Skip to content

install-by-platform.md

Latest commit

 

History

History
373 lines (252 loc) · 10.6 KB

install-by-platform.md

File metadata and controls

373 lines (252 loc) · 10.6 KB
Raw

Install by Platform

Step-by-step install guide for each supported platform. One command per platform — then run any aicrew action immediately.

Source of truth: ~/Agents/ (populated for every platform) Full action matrix: platform-entry-points.md

Claude Code

Install

npx aicrew install claude
# or, after cloning and installing globally:
aicrew install claude

What gets installed

Location Contents
~/Agents/ Shared source of truth — commands, agents, hooks, docs
~/.claude/commands/ Symlinks → ~/Agents/commands/*.md (slash commands)
~/.claude/skills/ Merged copy of ~/Agents/ (Claude skill reader)
~/.claude/settings.json Merged hook entries (session-memory.py, security-guard.py)
~/.claude/.mcp.json Symlink → config/mcp/claude.json

First command to run

/dev I want to add a login page with email + password

Available slash commands

Core:

Command What it does
/dev Full 9-phase TDD pipeline
/fix Fast bug-fix flow
/quick Scout → Act (token-efficient)
/brainstorm 3 design options with trade-offs

Utilities:

Command What it does
/conclude Wrap up session, capture learnings
/update-skills Evolve project-level skill overrides
/harness-audit Audit skill health and hook wiring
/session Name this task (avoids state file collisions)
/handoff Cross-tool state checkpoint
/benchmark Token savings estimate

Setup / maintenance:

Command What it does
/install Re-run install / update
/update Pull latest skills from package
/status Show install state per platform
/agent-kit Scaffold .mdc rules shared across repos
/cursor-plugin Scaffold Cursor multi-tool extension

Output style: Terse by default (caveman). Use /normal to go verbose; /lean on or /terse to return to terse.

Hooks

Hook Event Role
session-memory.py Stop Journals changed files; strips <private> blocks
security-guard.py PreToolUse Blocks secrets; warns on risky patterns

Tune with ECC_HOOK_PROFILE=minimal|standard|strict.

MCP

~/.claude/.mcp.json → config/mcp/claude.json (symlinked). Core servers: codebase-memory-mcp, context-mode, token-optimizer-mcp.

aicrew install only wires the config — it does not install server binaries. After install, run:

npm install -g codebase-memory-mcp   # graph index binary (required for graph queries)
npm install -g token-optimizer-mcp   # optional token budgeting server
# context-mode: no install needed — auto-downloaded via npx on first use

Or run aicrew install mcp for the full checklist.

Interactive checkpoints

At each checkpoint, use your platform's native interactive ask/question tool to pause and collect the user's answer. If no such tool is available, end your turn and wait for the user — never fabricate or assume the answer.

Known tool (use if available): AskUserQuestion.

Cursor

Install

npx aicrew install cursor
# or after global install:
aicrew install cursor

What gets installed

Location Contents
~/Agents/ Shared source of truth
~/.cursor/mcp.json Symlink → config/mcp/cursor.local.json
config/mcp/cursor.local.json Seeded from template (gitignored — fill real API keys here)

First command to run

Ask the Cursor agent with the shared ~/Agents/ rules:

/dev I want to add a login page with email + password

Or use the CLI scaffold commands:

aicrew agent-kit init        # share .mdc rules across repos
aicrew cursor-plugin init    # scaffold a multi-tool terminal extension

Available actions

Action How
/dev, /fix, /quick, /conclude, etc. Via Claude integration (rules in ~/Agents/)
MCP tools Wired via ~/.cursor/mcp.json
Agent-kit rules Run aicrew agent-kit init → shares .mdc across repos
Cursor plugin Run aicrew cursor-plugin init → scaffolds local extension

MCP

~/.cursor/mcp.json symlinks to config/mcp/cursor.local.json. config/mcp/cursor.json is the committed template (no secrets). Fill API keys in cursor.local.json (gitignored).

aicrew install only wires the config — install the core servers first:

npm install -g codebase-memory-mcp   # required for graph queries
npm install -g token-optimizer-mcp   # optional; Cursor config uses the global install path
# context-mode: no install needed — auto-downloaded via npx on first use

Optional servers: GitHub, filesystem, memory, Brave, Playwright, SQLite, Postgres, GitKraken, Perplexity.

Interactive checkpoints

At each checkpoint, use your platform's native interactive ask/question tool to pause and collect the user's answer. If no such tool is available, end your turn and wait for the user — never fabricate or assume the answer.

Known tool (use if available): askFollowupQuestion.

Codex

Install

npx aicrew install codex
# or after global install:
aicrew install codex

What gets installed

Location Contents
~/Agents/ Shared source of truth
~/.codex/skills/ Codex-native skill folders (aicrew-dev, aicrew-fix, …)
~/.codex/config.toml Patched with MCP server entries

First command to run

Invoke a skill by name in Codex:

aicrew-dev

Available Codex skills

Core:

Skill name Equivalent
aicrew-dev /dev
aicrew-fix /fix
aicrew-quick /quick
brainstorm /brainstorm

Utilities:

Skill name Equivalent
aicrew-conclude /conclude
aicrew-update-skills /update-skills
aicrew-harness-audit /harness-audit
aicrew-session /session
aicrew-handoff /handoff
aicrew-benchmark /benchmark

Setup / maintenance:

Skill name Equivalent
aicrew-install /install
aicrew-update /update
aicrew-status /status
aicrew-agent-kit /agent-kit
aicrew-cursor-plugin /cursor-plugin

Output style: Terse by default (caveman). Use aicrew-normal to go verbose; lean or aicrew-terse to return to terse.

MCP (config.toml)

aicrew install codex patches ~/.codex/config.toml with MCP server entries. Core: codebase-memory-mcp, context-mode, token-optimizer-mcp.

aicrew install only wires the config — install the core servers first:

npm install -g codebase-memory-mcp   # required for graph queries
# context-mode: no install needed — auto-downloaded via npx on first use

Interactive checkpoints

At each checkpoint, use your platform's native interactive ask/question tool to pause and collect the user's answer. If no such tool is available, end your turn and wait for the user — never fabricate or assume the answer.

Known tool (use if available): native ask tool (e.g. ask_human).

Gemini CLI

Install

npx aicrew install gemini
# or after global install:
aicrew install gemini

What gets installed

Location Contents
~/Agents/ Shared source of truth (always populated)

Gemini CLI prints post-install instructions for wiring commands manually (varies by Gemini CLI version).

Wire manually

Point your Gemini CLI commands config to ~/Agents/commands/ or reference files directly:

# Example: reference a command inline
cat ~/Agents/commands/dev.md

First command to run

After wiring, use the command text from ~/Agents/commands/dev.md as your Gemini prompt, or use it as a slash command if your Gemini CLI version supports that.

Available actions

All commands in ~/Agents/commands/ are available: /dev, /fix, /quick, /conclude, /update-skills, /harness-audit, /session, /handoff, /benchmark, /brainstorm, /lean on, /lean off, /terse, /normal.

Interactive checkpoints

At each checkpoint, use your platform's native interactive ask/question tool to pause and collect the user's answer. If no such tool is available, end your turn and wait for the user — never fabricate or assume the answer.

Known tool (use if available): native ask tool (e.g. ask_human).

Antigravity

Install

Antigravity reads ~/Agents/commands/ directly.

npx aicrew install           # populates ~/Agents/ (all platforms)
# or:
aicrew install               # same

aicrew install gemini also works — both just populate ~/Agents/.

What to configure

In your Antigravity config, point the commands reference to ~/Agents/commands/.

First command to run

/dev I want to add a login page with email + password

Available actions

All slash commands from ~/Agents/commands/: /dev, /fix, /quick, /conclude, /update-skills, /harness-audit, /session, /handoff, /brainstorm, /lean on, /lean off, /terse, /normal.

Interactive checkpoints

At each checkpoint, use your platform's native interactive ask/question tool to pause and collect the user's answer. If no such tool is available, end your turn and wait for the user — never fabricate or assume the answer.

Known tool (use if available): native ask tool.

All platforms at once

npx aicrew install         # installs for all supported platforms
npx aicrew install all     # same

Verify any platform

aicrew status

Expected output:

=== aicrew status ===

Shared assets (~/Agents/):     ✓ populated
Claude Code:                   ✓ ~/.claude/commands/ (17 symlinks), hooks registered
Cursor:                        ✓ ~/.cursor/mcp.json linked
Codex:                         ✓ ~/.codex/skills/ (18 skill folders)
Gemini CLI:                    ⚠ ~/Agents/ populated; manual Gemini config needed

Re-install / update skills

aicrew update               # pull latest skills from package → ~/Agents/
# or per platform:
aicrew install claude       # re-runs claude install (merge, not clobber)
aicrew install codex        # re-runs codex install

See also