direnv

name	direnv
description	Set up direnv/.envrc for per-project Claude Code environments — switch API keys, providers, models, and feature flags automatically when entering a directory. Use when the user wants to manage multiple Claude accounts, switch between Anthropic/Bedrock/Vertex per project, or configure per-directory environment variables.
allowed-tools	Bash, Read, Write, Edit, Glob

direnv — Per-Project Claude Code Environments

Set up .envrc files so Claude Code automatically loads the right API keys, provider, model, and feature flags when you cd into a project directory.

Prerequisites

Check that direnv is installed and hooked into the shell:

which direnv && direnv version
# If missing: suggest installation for their platform

Check the shell hook is active:

grep -q 'direnv' ~/.zshrc ~/.bashrc 2>/dev/null && echo "hook found" || echo "hook missing"

If the hook is missing, add it:

• zsh: echo 'eval "$(direnv hook zsh)"' >> ~/.zshrc
• bash: echo 'eval "$(direnv hook bash)"' >> ~/.bashrc

Creating a .envrc

Ask the user which scenario they need, then generate the .envrc:

Direct API Key (personal/hobby)

# .envrc
export ANTHROPIC_API_KEY="sk-ant-..."

AWS Bedrock

# .envrc
export CLAUDE_CODE_USE_BEDROCK=1
export AWS_PROFILE=my-bedrock-profile
export AWS_REGION=us-east-1
# Optional: override model
# export ANTHROPIC_MODEL="us.anthropic.claude-sonnet-4-20250514-v1:0"

Google Vertex AI

# .envrc
export CLAUDE_CODE_USE_VERTEX=1
export CLOUD_ML_REGION=us-east5
export ANTHROPIC_VERTEX_PROJECT_ID=my-project-id

Microsoft Azure Foundry

# .envrc
export CLAUDE_CODE_USE_FOUNDRY=1
export ANTHROPIC_FOUNDRY_BASE_URL="https://my-resource.services.ai.azure.com/api"
export ANTHROPIC_FOUNDRY_API_KEY="..."

Feature Flags / Tuning

# .envrc — append to any of the above
export ANTHROPIC_MODEL="claude-opus-4-8"   # pin a specific model snapshot; update as releases land
export CLAUDE_CODE_MAX_OUTPUT_TOKENS=64000
export CLAUDE_CODE_EFFORT_LEVEL=high
export CLAUDE_CODE_AUTOCOMPACT_PCT_OVERRIDE=99

Security Checklist

After creating the .envrc:

1. Allow it: direnv allow (required after every edit)
2. Gitignore it: Verify .envrc is in .gitignore — it may contain secrets

grep -q '.envrc' .gitignore 2>/dev/null || echo '.envrc' >> .gitignore

3. Never commit API keys — if the project is shared, use direnv allow locally and keep .envrc out of version control.

Verifying

After setup, verify the environment loads:

cd /path/to/project
direnv allow
env | grep -E 'ANTHROPIC|CLAUDE_CODE|AWS_PROFILE|CLOUD_ML'

Then start claude and confirm the right provider/model is active.

Common Patterns

Layered configs with .envrc.local

# .envrc (committed, non-secret defaults)
export CLAUDE_CODE_EFFORT_LEVEL=high
export CLAUDE_CODE_MAX_OUTPUT_TOKENS=64000

# Source local overrides if present
source_env_if_exists .envrc.local

# .envrc.local (gitignored, secrets)
export ANTHROPIC_API_KEY="sk-ant-..."

Switching between projects

No action needed — direnv automatically loads/unloads when you cd. Just set up each project's .envrc once.

Reference: Claude Code env vars

The full, current list of Claude Code environment variables — providers, model, effort, token limits, config dir — lives in the canonical settings reference. Don't reproduce it here; it drifts:

https://code.claude.com/docs/en/settings.md

The examples above cover the common provider / model / flag cases.

Not for

• Global, single-account setups — set the env vars normally; direnv earns its keep only when config differs per project.
• Managing non-Claude-Code environment variables — that's plain direnv usage, not this skill.