~ / rigs / storytree-ai / storytree

storytree-ai/storytree

Storytree 0.3: a map for you and your agents

↗ GitHub ★ 1 Apache-2.0 updated today project Claude CodeCodex Claude plugin
share on X
ARCHETYPE
Orchestrator
A bench of specialised subagents. The main agent mostly delegates.
CONTEXT TAX · EVERY TURN
up to ~12.5k tokens
Claude Code loads MCP tools on demand, so the real cost is usually lower.
Heavy · median rig: 2.3k · breakdown
GUARDRAILS
2/5
Pre-tool screening hook · No YOLO mode · details
Quality 25/100 Guardrails v2

Copy this rig

# review before running: this installs third-party code
$ npx degit storytree-ai/storytree/.claude ./rig-storytree  # inspect, then merge into .claude/

MCP servers are added to Claude Code at local scope; env vars are shown as YOUR_… placeholders — we never store values. Files are fetched with degit into a separate folder so you can review before merging.

$ npx degit storytree-ai/storytree/.agents/skills/grill-me .claude/skills/grill-me
$ curl -fsSL --create-dirs -o .claude/agents/blind-reviewer.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/blind-reviewer.md
$ curl -fsSL --create-dirs -o .claude/agents/builder.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/builder.md
$ curl -fsSL --create-dirs -o .claude/agents/change-reviewer.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/change-reviewer.md
$ curl -fsSL --create-dirs -o .claude/agents/corpus-investigator.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/corpus-investigator.md
$ curl -fsSL --create-dirs -o .claude/agents/explorer.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/explorer.md
$ curl -fsSL --create-dirs -o .claude/agents/frontend-builder.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/frontend-builder.md
$ curl -fsSL --create-dirs -o .claude/agents/librarian-curator.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/librarian-curator.md
$ curl -fsSL --create-dirs -o .claude/agents/story-author.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/story-author.md
$ curl -fsSL --create-dirs -o .claude/agents/test-writer.md https://raw.githubusercontent.com/storytree-ai/storytree/main/.claude/agents/test-writer.md

Merge into .claude/settings.json (project) or ~/.claude/settings.json (user). Hook commands reference scripts in the source repo — copy those too.

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Edit|Write|MultiEdit|NotebookEdit",
        "hooks": [
          {
            "type": "command",
            "command": "node --import tsx packages/app-setup/src/bins/storytree-hook.ts claude-code --test-guard by-agent"
          }
        ]
      }
    ]
  }
}

MCP servers (1)

serversourceest. tokens
storytree local / custom 2.5k

Skills (1)

Subagents (9)

blind-reviewer
model: opus
Independent reviewer that records its own verdict, with everything it read, before it may see any other reviewer's; it never reads an arc's increment log, a parent or sibling increment, or another rev
builder
model: inherit
Writes the least production code that turns the test writer's red tests green for one increment of storytree's own development; never edits a test, and hands one it judges wrong back through the sessi
change-reviewer
model: sonnet
Quality control on one green change before the gate: a fresh-context reviewer that answers every live check and every contract its increment names from the review's brief alone, and records what it fi
corpus-investigator
model: sonnet
Read-only, single-question verifier: checks each claim in a brief against the authoritative source and returns findings, assumption violations and a short summary.
explorer
model: sonnet
Read-only search subagent: sweeps the repo and the library for a question whose answer's shape is unknown, and returns a short digest with file:line citations.
frontend-builder
model: sonnet
Builds a visual unit in two stages — green on the test writer's red geometry and behaviour tests, then a captured, principle-cited look that lands with its pictures — never editing a test or self-acce
librarian-curator
model: opus
The librarian's hat, put on at a landing: dedupe, link, keep the decision log true, graduate memory, drain the queues — structure and history, not work units.
story-author
model: opus
Authors and amends the plan — story › capability › contract — bounding each story as a vertical slice in one package, placing pieces by afferent coupling, wiring real dependencies, and never implement
test-writer
model: inherit
Writes the failing tests for one increment of storytree's own development, sees each red, commits them and returns a test-to-contract table; a hook lets it edit test code and nothing else.

Hooks (2)

eventmatcherruns
SessionStart*node packages/base-dev-loop/src/provision-worktree.mjs --hook
PreToolUseEdit|Write|MultiEdit|NotebookEditnode --import tsx packages/app-setup/src/bins/storytree-hook.ts claude-code --test-guard by-agent

Permissions

deny (0)
—
ask (0)
—
allow (3)
mcp__storytree
Bash(pnpm storytree:*)
Bash(pnpm -s storytree:*)

Similar rigs

copied ✓