Tool Outputs¶
This page documents what lazyai-cli writes for each configured companion tool, grounded in packages/cli/internal/adapter.
Comparison¶
| Tool | Root file | Config dir | Agents | Skills | Hooks | MCP |
|---|---|---|---|---|---|---|
| OpenCode | AGENTS.md |
.opencode/ |
Yes (.opencode/agents) |
Yes (.opencode/skills) |
Yes (.opencode/plugins/vibe-lab-hooks.js) |
opencode.json |
| Claude Code | AGENTS.md (plus CLAUDE.md compat handling) |
.claude/ |
Yes (.claude/agents) |
Yes (.claude/skills) |
Yes (.claude/hooks/*) |
.mcp.json (+ settings.local when local-secrets) |
| GitHub Copilot | .github/copilot-instructions.md |
.github/ |
Yes (.github/agents) |
Yes (.github/skills) |
Yes (.github/hooks/*) |
.vscode/mcp.json, optional ~/.copilot/mcp-config.json |
| Pi | AGENTS.md |
.pi/ |
Yes (.pi/agents) |
Yes (.pi/skills) |
none (.pi/extensions/*.ts) |
none |
| OMP | AGENTS.md |
.omp/ |
Yes (.omp/agents) |
Yes (.omp/skills) |
Yes (.omp/hooks/pre/*.ts) |
.omp/mcp.json |
| Antigravity | AGENTS.md |
.gemini/ + .agents/ |
none | Yes (.agents/skills) |
Yes (.gemini/hooks/lazyai/* and .agents/hooks.json) |
~/.gemini/config/mcp_config.json |
| Kiro | AGENTS.md |
.kiro/ |
Yes (.kiro/agents) |
Yes (.kiro/skills) |
Yes (.kiro/hooks/*.json) |
.kiro/settings/mcp.json |
OpenCode¶
Surface summary: emits OpenCode-native agents, skills, commands/modes, MCP, and managed hook plugin.
- Root instructions:
AGENTS.md. - Config directory:
.opencode/ - Config tree (key files):
.
├── AGENTS.md
└── .opencode/
├── agents/guide.md
├── agents/<agent>.md
├── skills/<skill>/SKILL.md
├── commands/<command>.md
├── modes/<mode>.md
├── plugins/vibe-lab-hooks.js
└── opencode.json
- Agents surface:
.opencode/agents/<name>.md. - Skills surface:
.opencode/skills/<name>/SKILL.md. - Commands / prompts / chat modes: commands in
.opencode/commands; chat modes in.opencode/modes; no prompt surface. - Hook runtime surface:
.opencode/plugins/vibe-lab-hooks.js. - MCP output files:
opencode.json(with merged managedmcp). - Global scope support:
yes. - Maturity note: stable.
Claude Code¶
Surface summary: emits root instructions plus AGENTS, agents, skills, commands, output styles, hooks, and MCP config.
- Root instructions:
AGENTS.md+ project/workspaceCLAUDE.mdcompatibility handling. - Config directory:
.claude/ - Config tree (key files):
.
├── AGENTS.md
└── .claude/
├── agents/<agent>.md
├── skills/<skill>/SKILL.md
├── commands/<command>.md
├── output-styles/<style>.md
├── hooks/block-destructive-shell.sh
├── hooks/objective-workflow-gate.sh
├── rules/typescript.md
└── settings.json
- Agents surface:
.claude/agents/<name>.md. - Skills surface:
.claude/skills/<name>/SKILL.md. - Commands / prompts / chat modes: commands in
.claude/commands; output styles in.claude/output-styles; no prompt/chat-mode directories. - Hook runtime surface:
.claude/hooks/*.shand hook wiring in.claude/settings.json. - MCP output files:
.mcp.json(project/workspace), optionally.claude/settings.local.jsonwhen local-secrets mode is enabled. - Global scope support:
yes.
GitHub Copilot¶
Surface summary: emits project instructions, agents, skills, prompts, chat modes, and MCP outputs.
- Root instructions:
.github/copilot-instructions.md. - Config directory:
.github/ - Config tree (key files):
.
├── .github/
│ ├── copilot-instructions.md
│ ├── agents/<agent>.agent.md
│ ├── instructions/<instruction>.md
│ ├── prompts/<prompt>.prompt.md
│ ├── hooks/*.json
│ ├── hooks/*.sh
│ ├── chatmodes/<chatmode>.chatmode.md
│ └── skills/<skill>/SKILL.md
└── .vscode/mcp.json
# if probe-gated global install succeeds:
# ~/.copilot/mcp-config.json
- Agents surface:
.github/agents/<name>.agent.md. - Skills surface:
.github/skills/<name>/SKILL.md. - Commands / prompts / chat modes: no command surface; prompts at
.github/prompts/*.prompt.md; chat modes at.github/chatmodes/*.chatmode.md. - Hook runtime surface:
.github/hooks/*.jsonand.github/hooks/*.sh(project/workspace only). - MCP output files:
.vscode/mcp.json, optional~/.copilot/mcp-config.json(global only when probe passes:copilotbinary or~/.copilot/present). - Global scope support:
yes(probe-gated).
Pi¶
Surface summary: emits agents, skills, and prompt templates under .pi/, plus extension-based safety hooks. No MCP config is written.
- Root instructions:
AGENTS.md. - Config directory:
.pi/ - Config tree (key files):
.
├── AGENTS.md
└── .pi/
├── agents/<agent>.md
├── skills/<skill>/SKILL.md
├── prompts/<prompt>.md
└── extensions/*.ts
- Agents surface:
.pi/agents/<name>.md(Pi subagent extension reads markdown agent definitions). - Skills surface:
.pi/skills/<name>/SKILL.md. - Commands / prompts / chat modes: no commands or chat modes; prompt templates at
.pi/prompts/*.md. - Hook runtime surface: none as a
.pi/hooksdirectory; safety hooks ship as Pi extensions at.pi/extensions/*.ts. - MCP output files: none.
- Global scope support:
no.
OMP¶
Surface summary: emits agents, skills, commands, prompts, hook factories, and MCP.
- Root instructions:
AGENTS.md. - Config directory:
.omp/ - Config tree (key files):
.
├── AGENTS.md
└── .omp/
├── agents/<agent>.md
├── skills/<skill>/SKILL.md
├── commands/<command>.md
├── prompts/<prompt>.md
├── hooks/pre/<factory>.ts
└── mcp.json
- Agents surface:
.omp/agents/<name>.md. - Skills surface:
.omp/skills/<name>/SKILL.md. - Commands / prompts / chat modes: commands in
.omp/commands; prompts in.omp/prompts; no chat modes. - Hook runtime surface:
.omp/hooks/pre/<name>.ts. - MCP output files:
.omp/mcp.json. - Global scope support:
yes. - Maturity note: stable (all emitted surfaces verified against authoritative OMP docs; see ../adapters/snapshots/beta-adapter-verification-2026-06.md).
- Non-goals:
.omp/tools/and.omp/extensions/are documented OMP-native discovery surfaces (seeomp://custom-tools.mdandomp://extension-loading.md) but are not emitted by LazyAI; executable-module generation is out of current product scope and user-authored tools/extension modules remain user-managed.
Antigravity¶
Surface summary: emits minimal .gemini settings + hook surface, .agents-scoped Agent Skills, and user-level Gemini MCP config.
- Root instructions:
AGENTS.md. - Config directory:
.gemini/ - Config tree (key files):
.
├── AGENTS.md
└── .gemini/
├── settings.json
└── hooks/lazyai/
├── block-destructive-shell.sh
└── objective-workflow-gate.sh
.agents/
├── hooks.json
└── skills/<skill>/SKILL.md
- Agents surface: none.
- Skills surface:
.agents/skills/<name>/SKILL.md. - Commands / prompts / chat modes: none.
- Hook runtime surface:
.gemini/hooks/lazyai/*.shand.agents/hooks.json. - MCP output files:
~/.gemini/config/mcp_config.json. - Global scope support:
no. - Maturity note: stable (all emitted surfaces verified against official Antigravity/Gemini docs; see ../adapters/snapshots/beta-adapter-verification-2026-06.md).
Kiro¶
Surface summary: emits agents, skills, prompt templates, native Kiro v3 hook JSON, and MCP config under .kiro/; no commands or chat modes.
- Root instructions:
AGENTS.md. - Config directory:
.kiro/ - Config tree (key files):
.
├── AGENTS.md
└── .kiro/
├── agents/<agent>.json
├── skills/<skill>/SKILL.md
├── prompts/<prompt>.md
├── hooks/block-destructive-shell.json
├── hooks/block-destructive-shell.sh
└── settings/mcp.json
- Agents surface:
.kiro/agents/<name>.json(JSON required by Kiro CLI v3; emitted viaRewriteAgentForKirowithtools/allowedToolsfrom canonicaltools:frontmatter). - Skills surface:
.kiro/skills/<name>/SKILL.md. - Commands / prompts / chat modes: prompts at
.kiro/prompts/*.md; no commands or chat modes. - Hook runtime surface:
.kiro/hooks/*.jsonplus any referenced hook scripts. Only source-verified Kiro v3 trigger mappings are emitted. - MCP output files:
.kiro/settings/mcp.json. - Global scope support:
yes. - Non-goals: no emitted
.kiro/specs/or.kiro/steering/; no repo-local permissions file; no direct.kiro/powers/output.
How outputs are generated¶
lazyai-cli compile regenerates MCP output from ./.ai/mcp.json for configured targets. The emit and surface rules come from:
packages/cli/internal/adapter/output_mapping.gopackages/cli/internal/adapter/output_contract.gopackages/cli/internal/adapter/capabilities.gopackages/cli/internal/adapter/<tool>.go
For baseline target-level behavior, see ../concepts/tools.md.