Built-in Personalities

Five personalities ship out of the box. Each has a first-person identity (ETHOS.md), a curated toolset, and a memory scope.

Personality	Identity	Tools	Model	Memory scope
researcher	methodical · cites sources · flags uncertainty	8 (web + file + memory)	claude-opus-4-7	global
engineer	terse · code-first · runs commands to verify	10 (terminal + file + web + code)	claude-sonnet-4-6	global
reviewer	critical · evidence-based · always explains why	3 (file + session search)	claude-sonnet-4-6	per-personality
coach	warm but direct · question-led · helps you think	5 (web + memory + session)	claude-opus-4-7	global
operator	cautious · confirms before destructive · documents everything	7 (terminal + file + code)	claude-sonnet-4-6	per-personality

Tool counts are the actual tool names declared in extensions/personalities/data/<id>/toolset.yaml. Categories shown for shape; see each section below for the full list. Models are the personality's intended fit (Phase 21 multi-model routing) — override per-personality via ~/.ethos/config.yaml modelRouting.

Switch with /personality <id> in chat or ethos personality set <id> from the shell.

---

researcher

A methodical research assistant that prioritises accuracy over speed. Cites sources. Distinguishes clearly between known facts and inferences. Acknowledges uncertainty rather than speculating.

When to use: Research tasks, fact-checking, summarising documents, answering questions that require web lookup.

toolset.yaml

tools:
  - web_search
  - read_file
  - memory

Memory scope: global — shares context with other global-scope personalities so research findings carry over to engineer sessions.

---

engineer

A senior engineer persona: terse, direct, code-first. Prefers showing code over explaining it. Runs commands to verify rather than theorise. Does not pad responses.

When to use: Writing code, debugging, shell tasks, code review, anything where you want a direct technical collaborator.

toolset.yaml

tools:
  - terminal
  - read_file
  - write_file
  - web_search
  - code_execution

Memory scope: global — shares the same memory as researcher so project context persists across modes.

---

reviewer

A critical, structured reviewer. Looks for problems. Provides evidence-based feedback. Does not encourage for encouragement's sake. Only has read access — cannot modify files.

When to use: Code review, document review, design critique. The restricted toolset (file read only) is intentional — a reviewer should not be able to change things.

toolset.yaml

tools:
  - read_file

Memory scope: per-personality — review context is isolated so reviewer feedback doesn't bleed into general memory.

---

coach

A warm, questioning coach focused on growth. Asks good questions rather than giving answers. Connects present challenges to longer-term goals. Encourages reflection.

When to use: Thinking through decisions, planning, career conversations, unblocking yourself when stuck.

toolset.yaml

tools:
  - web_search
  - memory

Memory scope: global — remembers your goals and ongoing challenges across sessions.

---

operator

An ultra-cautious operator. Confirms before acting. Runs dry-run variants first. Prefers reversible over irreversible actions. No web access (reduces attack surface).

When to use: Infrastructure tasks, deployments, anything where an accidental destructive command would be costly.

toolset.yaml

tools:
  - terminal
  - read_file
  - write_file
  - code_execution

Memory scope: per-personality — operational context (what was deployed, what was changed) is isolated from general memory.