juggernaut
jpvelasco/juggernaut
Juggernaut is a one-command setup tool that configures Claude Code to use AWS Bedrock, with cross-platform support (bash/zsh/fish/PowerShell), dry-run mode, and validation scripts.
Instalação
npx convez-ai jpvelasco/juggernaut1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
Single cross-platform binary that configures Claude Code to route through Amazon Bedrock instead of Anthropic's direct API — IAM, SSO, or Bedrock API key auth.
Install from npm: juggernaut-bedrock — npm install -g juggernaut-bedrock
Why Bedrock?
| Feature | Direct Anthropic API | Amazon Bedrock |
|---|---|---|
| Billing | Separate Anthropic account | Consolidated AWS billing |
| Authentication | API keys only | IAM, SSO, roles, federation |
| Data Residency | Anthropic infrastructure | Your chosen AWS region |
| Compliance | Anthropic's certifications | AWS compliance (SOC, HIPAA, FedRAMP) |
| Network | Public internet | VPC endpoints, PrivateLink |
| Governance | Limited | IAM policies, CloudTrail, quotas |
Install
Via npm (recommended)
Published as juggernaut-bedrock on npm. Works everywhere Claude Code is installed.
npm install -g juggernaut-bedrock
curl (Unix / macOS / Linux / Git Bash / WSL):
curl -fsSL https://raw.githubusercontent.com/jpvelasco/juggernaut/latest/scripts/install.sh | bash
PowerShell (Windows):
irm https://raw.githubusercontent.com/jpvelasco/juggernaut/latest/scripts/install.ps1 | iex
Configure
# IAM / SSO (recommended)
juggernaut apply --auth=iam
# Bedrock API key (stored securely in OS keychain)
juggernaut apply
# Interactive first-run — omit flags for a guided prompt
juggernaut apply
juggernaut apply will not write CLAUDE_CODE_USE_BEDROCK=1 unless a valid credential source is confirmed.
Common options:
juggernaut apply --auth=iam --region=us-east-1
juggernaut apply --auth=iam --opusplan # Opus in /plan, Sonnet in execute
juggernaut apply --auth=iam --effort=xhigh # low | medium | high | xhigh | max
juggernaut apply --auth=iam --mode=auto # enable agentic safety-classifier mode
juggernaut apply --auth=iam --always-thinking # extended thinking on by default
juggernaut apply --auth=iam --service-tier=flex # Bedrock service tier: default | flex | priority
juggernaut apply --auth=iam --no-mantle # disable Mantle routing
juggernaut apply --auth=iam --dry-run # preview without writing
juggernaut apply --auth=iam --scope=project # write to ./.claude/settings.json
Launch
claude
Juggernaut installs a claude shim (~/.local/bin/claude on Unix, claude.cmd on Windows) that reads your bearer token from the OS keychain and injects it before launching Claude Code. No manual environment setup required.
Commands
| Command | Description |
|---|---|
apply | Write Juggernaut config to settings.json and install the launcher shim. |
show | Print the current Juggernaut block from user and project scopes. |
doctor | Read-only diagnostics — checks block, credentials, launcher shim. |
uninstall | Remove the Juggernaut block, bearer token, and launcher shim. |
migrate | Migrate from v3 (shell-based) to v4 (Go). Runs automatically on first apply. |
version | Print the installed version. |
Default Models
| Tier | Model | Global CRIS Profile |
|---|---|---|
| Primary | Claude Sonnet 4.6 | global.anthropic.claude-sonnet-4-6 |
| Opus | Claude Opus 4.8 | global.anthropic.claude-opus-4-8 |
| Fast / subagent | Claude Haiku 4.5 | global.anthropic.claude-haiku-4-5-20251001-v1:0 |
Override any tier:
juggernaut apply --auth=iam --opus-model=us.anthropic.claude-opus-4-8
juggernaut apply --auth=iam --model=global.anthropic.claude-sonnet-4-6 # override all
Effort Levels
Controls adaptive thinking depth. Valid for all Claude 4 models on Bedrock.
| Level | Behavior |
|---|---|
low | Minimal thinking — fastest, lowest cost |
medium | Moderate thinking |
high | Almost always thinks |
xhigh | Always thinks deeply (default) |
max | Maximum thinking — deepest reasoning, highest cost |
juggernaut apply --auth=iam --effort=max
On Opus 4.8 and 4.7, only adaptive thinking is supported. Manual thinking mode is rejected by the API.
Permission Modes
Controls how Claude Code handles tool-use approvals. Set with --mode.
| Mode | Behavior |
|---|---|
default | Prompts for permission on each action (Claude Code default) |
acceptEdits | Auto-approves file edits and common filesystem commands |
plan | Propose changes only — no execution without explicit approval |
auto | Agentic safety classifier — auto-approves safe actions, blocks destructive ones |
dontAsk | Auto-deny unless pre-approved via rules |
bypassPermissions | Skip all prompts — containers/VMs only |
juggernaut apply --auth=iam --mode=auto
Bedrock note:
automode requiresCLAUDE_CODE_ENABLE_AUTO_MODE=1. Juggernaut sets this automatically — no manual env var needed.
What Gets Written
Juggernaut writes only to ~/.claude/settings.json (user scope) or ./.claude/settings.json (project scope). No shell profile modification.
{
"juggernaut": { "auth": { "mode": "iam", "region": "us-west-2" }, "meta": { ... } },
"effortLevel": "xhigh",
"skipWebFetchPreflight": true,
"permissions": { "defaultMode": "default" },
"env": {
"CLAUDE_CODE_USE_BEDROCK": "1",
"AWS_REGION": "us-west-2",
"CLAUDE_CODE_MAX_OUTPUT_TOKENS": "32768",
"MAX_THINKING_TOKENS": "65536",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "global.anthropic.claude-opus-4-8",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "global.anthropic.claude-sonnet-4-6",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "global.anthropic.claude-haiku-4-5-20251001-v1:0",
"CLAUDE_CODE_EFFORT_LEVEL": "xhigh",
"ENABLE_PROMPT_CACHING_1H": "1",
"DISABLE_TELEMETRY": "1"
}
}
Migrating from v3
If you have a v3 (shell-based) installation, just install v4 and run:
juggernaut migrate
Or simply run juggernaut apply — migration runs automatically on first use when a v3 config is detected. Credentials are transferred, shell launcher blocks are removed from your profiles, and no re-entry of credentials is required.
Minimum supported migration source: v3.2.3. If you're on an older version, upgrade the shell version first:
curl -fsSL https://raw.githubusercontent.com/jpvelasco/juggernaut/v3.2.3/install.sh | bash
The v3 shell-based release is preserved on the
legacy/v3branch and taggedv3.2.3. All v3 install URLs continue to work.
IAM Permissions
{
"Effect": "Allow",
"Action": [
"bedrock:InvokeModel",
"bedrock:InvokeModelWithResponseStream",
"bedrock:ListInferenceProfiles"
],
"Resource": "*"
}
See iam-policy.json for the complete policy. For tighter security, restrict the resource to specific regions.
Uninstall
# Remove config and launcher shim
juggernaut uninstall
# Preview first
juggernaut uninstall --dry-run
# Remove everything including the shim binary
juggernaut uninstall --full
Troubleshooting
403 Access Denied — Complete the Anthropic model access request in the AWS Bedrock console.
Model not found — Use global inference profile IDs (Juggernaut does this by default).
Keychain unavailable on headless Linux — IAM auth works without a keychain. For Bedrock API key auth, use --storage=profile to store in a local file instead.
SSO session expired — aws sso login --profile=<your-profile> and re-run claude.
juggernaut doctor — always the first diagnostic step.
Notes
/loginand/logoutare disabled when using BedrockAWS_REGIONis set explicitly by Juggernaut (Claude Code does not read it from~/.aws/config)- Juggernaut is an independent tool, not affiliated with Anthropic or Amazon Web Services