Inspirations

Inspirations

Chimera ships seven coding-agent CLIs, each modelled on a real-world tool. The “inspired by” column below names the upstream that motivated the design. Chimera reimplements the posture — sandbox-first, server-first, etc. — using its own primitives (the 8-phase architecture documented in docs/architecture.md).

Codename	Alias	Inspired by	One-liner
mink	tui	Claude Code (Anthropic)	TUI-first interactive coding agent; ingests ~/.claude/settings.json
otter	multi	opencode	Server-first, multi-client (HTTP + SSE + ACP); plugin-driven
ferret	sandbox	codex (OpenAI)	Sandbox-first execution + IDE-flagship + OpenAI-default provider chain
weasel	mini	pi (pi-mono)	Minimal harness, four operating modes (interactive / print / RPC / SDK)
shrew	tiny	little-coder	Tuned for small local models (llama.cpp, Qwen MoE, scaffold-fit)
stoat	shell	kimi-cli (Moonshot)	Shell-mode toggle (Ctrl-X); Kimi-tuned defaults
badger	strict	claw-code	Harness-rewrite posture: tighter defaults, parity tracking, --rerun-on-failure

Codenames are stable shipping names (used in commits, release notes, parity matrices). Aliases are for discoverability: chimera mink and chimera tui are exactly equivalent.

Per-codename detail

mink ↔ tui (Claude Code)

The TUI-first ergonomic. Lives at chimera/mink/ with a slash-command palette mirroring the upstream’s /help, /clear, /cost, /context, etc. Reads ~/.claude/settings.json for hooks + permissions + MCP servers. Ships per-CLI presets, sessions list/show/cost/share, run persistence, and a stream-json output mode.

Chimera’s mink adopted the TUI ergonomics + settings ingest. It diverged on persistence (event-sourced JSONL + SQLite snapshots) and provider routing (multi-provider chain, not Anthropic-only).

See docs/mink/parity-matrix.md.

otter ↔ multi (opencode)

The server-first, multi-client posture. Lives at chimera/otter/. Default serve transport is HTTP + SSE on port 5173; alternative is ACP (serve --acp) for IDE-style JSON-RPC over stdio. Plugin-driven extension model with ~/.opencode/plugin/ discovery, MCP server config ingest, custom .opencode/command/*.md slash commands.

Chimera’s otter adopted the server + multi-transport architecture. It diverged on file layout (eventlog under ~/.chimera/eventlog/otter-* rather than upstream’s session DB) and on the default cost-tracking surface.

See docs/otter/parity-matrix.md.

ferret ↔ sandbox (codex)

The sandbox-first, IDE-flagship posture. Lives at chimera/ferret/. Every shell command runs through a SandboxedEnvironment by default (read-only / workspace-write / workspace-write-network); approval presets bundle the entire permission stance into one flag. ACP is the default serve transport (IDE-shaped notifications: code/diff, editor/open_file, terminal/output); HTTP is opt-in via --http. macOS adds OS-level sandbox-exec wrapping.

Chimera’s ferret adopted the sandbox + IDE ergonomics + OpenAI-default provider chain. It diverged on the language (Python on Chimera primitives rather than the upstream’s Rust core).

See docs/ferret/parity-matrix.md.

weasel ↔ mini (pi-mono)

The minimal-harness posture. Lives at chimera/weasel/. Four operating modes (interactive REPL / one-shot print / RPC stdio / embeddable SDK). Skips features like sub-agents and plan mode by default; ships powerful defaults that can be extended via .weasel/extensions/*.{js,ts,py} (Python today, Node subprocess for JS/TS). Theme + prompt-template registries.

Chimera’s weasel adopted the four-mode architecture + npm-extensible philosophy. It diverged by translating the TS/JS extension model to a Python-first one with a Node bridge for shipped TS extensions.

See docs/weasel/parity-matrix.md.

shrew ↔ tiny (little-coder)

The small-model-tuned posture. Lives at chimera/shrew/. Built on top of weasel; pins small-model defaults (qwen3.6-35b-a3b via llama.cpp, --max-steps 30, restricted tool subset). Curated skill markdowns under chimera/shrew/skills/{knowledge,protocols,tools}/. Small-model-fit extensions (MoE-aware context sizing, scaffold-fit prompt wrapping, tool filtering for sub-9B models).

Chimera’s shrew adopted the scaffold-model-fit thesis (skills + extensions + tighter budgets). It diverged on packaging (Python module vs the upstream’s npm package).

See docs/shrew/parity-matrix.md.

stoat ↔ shell (kimi-cli)

The shell-mode-toggle posture. Lives at chimera/stoat/. The signature feature is /shell (or Ctrl-X) — flips the REPL between agent mode (stoat> prompt, prompts feed the LLM) and shell mode (stoat$ prompt, inputs run as bash -c). Kimi-first provider chain ($MOONSHOT_API_KEY → kimi-k2.6 → fallbacks).

Chimera’s stoat adopted the shell-mode toggle. It diverged on language (Python harness on Chimera primitives rather than the upstream’s Python harness on different primitives).

See docs/stoat/parity-matrix.md.

badger ↔ strict (claw-code)

The harness-rewrite posture. Lives at chimera/badger/. Tighter defaults (--max-steps 25 vs 50 elsewhere), opt-in --rerun-on-failure with bounded retries, and a distinctive chimera badger parity --against PARITY.md subcommand that diffs live agent state against a declared schema (rc=0 match / 1 drift / 2 usage error).

Chimera’s badger adopted the “better harness tools” thesis + parity-tracker UX. It diverged on language (Python vs the upstream’s Rust port).

See docs/badger/parity-matrix.md.

Trademark policy

Live source / docs / help text in Chimera never embeds upstream brand strings. This document and chimera agents discovery output are explicit factual references — they exist so users and contributors can know what they’re getting — and are excluded from each CLI’s trademark scrub via a SKIP_FILES regex.

See docs/<cli>/security-and-trademarks.md for the per-CLI policy and scripts/<cli>_trademark_scrub.sh for the enforcement script.

See also

• Architecture (8 phases) — how all 7 CLIs compose from the same Chimera primitives.
• Coding agents overview — comparison table + use-case decision guide.
• Per-CLI quickstarts under docs/<cli>/quickstart.md.