By Archit3115
Ultra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.
Generate terse caveman-style commit message
Drop the always-on caveman activation rule into the current repo for every IDE agent
One-line code review comments
Real session token usage + lifetime savings + USD. Tweetable line via --share.
Switch caveman intensity level (lite/full/ultra/wenyan)
Surgical 1-2 file edit. Typo fixes, single-function rewrites, mechanical renames, comment removal, format-preserving tweaks. Hard refuses 3+ file scope. Returns caveman diff receipt. Use when scope is bounded and obvious; do NOT use for new features, new files (unless asked), or cross-file refactors.
Read-only code locator. Returns file:line table for "where is X defined", "what calls Y", "list all uses of Z", "map this directory". Output is caveman-compressed so the main thread eats ~60% fewer tokens than vanilla Explore. Refuses to suggest fixes.
Diff/branch/file reviewer. One line per finding, severity-tagged, no praise, no scope creep. Output format `path:line: <emoji> <severity>: <problem>. <fix>.` Use for "review this PR", "review my diff", "audit this file". Skips formatting nits unless they change meaning.
Decision guide for delegating to caveman-style subagents. Tells the main thread WHEN to spawn `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit), or `cavecrew-reviewer` (diff review) instead of doing the work inline or using vanilla `Explore`. Subagent output is caveman-compressed so the tool-result injected back into main context is ~60% smaller — main context lasts longer across long sessions. Trigger: "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".
Ultra-compressed commit message generator. Cuts noise from commit messages while preserving intent and reasoning. Conventional Commits format. Subject ≤50 chars, body only when "why" isn't obvious. Use when user says "write a commit", "commit message", "generate commit", "/commit", or invokes /caveman-commit. Auto-triggers when staging changes.
Compress natural language memory files (CLAUDE.md, todos, preferences) into caveman format to save input tokens. Preserves all technical substance, code, URLs, and structure. Compressed version overwrites the original file. Human-readable backup saved as FILE.original.md. Trigger: /caveman-compress FILEPATH or "compress memory file"
Quick-reference card for all caveman modes, skills, and commands. One-shot display, not a persistent mode. Trigger: /caveman-help, "caveman help", "what caveman commands", "how do I use caveman".
Ultra-compressed code review comments. Cuts noise from PR feedback while preserving the actionable signal. Each comment is one line: location, problem, fix. Use when user says "review this PR", "code review", "review the diff", "/review", or invokes /caveman-review. Auto-triggers when reviewing pull requests.
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Seahorse turns Claude Code into an advisor→executor orchestration layer: one architect model plans, cheaper specialist models build, every unit of work is routed to the model that does it best — under a strict token diet, over a live project knowledge graph, with typeset output.
Seahorse = the whole framework this repo installs. The token-discipline layer (caveman = talk less, ponytail = write less code) ships vendored in
vendor/.
| Concern | Tooling |
|---|---|
| Talk less / write less code | caveman + ponytail (vendored, MIT) |
| Plan → delegate → verify | Fable advisor → Sonnet/Opus executors → GPT (Codex) review |
| Research | /research, /deep-research, dynamic Workflows, arXiv (alphaXiv MCP), primary sources |
| Knowledge graph | graphify → OKF |
| Hold work across turns | /goal, /workflows |
| Outputs | PDFs = LaTeX/tectonic, diagrams = Mermaid |
| CI/CD | GitHub Actions templates (lint → type → test → build) |
See docs/architecture.md for the flow diagrams and the full routing table.
Seahorse ships two ways, because Claude Code has two extension models and each has a trade-off.
/seahorse:seahorseclaude plugin marketplace add Archit3115/seahorse
claude plugin install seahorse@seahorse
# optional companion token-discipline plugins (vendored in this repo):
claude plugin install caveman@seahorse ponytail@seahorse
Or ask Claude to do it for you after cloning — those are the two commands it runs.
Auto-install on clone: this repo carries a .claude/settings.json that pre-registers the marketplace
and enables the plugins. When you open Claude Code inside the cloned repo and trust the folder, Claude
Code can offer to install them without you typing the commands. This only affects sessions started in this
directory, and — being third-party code — it still asks for your consent. There is no silent
install-on-git clone; that would be a security hole, and Claude Code does not do it. If your version
doesn't prompt, run the two commands above.
Local dev without installing: claude --plugin-dir /path/to/seahorse.
~/.claude config — gives you the bare /seahorse commandPlugin skills are always namespaced (/seahorse:seahorse). If you want the bare /seahorse, run the
installer, which copies the contract, agents, and commands into ~/.claude:
./install.sh
It merges (never clobbers) your global CLAUDE.md, installs the four agents and four commands (including
the bare /seahorse), and wires the SessionStart bootstrap hook. Optional tooling it points you at:
tectonic (LaTeX PDFs), uv + graphifyy (knowledge graph), codex CLI + codex login (GPT review).
Which to pick: plugin for clean, versioned, shareable install; installer if you specifically want the
bare /seahorse and the global contract. They coexist — a --plugin-dir copy wins for its session.
seahorse/
├── .claude-plugin/
│ ├── plugin.json # plugin manifest (the seahorse plugin)
│ └── marketplace.json # marketplace: seahorse + vendored caveman + ponytail
├── .claude/settings.json # auto-install pre-config (trust-prompt, this dir only)
├── commands/ # /seahorse · /research · /kg · /pdf (namespaced under the plugin)
├── agents/ # advisor · researcher · builder-heavy · builder-light
├── hooks/
│ ├── hooks.json # plugin SessionStart hook
│ ├── seahorse-activate.sh # injects the compact operating contract
│ └── seahorse-bootstrap.sh # standalone-install project-bootstrap signal (gated to ~/Work)
├── contract/
│ ├── CLAUDE.global.md # → ~/.claude/CLAUDE.md (standing contract)
│ ├── CLAUDE.project.template.md # per-project contract template
│ └── settings.snippet.json # SessionStart hook snippet for the standalone install
├── vendor/ # caveman + ponytail forks (MIT, attribution preserved)
├── benchmarks/ # SWE-bench harness: Fable vs Opus vs Sonnet vs Seahorse
├── docs/architecture.md # Mermaid diagrams + routing table
├── templates/ci/ # node.yml · python.yml
└── install.sh # standalone ~/.claude installer
npx claudepluginhub archit3115/seahorse --plugin cavemanAdvisor→executor orchestration for Claude Code. One architect model (Fable) plans, cheaper specialists (Sonnet/Opus) build, GPT reviews — routed to the model that does each chunk best, under a token diet, over a live knowledge graph, with typeset output.
Lazy senior dev mode. Forces the simplest, shortest solution that actually works: YAGNI, stdlib first, no unrequested abstractions.
Ultra-compressed communication mode. Cuts 65% of output tokens (measured) while keeping full technical accuracy by speaking like a caveman.
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Unified capability management center for Skills, Agents, and Commands.
Memory compression system for Claude Code - persist context across sessions
Standalone image generation plugin using Nano Banana MCP server. Generates and edits images, icons, diagrams, patterns, and visual assets via Gemini image models. No Gemini CLI dependency required.