Agent Guide
40 agents, 8 swarms, one quality standard
Forty specialized AI agents organized into eight swarms — each with a clear domain, a defined role, and a shared quality standard. See the full current roster on the Agents page.
Overview
The Bamwerks agent swarm is an AI organization designed to operate like a high-performance company. Each agent has a distinct specialty, a defined role (supervisor or worker), and a chain of accountability. They don't generalize — they specialize.
The swarm runs on FORGE: every non-trivial task follows a structured four-stage cycle (Reason → Act → Reflect → Verify). No single agent ships work without review. Quality gates run in parallel before anything reaches the Founder.
┌──────────────────────────────────────────────────────────────────┐
│ FOUNDER (Sirbam) │
│ │ │
│ ┌─────────────────────┼─────────────────────┐ │
│ │ │ │ │
│ Operations Engineering Intelligence │
│ (Atlas) (Stark) (Oracle) │
│ │ │ │ │
│ ┌──┴──┐ ┌────┴────┐ ┌───┴───┐ │
│ Forge Heimdall Ada Ratchet … Mercury Scribe … │
│ │
│ ┌─────────────────────┼─────────────────────┐ │
│ │ │ │ │
│ Finance Security Quality │
│ (Midas) (Sentinel) (Judge) │
│ │ │ │ │
│ ┌──┴──┐ ┌─────┴────┐ ┌────┴────┐ │
│ Penny Prophet Cipher Reaper … Hawk Testify … │
│ │
│ ┌────────────────────┬──────────────────────┐ │
│ │ │ │ │
│ Business Life │ │
│ (Sir) (Alfred) │ │
│ │ │ │ │
│ Chancellor Chronos Sage … │ │
│ Herald Nexus │ │
│ │ │
└─────────────────────────────────────────────────┴───────────────┘
Roles: Supervisors and Workers
Every agent is one of two types:
Supervisors own a domain. They coordinate work, prioritize requests, maintain quality standards, and are accountable for their swarm's output. They report directly to the Founder.
Workers are specialists. They receive task briefs, execute with depth in their area of expertise, and submit their work through the FORGE cycle before it's considered complete.
| Role | Accountability | Reports To |
|---|---|---|
| Supervisor | Domain ownership, swarm output | Founder |
| Worker | Task-level execution and quality | Swarm supervisor |
No worker ships directly to the Founder without supervisor review. No supervisor reviews their own work without an independent quality gate.
How Agents Are Dispatched
When the Founder submits a request, Sir (COO) orchestrates the response:
- Triage — Sir identifies which swarm(s) own the relevant domain
- Brief creation — A structured task brief is written (goal, constraints, context, output checklist)
- Specialist dispatch — The right worker agent receives the brief with role-specific context
- Parallel execution — Independent tasks are assigned simultaneously to compress time
- Review gates — Quality (Hawk) and Security (Sentinel/Cipher/Vault) review in parallel
- Delivery — After both gates pass, Sir delivers the verified output to the Founder
No agent gets more context than its task requires. Engineering agents get the codebase and build tools. Security agents get threat models. This separation keeps the system lean and focused.
FORGE at the Agent Level
Each agent runs FORGE on their own work before submitting it — this is an individual discipline, not just a system-level process.
An agent executing a task does:
| Stage | What the agent does |
|---|---|
| Reason | Reads the brief in full. Gathers all referenced context. Identifies any gaps before touching anything. |
| Act | Executes against the brief with full depth and specialty. Writes, builds, analyzes, or investigates. |
| Reflect | Reviews own output critically. Does it meet the brief's success criteria? What could be wrong? |
| Verify | Confirms deliverables (runs the build, tests the endpoint, checks the spec). Not just "it looks right" — it runs. |
The agent submits verified work to their supervisor. The supervisor and review agents then apply FORGE from their own perspectives — catching what the executing agent's assumptions would naturally miss. A security agent reviewing a frontend engineer's output finds things the engineer's own review would never flag.
This layering — agent-level FORGE + reviewer-level FORGE — is what makes autonomous AI work observable and trustworthy.
The Eight Swarms
Each swarm owns a domain and is led by a supervisor. For the current roster — every agent, their role, and their bio — see the Agents page; it's generated straight from the swarm registry, so it's always up to date.
| Swarm | Domain |
|---|---|
| 🔵 Operations | Infrastructure, tooling, and agent platform |
| 🟡 Engineering | Software design and development |
| 🟣 Intelligence | Research, communications, and knowledge management |
| 🟤 Business | Strategy, marketing, legal, and partnerships |
| 🟢 Finance | Financial strategy, cost management, and investment |
| 🩺 Quality | QA, testing, and process integrity |
| 🔴 Security | Protection, access control, and threat intelligence |
| 🩷 Life | Personal operations, wellness, and work-life balance |
Design Principles
Specialize, don't generalize. An agent that's good at everything is exceptional at nothing. Ratchet builds fast. Bishop writes clean. Wrench debugs relentlessly. This isn't redundancy — it's depth.
Multiple perspectives, by design. The same assumptions that shape implementation shape self-review. Sending Hawk and Sentinel to review code independently — without seeing each other's findings — produces the disagreement that catches what agreement would miss.
Anti-sycophancy. If reviewers unanimously approve, a contrarian review is triggered. Unanimous praise is a signal, not a conclusion. The swarm actively seeks the problem that optimism would hide.
Evidence over assertions. "I checked it and it's fine" is an assertion. A green build log, a Testify runtime report, and a Sentinel security sign-off are evidence. Bamwerks ships evidence.
Learn how these agents coordinate on every task in the FORGE Process, or see how the overall system is structured in Architecture.