Skip to main content
Forge Docs

Agents & Skills

Reference for all 33 specialized agents and 27 knowledge skills provided by the forge-plugin.

33 Specialized Agents

Forge agents are autonomous AI specialists. Each handles a specific domain. Claude Code selects the right agent based on your request, or you can invoke them directly through commands.

Core Workflow Agents

AgentModelWhen to Use
plannersonnetFeature planning, architecture design, task breakdown. Use when you need to decompose complex work into manageable tasks with dependencies and estimates.
buildersonnetFeature implementation, production-ready code, refactoring, test generation. Invoked after a plan is approved or when specific implementation is needed.
guardiansonnetQuality assurance, testing, security validation. Use after implementation is complete, when pre-commit quality gates need to run, or for code review.
orchestratoropusMulti-agent coordination. Activates the Forge developer experience — context restoration, feature planning, soundboard discussions, health checks.
master-architectopusSystem design decisions, architecture review, design pattern selection, infrastructure planning, and technical debt assessment. The senior architect who says "no, don't build it that way."

Domain Specialists

AgentModelWhen to Use
securitysonnetVulnerability scanning, OWASP Top 10 checks, secrets detection, auth/authz code review, CSP configuration, remediation guidance.
testingsonnetTest generation, coverage analysis, test infrastructure. Creates unit/integration/e2e tests, analyzes gaps, fixes flaky tests.
performancesonnetProfiling, bundle analysis, optimization. Identifies bottlenecks, reduces bundle size, optimizes renders, benchmarks.
apisonnetREST endpoint design, request/response handlers, validation, middleware, external API integration, OpenAPI specs.
databasesonnetSchema design, migration creation, query optimization, data modeling, indexing strategy, troubleshooting.
uisonnetReact components, responsive layouts, design system, animations, accessibility, UI performance.
refactorsonnetCode restructuring without behavior change. Extract functions, reduce complexity, eliminate duplication, apply patterns.
devopssonnetDocker, GitHub Actions, environment config, server setup, monitoring, deployment automation.
integrationsonnetExternal service connections. GitHub API, Sentry, webhooks, OAuth, MCP servers, service-to-service.
design-vanguardopusUI/UX design and frontend implementation. Component development, design systems, animations, accessibility audits, color/typography, dark mode, and design critiques.
dx-engineersonnetDeveloper experience evaluation and improvement. CLI UX audits, SDK API design review, onboarding flows, error message quality, and time-to-first-value analysis.
qa-sentinelsonnetTest STRATEGY and quality architecture — not test generation. Designs testing approaches, enforces CRUCIBLE gates, and recommends advanced techniques (mutation, property-based, load, contract).

Governance & Analysis Agents

AgentModelWhen to Use
detectivesonnetProject analysis, health checks, tech stack detection, gap analysis, code quality assessment, architectural review. "What's wrong with this codebase?"
analyticssonnetMetrics tracking, data analysis, reporting. Implementing analytics events, building dashboards, tracking KPIs.
compliancesonnetRegulatory compliance, license auditing, GDPR/privacy, accessibility (WCAG), export controls.
docssonnetDocumentation generation, README creation, API docs, changelog maintenance, stale doc detection.
learningsonnetSession analysis, pattern recognition, workflow optimization. Learns from past sessions to improve recommendations.

Specialized System Agents

AgentModelWhen to Use
release-sentinelopusDocumentation auditing after code changes, release preparation, changelog generation, stale section detection.
incident-commanderopusProduction incident triage and coordination. Outages, failed deployments, data corruption, security breaches, plus post-incident reviews and runbook generation.
crucible-detectivesonnetForensic test quality auditing. Detects hollow assertions, coverage gaming, mock proliferation, dead tests. Use before release gates or during spot audits.
governance-verifiersonnetAutomated verification for governance concerns flagged by hooks. Responds to quality check results.
oraclesonnetProactive governance sentinel for autonomous development. Monitors changes against project scope and architectural constraints.

Business & Growth Agents

AgentModelWhen to Use
product-strategistopusProduct strategy, market positioning, feature prioritization (RICE/MoSCoW), product-market fit evaluation, roadmaps, and go-to-market planning.
growth-enginesonnetGrowth strategy, launch playbooks (Product Hunt, Hacker News), content marketing, SEO audits, community building, and developer relations.
revenue-architectopusPricing strategy, monetization design, Stripe integration, revenue modeling, and enterprise sales planning. The money brain.
scoutsonnetCompetitive intelligence and market analysis. Competitor feature matrices, pricing comparisons, market sizing (TAM/SAM/SOM), SWOT and Five Forces analysis.
wordsmithsonnetUser-facing writing that needs to work. Landing page copy, technical blog posts, READMEs, email sequences, error messages, CLI help text, changelogs.

Executive Agent

AgentModelWhen to Use
nxtg-ceo-loopopusCEO Digital Twin for autonomous strategic decisions. Product direction, architecture decisions, resource allocation, risk assessment. Only escalates CRITICAL decisions to humans.

27 Knowledge Skills

Skills are contextual knowledge documents that Claude Code loads automatically when relevant. They encode patterns, conventions, and best practices so agents make informed decisions without being told the same things repeatedly.

Core Skills

SkillWhat It Provides
core-nxtg-forgeForge-specific conventions, MCP tool usage, governance patterns
ceo-loopThe ORBIT execution protocol (OBSERVE → REASON → BUILD → INSPECT → TURN) for autonomous product governance iterations
parallel-executionPlan Mode and Agent Teams force-multipliers, the canonical forge agent roster, and delegation-routing rules

Domain Knowledge

SkillWhat It Provides
architectureSystem design patterns, component relationships, data flow
coding-standardsLanguage-specific conventions, formatting, documentation standards
documentationDoc structure, API documentation patterns, README conventions
owasp-securityOWASP vulnerability detection and remediation, plus the Forge plugin's own security guards and how to resolve a SECURITY BLOCK
testingTest framework usage, assertion patterns, coverage tools
crucible-auditForensic test-quality auditing — detects hollow assertions, coverage-omit gaming, mock proliferation, and dead tests that produce false green signals

Framework Skills

SkillWhat It Provides
claude-code-frameworkClaude Code capabilities, tool usage, agent patterns
codex-frameworkCodex CLI conventions, task mode, AGENTS.md format
gemini-frameworkGemini CLI conventions, GEMINI.md format

Workflow Skills

SkillWhat It Provides
dev-environment-patternsEnvironment setup, toolchain configuration, CI/CD patterns
git-workflowBranching strategy, commit conventions, PR patterns
runtime-validationInput validation, error boundaries, runtime safety patterns

Performance & Debugging

SkillWhat It Provides
optimizationPerformance profiling, bundle optimization, caching strategies
browser-debuggingDevTools usage, network analysis, rendering profiling

Knowledge & Learning

SkillWhat It Provides
skill-developmentHow to create new skills, skill structure, progressive disclosure
domain-knowledgeProject-specific domain concepts, business rules
verify-governanceGovernance verification patterns, health score interpretation

Agent Role Skills

SkillWhat It Provides
agent-lead-architectSystem design leadership, architecture decisions, trade-off analysis
agent-backend-masterBackend implementation patterns, API design, data layer
agent-cli-artisanCLI tool design, argument parsing, terminal UX
agent-platform-builderPlatform engineering, infrastructure, deployment
agent-qa-sentinelQuality assurance leadership, test strategy, coverage analysis
agent-developmentAgent creation patterns, system prompt design
agent-integration-specialistService integration patterns, API connections, webhooks

How Skills Work

Skills load automatically based on context. When you're working on tests, the testing skills activate. When you're building an API, the API and backend skills activate. You don't need to invoke them manually.

Each skill has a SKILL.md with YAML frontmatter describing when it's relevant and markdown content providing the knowledge. Claude Code matches your current task context against skill descriptions and loads the most relevant ones.