A lightweight, low-token-tax workflow plugin for Claude Code, Codex, Antigravity (agy), Cursor, and Kimi Code.
If heavyweight agent frameworks feel too rigid or burn too many tokens on routine work, tea-stack gives you the minimal production guardrails that actually prevent costly mistakes: ~4 KB of baseline rules, 5 focused skills, and a read-only independent code reviewer.
- Low ceremony, proportional to risk: Move straight to edits on clear small fixes; invoke structured planning and independent review only when changes cross modules or touch contracts.
- Low token tax, no silent truncation: Compact rules (~4 KB) and small on-demand skills (~1.5–3 KB each) stay well below host context limits.
- Real independence on reviews: Ships a dedicated read-only
code-reviewersub-agent so critical diffs are audited outside the authoring context. - Native multi-tool packaging: One clean repository supporting Claude Code, Codex, Antigravity, Cursor, and Kimi Code out of the box—plus a zero-magic fallback installer.
/plugin marketplace add futuretea/tea-stack
/plugin install tea-stack@tea-stack
agy plugin install https://github.com/futuretea/tea-stackcodex plugin marketplace add https://github.com/futuretea/tea-stack.git
codex plugin add tea-stack@tea-stack(Or symlink skills directly via ./install.sh --target codex)
/add-plugin https://github.com/futuretea/tea-stack
(Or symlink skills via ./install.sh --target cursor)
- TUI Plugin Manager: run
/pluginsinside Kimi Code and install fromhttps://github.com/futuretea/tea-stack.git(powered bykimi.plugin.json). - CLI Skills Directory: pass
--skills-dir /path/to/tea-stack/skills, or symlink once via./install.sh --target kimi.
# Preview symlinks without modifying anything
./install.sh --target all --dry-run
# Supported targets: claude | codex | cursor | kimi | all
./install.sh --target kimi
# Or install into any custom skills directory
./install.sh --skills-dir ~/.config/agents/skills
install.shonly symlinksskills/*and skips existing files/symlinks. It never overwrites your existing globalAGENTS.md/CLAUDE.md.
Task / Requirement
│
├─ Multi-file / Contract / Risky ──▶ write-plan (align scope & verification first)
│ │
└─ Small & Well-scoped ──────────────────┴──▶ Implement + Run Tests
│
Unexpected failure ◀─────┤
│ │
debug ▼
review-code (fresh session / code-reviewer)
│
▼
feature-ready (gates + PR summary)
| Path | Purpose |
|---|---|
skills/using-tea-stack/SKILL.md |
Lightweight workflow router & 5 non-negotiable guardrails |
skills/write-plan/SKILL.md |
One-page implementation plan before editing across modules or changing contracts |
skills/debug/SKILL.md |
Scientific debugging loop: reproduce -> hypothesize -> isolate -> minimal root-cause fix |
skills/review-code/SKILL.md |
Evidence-based P0/P1/P2 code review designed for independent review |
skills/feature-ready/SKILL.md |
Pre-PR readiness check (diff hygiene, deterministic gates, risk review, PR summary draft) |
agents/code-reviewer.md |
Read-only independent reviewer sub-agent definition |
rules/AGENTS.md |
Compact global rule baseline (~4 KB; auto-loaded by Antigravity plugin or merged manually) |
templates/PROJECT-AGENTS.md |
Fill-in-the-blank <150-line repo AGENTS.md / CLAUDE.md template |
examples/ |
Filled-in repo AGENTS.md example + concrete review-code output example |
docs/TEAM-PLAYBOOK.md |
Team rollout guide, anti-patterns, and 2-week pilot plan (中文版) |
- Claude Code:
.claude-plugin/plugin.json+.claude-plugin/marketplace.json - Antigravity (
agy):.antigravity-plugin/plugin.json - Codex:
.codex-plugin/plugin.json+.agents/plugins/marketplace.json - Cursor:
.cursor-plugin/plugin.json+.cursor-plugin/marketplace.json - Kimi Code:
kimi.plugin.json
- Install the plugin in your preferred coding agent above.
- Copy
templates/PROJECT-AGENTS.mdto your repo root asAGENTS.md(orCLAUDE.md) and fill in your build/test/lint commands (seeexamples/project-agents-example.md). - Optionally merge clauses you like from
rules/AGENTS.mdinto your global instructions.