mise Configuration as Single Source of Truth

Use mise [env] as centralized configuration with backward-compatible defaults.

When to Use This Skill

Use this skill when:

• Centralizing environment variables in mise.toml
• Setting up Python venv auto-creation with mise
• Implementing hub-spoke configuration for monorepos
• Creating backward-compatible environment patterns

Core Principle

Define all configurable values in .mise.toml [env] section. Scripts read via environment variables with fallback defaults. Same code path works WITH or WITHOUT mise installed.

Key insight: mise auto-loads [env] values when shell has mise activate configured. Scripts using os.environ.get("VAR", "default") pattern work identically whether mise is present or not.

Quick Reference

Language Patterns

Special Directives

Python Venv Auto-Creation (Critical)

Auto-create and activate Python virtual environments:

[env]
_.python.venv = { path = ".venv", create = true }

This pattern is used in ALL projects. When entering the directory with mise activated:

1. Creates .venv if it doesn't exist
2. Activates the venv automatically
3. Works with uv for fast venv creation

Alternative via [settings]:

[settings]
python.uv_venv_auto = true

Hub-Spoke Architecture (CRITICAL)

Keep root mise.toml lean by delegating domain-specific tasks to subfolder mise.toml files.

Wiki Reference: Pattern-mise-Configuration - Complete documentation with CLAUDE.md footer prompt

When to Use

• Root mise.toml exceeds ~50 lines
• Project has multiple domains (packages, experiments, infrastructure)
• Different subfolders need different task sets

Spoke Scenarios

Hub-spoke applies to any multi-domain project, not just packages:

Directory Structure Examples

Monorepo:

project/
├── mise.toml              # Hub: [tools] + [env] + orchestration
├── packages/
│   ├── api/mise.toml      # Spoke: API tasks
│   └── web/mise.toml      # Spoke: Web tasks
└── scripts/mise.toml      # Spoke: Utility scripts

ML/Research Project:

ml-project/
├── mise.toml              # Hub: python, cuda, orchestration
├── experiments/
│   ├── baseline/mise.toml # Spoke: baseline experiment
│   └── ablation/mise.toml # Spoke: ablation study
├── training/mise.toml     # Spoke: training pipelines
└── evaluation/mise.toml   # Spoke: metrics, benchmarks

Infrastructure:

infra/
├── mise.toml              # Hub: terraform, kubectl, helm
├── terraform/
│   ├── prod/mise.toml     # Spoke: production infra
│   └── staging/mise.toml  # Spoke: staging infra
└── kubernetes/mise.toml   # Spoke: k8s manifests

Hub Responsibilities (Root mise.toml)

# mise.toml - Hub: Keep this LEAN

[tools]
python = "<version>"
uv = "latest"

[env]
PROJECT_NAME = "my-project"
_.python.venv = { path = ".venv", create = true }

# Orchestration: delegate to spokes
[tasks.train-all]
run = """
cd experiments/baseline && mise run train
cd experiments/ablation && mise run train
"""

[tasks."build:api"]
run = "cd packages/api && mise run build"

Spoke Responsibilities (Subfolder mise.toml)

# experiments/baseline/mise.toml - Spoke

[env]
EXPERIMENT_NAME = "baseline"
EPOCHS = "<num>"           # e.g., 100
LEARNING_RATE = "<float>"  # e.g., 0.001

[tasks.train]
run = "uv run python train.py"
sources = ["*.py", "config.yaml"]
outputs = ["checkpoints/*.pt"]

[tasks.evaluate]
depends = ["train"]
run = "uv run python evaluate.py"

Inheritance Rules

• Spoke mise.toml inherits hub's [tools] automatically
• Spoke [env] extends hub's [env] (can override per domain)
• .mise.local.toml applies at directory level (secrets stay local)

Anti-Patterns

Monorepo Workspace Pattern

For Python monorepos using uv workspaces, the venv is created at the workspace root. Sub-packages share the root venv.

# Root mise.toml
[env]
_.python.venv = { path = ".venv", create = true }

Hoisted Dev Dependencies (PEP 735)

Dev dependencies (pytest, ruff, jupyterlab, etc.) should be hoisted to workspace root pyproject.toml using [dependency-groups]:

# SSoT-OK: example workspace configuration
# Root pyproject.toml
[tool.uv.workspace]
members = ["packages/*"]

[dependency-groups]
dev = [
    "pytest>=<version>",
    "ruff>=<version>",
    "jupyterlab>=<version>",
]

Why hoist? Sub-package [dependency-groups] are NOT automatically installed by uv sync from root. Hoisting ensures:

• Single command: uv sync --group dev
• No "unnecessary package" warnings
• Unified dev environment across all packages

Reference: bootstrap-monorepo.md for complete workspace setup

Special Directives

Load from .env Files (_.file)

[env]
# Single file
_.file = ".env"

# Multiple files with options
_.file = [
    ".env",
    { path = ".env.secrets", redact = true }
]

Extend PATH (_.path)

[env]
_.path = [
    "{{config_root}}/bin",
    "{{config_root}}/node_modules/.bin",
    "scripts"
]

Source Bash Scripts (_.source)

[env]
_.source = "./scripts/env.sh"
_.source = { path = ".secrets.sh", redact = true }

Lazy Evaluation (tools = true)

By default, env vars resolve BEFORE tools install. Use tools = true to access tool-generated paths:

[env]
# Access PATH after tools are set up
GEM_BIN = { value = "{{env.GEM_HOME}}/bin", tools = true }

# Load .env files after tool setup
_.file = { path = ".env", tools = true }

Template Syntax (Tera)

mise uses Tera templating. Delimiters: {{ }} expressions, {% %} statements, {# #} comments.

Built-in Variables

Functions

[env]
# Get env var with fallback
NODE_VER = "{{ get_env(name='NODE_VERSION', default='20') }}"

# Execute shell command
TIMESTAMP = "{{ exec(command='date +%Y-%m-%d') }}"

# System info
ARCH = "{{ arch() }}"      # x64, arm64
OS = "{{ os() }}"          # linux, macos, windows
CPUS = "{{ num_cpus() }}"

# File operations
VERSION = "{{ read_file(path='VERSION') | trim }}"
HASH = "{{ hash_file(path='config.json', len=8) }}"

Filters

[env]
# Case conversion
SNAKE = "{{ name | snakecase }}"
KEBAB = "{{ name | kebabcase }}"
CAMEL = "{{ name | lowercamelcase }}"

# String manipulation
TRIMMED = "{{ text | trim }}"
UPPER = "{{ text | upper }}"
REPLACED = "{{ text | replace(from='old', to='new') }}"

# Path operations
ABSOLUTE = "{{ path | absolute }}"
BASENAME = "{{ path | basename }}"
DIRNAME = "{{ path | dirname }}"

Conditionals

[env]
{% if env.DEBUG %}
LOG_LEVEL = "debug"
{% else %}
LOG_LEVEL = "info"
{% endif %}

Required & Redacted Variables

Required Variables

Enforce variable definition with helpful messages:

[env]
DATABASE_URL = { required = true }
API_KEY = { required = "Get from https://example.com/api-keys" }

Redacted Variables

Hide sensitive values from output:

[env]
SECRET = { value = "my_secret", redact = true }
_.file = { path = ".env.secrets", redact = true }

# Pattern-based redactions
redactions = ["*_TOKEN", "*_KEY", "PASSWORD"]

[settings] Section

[settings]
experimental = true              # Enable experimental features
python.uv_venv_auto = true       # Auto-create venv with uv

[tools] Version Pinning

Pin tool versions for reproducibility:

[tools]
python = "3.11"  # minimum baseline; use 3.12, 3.13 as needed
node = "latest"
uv = "latest"

# With options
rust = { version = "1.75", profile = "minimal" }

min_version: Enforce mise version compatibility:

min_version = "2024.9.5"

Implementation Steps

1. Identify hardcoded values - timeouts, paths, thresholds, feature flags
2. Create .mise.toml - add [env] section with documented variables
3. Add venv auto-creation - _.python.venv = { path = ".venv", create = true }
4. Update scripts - use env vars with original values as defaults
5. Add ADR reference - comment: # ADR: 2025-12-08-mise-env-centralized-config
6. Test without mise - verify script works using defaults
7. Test with mise - verify activated shell uses .mise.toml values

GitHub Token Multi-Account Patterns (MANDATORY for Multi-Account Setups) {#github-token-multi-account-patterns}

For multi-account GitHub setups, mise [env] provides per-directory token configuration that overrides gh CLI's global authentication.

Token Storage

Store tokens in a centralized, secure location:

mkdir -p ~/.claude/.secrets
chmod 700 ~/.claude/.secrets

# Create token files (one per account)
gh auth login  # authenticate as account
gh auth token > ~/.claude/.secrets/gh-token-accountname
chmod 600 ~/.claude/.secrets/gh-token-*

Per-Directory Configuration

# ~/.claude/.mise.toml (terrylica account)
[env]
GH_TOKEN = "{{ read_file(path=config_root ~ '/.secrets/gh-token-terrylica') | trim }}"
GITHUB_TOKEN = "{{ read_file(path=config_root ~ '/.secrets/gh-token-terrylica') | trim }}"
GH_ACCOUNT = "terrylica"  # For human reference only

# ~/eon/.mise.toml (terrylica account - different directory)
[env]
GH_TOKEN = "{{ read_file(path=env.HOME ~ '/.claude/.secrets/gh-token-terrylica') | trim }}"
GITHUB_TOKEN = "{{ read_file(path=env.HOME ~ '/.claude/.secrets/gh-token-terrylica') | trim }}"
GH_ACCOUNT = "terrylica"

Variable Naming Convention

Rule: Always set BOTH variables in mise [env] pointing to the same token file. Different tools check different variable names.

Alternative: 1Password Integration

For enhanced security with automatic token rotation:

[env]
GH_TOKEN = "{{ op_read('op://Engineering/GitHub Token/credential') }}"

With caching for performance:

[env]
GH_TOKEN = "{{ cache(key='gh_token', duration='1h', run='op read op://Engineering/GitHub Token/credential') }}"

Verification

/usr/bin/env bash << 'MISE_EOF'
for dir in ~/.claude ~/eon ~/own ~/scripts ~/459ecs; do
  cd "$dir" && eval "$(mise hook-env -s bash)" && echo "$dir → $GH_ACCOUNT"
done
MISE_EOF

Pattern: Based on GitHub Multi-Account Authentication ADR (mise [env] per-directory token loading).

SSH ControlMaster Warning: If using multi-account SSH, ensure ControlMaster no is set for GitHub hosts in ~/.ssh/config. Cached connections can authenticate with the wrong account. See SSH ControlMaster Cache for troubleshooting.

Anti-Patterns

Critical: mise [env] Secrets and Non-Interactive Shells

Do NOT put secrets in mise.toml [env] if they will be consumed by pueue jobs, cron jobs, or systemd services. These execution contexts run in clean shells without mise activation — [env] variables are invisible.

Preferred pattern: Use mise.toml for task definitions and non-secret configuration only. Put secrets in a .env file (gitignored) and load them with python-dotenv at runtime:

# mise.toml — tasks only, no secrets in [env]
[env]
DATABASE_NAME = "mydb"       # OK: non-secret defaults

[tasks.backfill]
run = "bash scripts/backfill.sh"

# .env (gitignored) — secrets loaded by python-dotenv
API_KEY=sk-abc123
DATABASE_PASSWORD=hunter2

This works identically in interactive shells, pueue jobs, cron, systemd, and across macOS/Linux.

Cross-reference: See devops-tools:distributed-job-safety — G-15, AP-16

Task Orchestration Integration

When you detect workflow opportunities during mise configuration, prescriptively invoke the mise-tasks skill.

Detection Triggers

• Multi-step workflows (test + build + deploy)
• Repeatable project commands (lint, format, release)
• Dependency chains (db-migrate depends on db-init)
• File-tracked builds (sources/outputs caching)

Prescriptive AI Reminder

PRESCRIPTIVE: When detecting multi-step project workflows, invoke mise-tasks skill to generate appropriate [tasks] definitions with dependency management.

Example: From Environment to Tasks

Step 1: Configure environment (this skill):

[env]
DATABASE_URL = "postgresql://localhost/mydb"
_.python.venv = { path = ".venv", create = true }

Step 2: Define tasks (mise-tasks skill):

[tasks.test]
depends = ["lint"]
run = "pytest tests/"

[tasks.deploy]
depends = ["test", "build"]
run = "deploy.sh"

Tasks automatically inherit [env] values.

Additional Resources

For complete code patterns and examples, see: references/patterns.md

For task orchestration, see: mise-tasks skill - Dependencies, arguments, file tracking, watch mode

Wiki Documentation: Pattern-mise-Configuration - Copyable CLAUDE.md footer prompt, hub-spoke architecture, quick reference

ADR Reference: When implementing mise configuration, create an ADR at docs/adr/YYYY-MM-DD-mise-env-centralized-config.md in your project.

Troubleshooting