superclaude-org--superclaude_framework
kazukiSuperClaude Plugin for Claude Code
Eval-gated toolkit — a few skills that beat native behavior, deterministic hooks, and optional MCP integrations. Less framework, more leverage.
Installation
From marketplace (when published)
/plugin marketplace add SuperClaude-Org/SuperClaude_Framework
/plugin install superclaude@SuperClaude-Org/SuperClaude_Framework --scope user
Local development
claude --plugin-dir ./plugins/superclaude
What's Included
4 Skills
| Skill | Auto-triggers on |
|---|---|
confidence-check |
Pre-implementation readiness gate (+ confidence.ts) |
spec-panel |
Reviewing specs, requirements, API contracts via expert panel |
socratic |
"Teach me / help me understand" — questions instead of answers |
pm-reflexion |
Resuming cross-session work, post-mortems after failures |
1 Agent
explore-haiku— cheap codebase exploration on Haiku
5 Hooks (hooks/hooks.json)
| Hook | Event | Behavior |
|---|---|---|
session-restore |
SessionStart | TASK.md excerpt, git state, mindbase probe (silent degrade) |
confidence-gate |
PreToolUse (Write|Edit) |
Confidence criteria reminder for non-trivial work |
session-summary |
Stop | Outcome + remaining-work summary |
reflexion-trigger |
Stop | Failure reflexion via pm-reflexion skill |
tab-title |
session lifecycle | Terminal tab state, opt-in via SUPERCLAUDE_TAB_TITLE=1 |
MCP Servers (optional)
.mcp.json ships optional server configs. Everything works with zero MCP
servers; see the repository README for the transparency policy on
Agile-Tech-maintained integrations.
New components must pass the eval gate in eval/ at the repository root —
see eval/preregister.yaml for the survive rules.
Version
5.0.0-alpha.1