Bundled Roles Reference
58 roles ship with agents-fleet. A role is an agent persona — it defines a system-prompt body and selects an agent type for a coordinator or worker. The role does not own tool permissions. agent_type selects a runtime baseline; the runtime permission profile, operator configuration, and approved spawn overrides determine effective access. Role or skill prose cannot grant tools or widen permissions. Bundled and user roles are always loaded. Project .fleet/roles/ are opt-in via --trust-project-skills (or AGENTS_FLEET_TRUST_PROJECT_SKILLS=1) and override lower tiers by name. save_role writes the project tier and updates the live registry; after manual file edits use reload_registry. List live roles with list_features --kind role.
See also: Bundled Skills, Crews, Security Autopilots.
Generated for agents-fleet v0.60.0 · 58 bundled roles.
Serialized .role.md schema
save_role accepts camelCase tool arguments but serializes the snake_case YAML fields below. The Markdown after the closing frontmatter delimiter is the role system prompt.
| Serialized field | Type | Required / default | Behavior |
|---|---|---|---|
name | string | Required | Kebab-case for save_role; must match the .role.md filename when loaded. |
description | string | Required by save_role; parser default: "" | Short human-readable role summary. |
agent_type | AgentType | Required by save_role; parser default: "general-purpose" | Selects the runtime permission baseline; it does not embed permissions in the role. |
version | string or number | Serializer-managed; parser default: "1" | Starts at 1 and increments when save_role overwrites the canonical file. |
model | string | Optional; default: session model | Preferred model when no higher-precedence runtime selection overrides it. |
when_to_use | string | Optional | Coordinator guidance for choosing the role. |
output_file | string | Optional | Declares the output filename expected from the role. |
Markdown body | Markdown | Required by save_role as systemPrompt | Persona and task guidance; cannot change runtime permissions or topology. |
Supported agent_type values for save_role are general-purpose, explorer, coder, reviewer, tester, custom, coordinator, and researcher. coordinator is not spawnable as a worker. Runtime-owned permission profiles live in src/fleet/agentPermissionProfiles.ts; configured and per-spawn overrides are resolved outside role files and privilege-increasing overrides require the explicit runtime gate.
Minimal serialized role
yaml
---
name: docs-writer
description: Writes project documentation
agent_type: coder
version: 1
---
You write accurate documentation from the implementation.Saving through save_role updates the running registry immediately. A manual file edit requires reload_registry; reload reapplies the session trust policy, so an untrusted project role is skipped even if a trusted save operation created it earlier in the session. Role changes affect later composition/spawns, not workers that are already running.
Bundled roles
Default Agent Roles
| Role | Agent Type | Description |
|---|---|---|
default-coder | coder | Default system prompt for coder worker agents |
default-coordinator | coordinator | Default project-neutral coordinator persona and response format |
default-explorer | explorer | Default system prompt for explorer worker agents |
default-general-purpose | general-purpose | Default system prompt for general-purpose worker agents |
default-researcher | researcher | Default system prompt for researcher worker agents |
default-reviewer | reviewer | Default system prompt for reviewer worker agents |
default-tester | tester | Default system prompt for tester worker agents |
Fleet-Specialized Roles
| Role | Agent Type | Description |
|---|---|---|
fleet-architect | reviewer | Architecture reviewer specialized for agents-fleet coordinator-worker patterns |
fleet-ci-engineer | coder | Infrastructure/CI engineer — builds & maintains CI pipelines (test, coverage, regression, build gates) |
fleet-ci-reviewer | reviewer | CI/build-integrity reviewer — ensures tests, coverage, and green builds are always in place |
fleet-coder | coder | Coder agent specialized for agents-fleet (TypeScript/ESM/Ink/Copilot SDK) |
fleet-correctness-reviewer | reviewer | Correctness reviewer specialized for agents-fleet type safety and async patterns |
fleet-explorer | explorer | Read-only explorer specialized for investigating the agents-fleet codebase |
fleet-maintainability-reviewer | reviewer | Maintainability reviewer specialized for agents-fleet code hygiene and conventions |
fleet-minimalism-reviewer | reviewer | Minimalism / over-engineering reviewer — finds what to delete, replace with stdlib/native, or shrink |
fleet-performance-reviewer | reviewer | Performance reviewer specialized for agents-fleet runtime efficiency and resource usage |
fleet-report-writer | general-purpose | Report-writing specialist — produces polished, interactive HTML reports (agents-fleet light brand theme) with SVG/PNG/interactive diagrams and on-demand PDF export |
fleet-researcher | researcher | Research and investigation agent with scoped write access for the agents-fleet codebase |
fleet-security-reviewer | reviewer | Security reviewer specialized for agents-fleet threat model |
fleet-tester | tester | Tester agent specialized for agents-fleet (Vitest/better-sqlite3/real implementations) |
fleet-workflow-architect | coder | Expert workflow author — composes the workflow-authoring skill, always validates via /wf validate, discovers reusable components |
Code Review Roles
| Role | Agent Type | Description |
|---|---|---|
architecture-reviewer | reviewer | Architecture and design pattern reviewer |
correctness-reviewer | reviewer | Logic and correctness focused code reviewer |
performance-reviewer | reviewer | Performance and efficiency focused reviewer |
security-reviewer | reviewer | OWASP-focused security audit reviewer |
test-reviewer | reviewer | Test coverage and quality reviewer for agents-fleet (TypeScript/Vitest) |
Investigation Roles (used by /init)
| Role | Agent Type | Description |
|---|---|---|
init-api-explorer | coder | Investigates API surface and contracts |
init-arch-explorer | coder | Investigates architecture and project structure |
init-build-explorer | coder | Investigates build system, test framework, and dev workflow |
init-critic | general-purpose | Adversarial critic that verifies investigation findings |
init-decision-explorer | coder | Investigates key design decisions and rationale |
init-dep-explorer | coder | Investigates dependencies and external services |
init-pattern-explorer | coder | Investigates coding patterns and conventions |
init-security-explorer | coder | Investigates security model and authentication |
init-synthesizer | general-purpose | Synthesis writer that produces the final /init completion artifacts |
Research, Brainstorm & Planning
| Role | Agent Type | Description |
|---|---|---|
context-researcher | explorer | Context and dependencies researcher - investigates surrounding context and historical background |
counter-researcher | explorer | Counter-arguments and alternatives researcher - investigates downsides and alternative approaches |
edge-case-researcher | explorer | Edge cases and failure modes researcher - investigates what could go wrong |
inquisitor-edge-cases | explorer | Adversarial edge-cases-angle inquisitor that grills the user with questions about failure modes, security, malformed input, and adversarial scenarios before a feature is built. Spawned by /brainstorm in round 1 alongside inquisitor-ux and inquisitor-technical. |
inquisitor-technical | explorer | Adversarial technical/integration-angle inquisitor that grills the user with questions about architecture, integration points, data shapes, and operational concerns before a feature is built. Spawned by /brainstorm in round 1 alongside inquisitor-ux and inquisitor-edge-cases. |
inquisitor-ux | explorer | Adversarial UX-angle inquisitor that grills the user with questions about intent, workflows, success criteria, and user experience before a feature is built. Spawned by /brainstorm in round 1 alongside inquisitor-technical and inquisitor-edge-cases. |
planning-coordinator | coordinator | Coordinator specialized for planning workflows - brainstorming, requirements, and design |
ponytail-coder | coder | Minimalist coder — a lazy senior developer who writes the least code that works |
primary-researcher | explorer | Primary research perspective - direct investigation of facts, code, documentation, examples |
research-coordinator | coordinator | Coordinator specialized for parallel research and investigation workflows |
research-critic | reviewer | Adversarial critic that cross-references and verifies research findings |
Security Roles
| Role | Agent Type | Description |
|---|---|---|
sec-binary-analyst | general-purpose | Static reverse engineer for a target binary — triages the file, maps parsing/decoder attack surface, patch-diffs patched-vs-unpatched builds, and produces an RE findings artifact plus a ranked list of fuzz targets |
sec-crash-triage-analyst | general-purpose | Triages fuzzing crashers and assesses realistic exploitability — reproduces, classifies (!exploitable / !analyze / ASan), determines the memory-safety primitive, weighs mitigations, writes the minimal impact-proving PoC (never weaponized), and produces a triage/exploitability artifact |
sec-exploit-poc-engineer | coder | Writes a local, sandboxed failing test that proves a vulnerability in the operator's own code |
sec-fixer | coder | Implements the minimal, correct security fix for a finding while preserving behavior |
sec-fuzzing-engineer | general-purpose | Builds and runs a coverage-guided fuzz harness for ONE target routine inside an isolated, disposable Azure VM — assembles seeds + dictionaries, fuzzes under sanitizers, measures coverage, minimizes crashers, and produces a fuzzing artifact |
sec-msrc-reporter | general-purpose | Produces the MSRC coordinated-disclosure report from a validated finding plus isolated-VM evidence, using the sec-msrc-report format; WRITES the report artifact |
sec-recon-mapper | general-purpose | Maps a codebase's attack surface and writes a structured recon artifact |
sec-reporter | general-purpose | Writes the final security report with per-finding detail and an executive summary |
sec-triage-analyst | general-purpose | Dedupes and prioritizes hunter findings and writes a triage summary + fix-target list |
sec-verifier | tester | Re-runs the PoC and relevant suites to confirm a fix resolves the finding with no regressions |
sec-vm-validator | general-purpose | Owns the Azure isolated-VM lifecycle — provision an isolated disposable VM, detonate the PoC, capture + hash evidence, assert the security impact, and ALWAYS tear down; uses az + powershell under hard isolation guardrails |
sec-vuln-hunter | reviewer | Senior appsec auditor that hunts one vulnerability class and returns evidence-backed findings |