Skip to content
SSCodeAI
日本語English

Projects

SSCodeAI's public projects and related repositories.

Updated View as Markdown

Projects

superpowers-ja

A package that adds planning, TDD, debugging, review, verification, and Japanese development workflows to AI coding agents.

agency-agents-ja

An AI specialist agent library for Japanese IT development, SIer, custom development, in-house services, SaaS, manufacturing DX, and public sector.

sscodeai-site

This website. An Astro + Nimbus based documentation, blog, tutorial, and video hub.

takumi

A model-agnostic orchestration platform for verifiable agentic software delivery.

  • Repository: sscodeai/takumi
  • Docs: takumi
  • Runtime: TypeScript / Node 20+
  • Core: workflows, traceability, evidence-native testing, agent evals, verify-before-done loops
  • Status: Developer Preview

img2threejs

An image-to-3D skill that reconstructs the object in a reference image as code-only procedural Three.js models.

keysmith

An MCP server and CLI for AI-agent-safe secret management, designed to keep plaintext secrets out of agent context.

  • Repository: sscodeai/keysmith
  • Docs: keysmith
  • Architecture: keysmith Architecture
  • Runtime: Go single static binary
  • Security model: age encrypted store, masked views, handle-passing, self-healing rotation, Vault short-TTL credentials

Proctor

An Agent Eval / LLMOps tool that evaluates recorded AI-agent trajectories and explains the failing step and root cause.

local-agent-toolkit

A local-first toolkit for agent memory, skill management, and run tracing.

aiitg

A zero-trust input security layer that detects hidden prompt injection, sanitizes it, labels trustworthiness, and enforces policy before untrusted documents reach an LLM.

  • Repository: sscodeai/aiitg
  • Docs: aiitg
  • Runtime: Python 3.11+
  • Core: 6 formats × 7 detectors, coordinate-level evidence, safe/caution/dangerous labels, allow/quarantine/human_approval/block policy, audit log and human-approval queue
  • Interfaces: aiitg CLI, MCP server (aiitg-mcp), Claude Code PreToolUse hook
  • Tests: 191 tests (make test)
  • License: MIT

ai-skill-maker

A meta-skill that turns rough intent, repository evidence, and repeatable workflows into AI skills with evidence labels, validation scripts, release gates, and refresh-safe user rule blocks.

  • Repository: sscodeai/ai-skill-maker
  • Docs: ai-skill-maker
  • Runtime: Node.js 18+
  • Output: functional / document / workflow skills, project maintainer skills, and adapter bundles for AGENTS.md, CLAUDE.md, Cursor, and Copilot
  • Guardrails: Trust Gate, 9,000-token file budget, protected core principles locked by SHA-256 fingerprint
  • License: Apache-2.0

ai-project-skill-maker

A meta-skill that generates durable project maintainer skills and instruction bundles for AI coding agents.

  • Repository: sscodeai/ai-project-skill-maker
  • Docs: ai-project-skill-maker
  • Runtime: Node.js scripts plus platform-neutral Markdown templates
  • Modes: genesis mode (new projects) and repo mode (existing repositories)
  • Evidence model: observed_fact, declared_intent, recommended_standard, inferred_assumption
  • License: Apache-2.0

skills-ja-JP

A Japanese localization of the Matt Pocock Agent Skills (sscodeai/skills) that reduces misread intent when working with coding agents in Japanese.

  • Repository: sscodeai/skills-ja-JP
  • Docs: skills-ja-JP
  • Content source of truth: sscodeai/skills
  • Install: npx skills@latest add sscodeai/skills-ja-JP
  • Skills: 20 (15 engineering + 5 productivity)
  • Policy: skill-guided content localization (directory names, skill names, commands, paths, URLs, and identifiers are preserved)
  • License: MIT

decision-workbench

A six-stage tool that expands the decision-maker’s cognition before they decide: it diverges options, reveals blind spots, structures the map, and hands the decision back to a human. It never picks a winner.

  • Repository: sscodeai/decision-workbench
  • Docs: decision-workbench
  • Runtime: Python (FastAPI) + React (mock mode needs no API keys)
  • Eval: against a hidden gold option set, single model 26% → naive multi-agent 78% → decision-workbench 100% (12 of 12)
  • Tests: 25 tests (uv run pytest)
  • License: MIT

decision-space-expansion

A three-stage decision methodology distilled from decision-workbench: widen the option space with a heterogeneous model committee, structure the divergence, fold constraints back in, and let a human decide.

  • Repository: sscodeai/decision-space-expansion
  • Docs: decision-space-expansion
  • Form: a SKILL.md in the Agent Skills format (no code, no dependencies)
  • Loop: Diverge → Structure → Converge (with a mandatory human gate)
  • Install: Claude Code, Cursor / Windsurf, Codex, Gemini CLI, or any Markdown-reading agent
  • License: MIT

eli5

A picture-first explainer generator that explains any topic with large SVG diagrams and minimal text, producing a single self-contained HTML file with zero external requests.

  • Repository: sscodeai/eli5
  • Docs: eli5
  • Spec: SKILL.md (procedure, language rules, HTML hard rules, verification checklist)
  • Output: single HTML file (interactive glossary, quiz, light/dark theme) plus PDF/PNG export
  • License: MIT

fusion-cache

A framework-agnostic LLM API caching layer that fuses exact, semantic, and prefix-cache accounting into one drop-in middleware.

  • Repository: sscodeai/fusion-cache
  • Docs: fusion-cache
  • Runtime: Python ≥ 3.10
  • Layers: L1 exact (canonicalized SHA-256 key), L2 semantic (embedding + cosine), L3 prefix accounting (DeepSeek ~31× discount)
  • Benchmark: 97% hit rate, P95 0.7 ms, P50 3.2 s → 0.1 ms
  • Extra: OpenAI-compatible gateway (fusion-cache serve), MCP server (fusion-cache-mcp), Redis store
  • License: Apache-2.0
Navigation

Type to search…

↑↓ navigate↵ selectEsc close