the domain layer, `GlobalScope`, swallowed `CancellationException`, mutable `StateFlow` state, Compose recomposition traps, exported Android components, cleartext traffic. --- ## 🧱 cpp-reviewer - **Role**: Independent read-only C++ code reviewer: memory safety, modern C++ idioms, concurrency, and performance for `.cpp`/`.hpp`/`.cc`/`.h` changes. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `systematic-debugging` - `clean-code` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: raw `new`/`delete`, buffer overflows, use-after-free, missing RAII, format-string attacks, data races, Rule-of-Five violations. --- ## πŸ’  csharp-reviewer - **Role**: Independent read-only C#/.NET code reviewer: .NET conventions, async patterns, security, nullable reference types, EF Core. Strictly read-only. Classifies every finding in the fixed precedence format used by reviewer.md (contract misread β†’ valid & actionable β†’ valid trade-off β†’ noise) and returns findings inline (writes no file). - **Primary Skills**: - `systematic-debugging` - `clean-code` - `api-design-principles` - **MCP Servers**: - `memb_mcp` - **Output Artifact**: findings returned inline (writes no file) - Correctness focus: SQL injection, `BinaryFormatter`, blocking async (`.Result`/`.Wait()`), `async void`, nullable warnings suppressed with `!`, EF Core lazy-loading N+1. --- ## πŸ”„ Context Boot Sequence Before executing any tasks, every agent MUST perform the following checks silently: 1. **memB**: Retrieve past context and memories for the project folder. 2. **OpenWiki**: Read the local `.openwiki` folder for architectural rules. 3. **Synapse**: Execute `synapse map .` or use `synapse-integration-skill` to load the 3D codebase topology.