Personas Cheatsheet
Quick reference for the SNIPER v3 persona system. In v3, agent roles are defined directly in agent definition files (.claude/agents/*.md), and cognitive personas are lightweight thinking-style mixins applied on top of any agent via configuration.
Agents
SNIPER v3 ships 13 agent definitions. Each agent has a dedicated role, a model assignment, and scoped write permissions.
| Agent | Model | Role | Write Scope |
|---|---|---|---|
| lead-orchestrator | opus | Coordinates agent teams through protocol phases. Delegates -- never codes. | .sniper/, .sniper-workspace/ |
| analyst | sonnet | Researches, analyzes, and produces discovery artifacts (specs, codebase overviews). | docs/, .sniper/ |
| architect | opus | Designs system architecture, component boundaries, and technical plans. | docs/ |
| product-manager | sonnet | Translates requirements into structured stories with EARS acceptance criteria. | docs/ |
| backend-dev | sonnet | Implements server-side code (APIs, services, workers) in an isolated worktree. | worktree (full) |
| frontend-dev | sonnet | Implements client-side code (components, pages, hooks) in an isolated worktree. | worktree (full) |
| fullstack-dev | sonnet | Handles both backend and frontend for smaller projects. Isolated worktree. | worktree (full) |
| code-reviewer | opus | Multi-faceted code review: scope validation, standards enforcement, risk scoring. | docs/ |
| qa-engineer | sonnet | Writes tests, analyzes coverage, validates EARS acceptance criteria. | test files only |
| gate-reviewer | haiku | Runs automated gate checks at phase boundaries. Lightweight and fast. | .sniper/gates/ |
| retro-analyst | sonnet | Post-protocol retrospective analysis. Captures learnings. | .sniper/ |
| doc-writer | sonnet | Incrementally updates project documentation after implementation. | CLAUDE.md, README.md, docs/ |
| memory-curator | sonnet | Curates and maintains project memory (conventions, anti-patterns, decisions). | .sniper/memory/ |
Agent groupings by protocol phase
Discover -- analyst
Define -- product-manager
Design -- architect
Solve -- product-manager
Implement -- fullstack-dev + qa-engineer (full protocol), or backend-dev + frontend-dev + qa-engineer (spawned by lead-orchestrator based on project config)
Review -- code-reviewer (gate-reviewer for automated checks)
Retro -- retro-analyst
Orchestration -- lead-orchestrator (active across all phases)
Cognitive Personas
Cognitive personas are thinking-style mixins. They modify how an agent reasons, not what it does. SNIPER v3 ships three cognitive personas.
devils-advocate
Challenges assumptions, stress-tests designs, and finds failure modes before they reach production.
- Questions the happy path -- asks "What happens when this fails?"
- Hunts for edge cases: boundary values, timing issues, scale breaks, partial failures
- Every challenge comes with a concrete scenario, not vague doubt
Best applied to: code-reviewer, qa-engineer, analyst
See full reference.
performance-focused
Evaluates every decision through an efficiency lens -- latency, throughput, memory, scalability.
- Measures before optimizing -- profiles first, optimizes second
- Checks for N+1 queries, unbounded collections, unnecessary computation, blocking I/O
- Weighs speedup against complexity cost
Best applied to: backend-dev, fullstack-dev, architect
See full reference.
security-first
Evaluates every decision through a security lens -- threat modeling, least privilege, defense in depth.
- Asks "How could this be abused?" for every external input and API endpoint
- Checks input validation, authentication, authorization, data exposure, injection, secrets
- Flags concerns explicitly rather than making assumptions
Best applied to: architect, code-reviewer, backend-dev
See full reference.
Applying Cognitive Personas
Cognitive personas are applied to agents via .sniper/config.yaml. Add entries under agents.mixins mapping agent names to an array of cognitive personas:
# .sniper/config.yaml
agents:
mixins:
architect: [security-first]
code-reviewer: [devils-advocate]
backend-dev: [performance-focused]Multiple cognitive personas can be applied to the same agent:
agents:
mixins:
architect: [security-first, performance-focused]When a cognitive persona is applied, its thinking framework is injected into the agent's context at spawn time. The agent retains all its original responsibilities but evaluates decisions through the additional cognitive lens.
Quick Comparison
| Concept | What it defines | Where it lives | How many |
|---|---|---|---|
| Agent | Role, model, write scope, responsibilities | packages/core/agents/*.md | 13 |
| Cognitive persona | Thinking style, decision framework | packages/core/personas/cognitive/*.md | 3 |
| Protocol | Phase sequence, which agents to spawn | packages/core/protocols/*.yaml | 7 |
Further Reading
- Personas Guide -- detailed walkthrough of the persona system
- Personas Reference -- full text of each cognitive persona