BMAD+ Persona Stack

Nestor ships with five opinionated personas — Atlas, Forge, Sentinel, Nexus, Shadow — each a specialist with triggers, handoff rules, and anti-patterns. They are drop-in prompt modules, not separate processes.

Overview

BMAD+ (Business, Management, Architecture, Delivery, plus) is a multi-role pattern Nestor inherits from BMAD-METHOD v6.2.0. The five personas live as markdown prompt files under packages/agent/src/prompts/agents/ — one file per role, all shipped with the npm package.

Each persona has a single responsibility, a short "when you activate" section, explicit handoff rules to the others, and a "silence" clause that tells the agent to say nothing when there is nothing to say. They are deliberately terse.

Atlas — Strategist

Source: packages/agent/src/prompts/agents/ATLAS.md. Atlas covers business analysis and product management. It challenges scope, thinks in trade-offs, and demands evidence over opinion. One-page briefs only.

Triggers: business, scope, priority, PM, roadmap, market, pricing, user research, competitor, positioning, ROI.

Handoff: code or deploy to Forge, test coverage to Sentinel, sprint coordination to Nexus, competitor intel to Shadow.

Forge — Architect-Dev

Source: packages/agent/src/prompts/agents/FORGE.md. Forge writes code, architecture, and docs. Core rule: read before you write — every file is read fully before it is changed. Forge also enforces the FTP deploy discipline for nestor.sh and the npm package name (nestor-sh).

Triggers: implement, code, build, deploy, architecture, refactor, fix bug, write tests, ship.

Handoff: scope questions to Atlas, QA to Sentinel, multi-step sprints to Nexus, OSINT to Shadow.

Sentinel — Quality

Source: packages/agent/src/prompts/agents/SENTINEL.md. Sentinel owns QA and UX review. It assumes regression until proven otherwise, enforces WCAG AA as a minimum, and refuses to sign off on a PR it has not exercised.

Triggers: test, QA, UX review, accessibility, a11y, regression, bug, broken, flaky, coverage.

Handoff: scope to Atlas, implementation to Forge, orchestration to Nexus, investigation to Shadow.

Nexus — Orchestrator

Source: packages/agent/src/prompts/agents/NEXUS.md. Nexus is the sprint manager, autopilot driver, and handoff router. It also serves as the fallback persona when no other trigger matches. Nexus parallelizes by default and serializes when two agents would touch the same file.

Triggers: sprint, orchestrate, autopilot, coordinate, handoff, parallel, who does what, status.

Handoff: routes to the other four based on the task; surfaces blockers rather than absorbing them.

Shadow — OSINT

Source: packages/agent/src/prompts/agents/SHADOW.md. Shadow does investigation, scraping, and psychoprofiling. Every claim needs a URL, timestamp, and retrieval method. Findings are tagged with confidence (high / medium / low / speculative). PII is redacted by default.

Triggers: OSINT, investigate, scrape, psychoprofile, research, dig, background check, competitor intel, lookup.

Handoff: returns findings to whichever persona asked; escalates legal/ToS blockers.

How activation works

There are two equally valid ways to wake a persona:

1. Name mention. Typing "Atlas, what's the smallest viable test?" or "Forge, implement this" loads the matching prompt file and pins the persona for the turn.
2. Trigger phrase. When the user message contains a word from a persona's trigger list, the orchestrator auto-routes to that persona. No explicit naming required.

If a message contains triggers for more than one persona, Nexus routes and announces the handoff out loud ("Forge takes the implementation, Sentinel reviews after"). When nothing matches, Nexus is the fallback.

Customize per project

Each BMAD+ persona is a plain markdown file. To adjust a persona for a specific project, copy the file out of the package and edit it:

# Copy Forge into your project as a local override
mkdir -p .agents/prompts
cp node_modules/nestor-sh/packages/agent/src/prompts/agents/FORGE.md \
   .agents/prompts/FORGE.md

# Edit the triggers, anti-patterns, or handoff rules
$EDITOR .agents/prompts/FORGE.md

Local persona files take precedence over the packaged defaults. Keep the top-level sections intact (Identity, When you activate, How you work, Anti-patterns, Handoff, Silence) so the orchestrator can still parse triggers and routing.

For teams with their own role vocabulary, the cleanest pattern is to add triggers rather than replace them. Atlas still fires on "roadmap"; you add "OKR review" alongside. The orchestrator unions all triggers.

See Agents for the runtime that loads these personas, and the documentation overview for the full stack.