EnvOrigin

Explain where environment variables actually come from — in Docker Compose projects and GitHub Actions workflows.

“Why is DATABASE_URL suddenly localhost in staging? It was fine yesterday.” “Which SHARED did that step actually see — the workflow env, the job env, the step env, or the env file?”

A variable in a Compose service can be defined in five places at once — your shell, .env, --env-file, services.*.env_file, and services.*.environment — and Compose resolves each one differently depending on how the others are written. A workflow variable can be defined in five more: workflow, job, and step env: blocks, env: file: references, and GITHUB_ENV writes at runtime. When the value changes, figuring out which layer won (and which layers you just wasted an hour editing) is manual archaeology.

EnvOrigin reads your project exactly the way the platform does and answers the question per variable, with the full provenance chain and line numbers:

$ envorigin explain P -s web --show-values

P for service web
state: present
value: "compose_explicit"
winner: service environment (compose.yaml:9)
candidates:
  - [shadowed] service env_file (env/web.env:1) = "from_web_env"
  - [shadowed] service env_file (env/overrides.env:1) = "from_overrides_env"
  - [winner] service environment (compose.yaml:9) = "compose_explicit"

Values are redacted by default — you get a short SHA-256 fingerprint instead of the secret. Pass --show-values only when you mean it.

Why

Docker's own precedence rules are subtle and their failure modes are loud:

Layer (low → high)	Writes a value	Notes
Compose file defaults	rarely	only via interpolation
Shell / process environment	when present	wins over .env
.env / --env-file	interpolation only	never injected into the container directly
services.*.env_file	injected	later files override earlier ones
services.*.environment	injected	explicit values override everything
docker compose config	ground truth	what Compose actually computed

Two consequences that bite teams every week:

1. .env is not an injection layer. Values there only reach the container through ${VAR} interpolation or an unset environment: - VAR entry. Editing .env and seeing no effect is the most common false fix.
2. The winner is invisible. If environment: overrides a value from env_file, Compose gives you no hint that the env_file line is dead code. EnvOrigin marks it shadowed so you can delete it with confidence.

Install

cargo install envorigin            # crates.io
brew install FIERsity/envorigin/envorigin  # Homebrew
# or from source
cargo build --release

Every GitHub Release also carries prebuilt binaries — no Rust toolchain or Homebrew needed:

Platform	Asset
Linux x86_64	envorigin-<version>-x86_64-unknown-linux-gnu
Linux arm64	envorigin-<version>-aarch64-unknown-linux-gnu
macOS arm64	envorigin-<version>-aarch64-apple-darwin
macOS x86_64	envorigin-<version>-x86_64-apple-darwin
Windows x86_64	envorigin-<version>-x86_64-pc-windows-msvc.exe

# e.g. CI cache or a Docker layer
curl -sL https://github.com/FIERsity/envorigin/releases/download/v1.12.0/envorigin-v1.12.0-x86_64-unknown-linux-gnu -o envorigin
chmod +x envorigin

Requires Rust 1.86+.

Quick start

brew install FIERsity/tap/envorigin      # or: cargo install envorigin
cd your-project
envorigin audit                          # ← that's it — the format is auto-detected

EnvOrigin figures out whether your project is Docker Compose, GitHub Actions, GitLab CI, CircleCI, or a plain dotenv project — no flags needed. Run it on your project right now:

$ envorigin audit

EnvOrigin 1.12.0 — audit
compose: /app/compose.yaml

2 error(s), 4 warning(s), 4 info

error [sensitive-value]: API_TOKEN is set to a concrete value; verify it is not a real credential (/app/compose.yaml:7)
warning [shadowed-env-line]: SHADOWED is shadowed by a higher-precedence layer and can be removed (/app/env/web.env:2)

Then answer "why is it that value" for one variable:

$ envorigin explain DATABASE_URL

DATABASE_URL for service web
state: present
value: <redacted sha256:61be8cd3>
winner: service environment (compose.yaml:9)

Common scenarios:

I want to…	Run
see every variable and where it comes from	envorigin scan
understand one variable	envorigin explain DATABASE_URL
get a health report with fixes	envorigin audit
gate my CI on the report	envorigin audit --fail-on error (or the GitHub Action)
set team conventions	envorigin init → edit envorigin.toml
compare environments	envorigin diff --project-a ./dev --project-b ./prod

Values are hidden by default — pass --show-values when you mean it. The backend-prefixed commands (envorigin gitlab audit, …) remain for when you want to pin the format explicitly.

Usage

EnvOrigin 1.12.0
Explain where environment variables come from

Usage: envorigin <COMMAND>

Commands:
  scan     List the variables configured for each Compose service
  explain  Explain the winning and shadowed sources for one variable
  audit    Audit a project for env health issues (sensitive values, dead code)
  diff     Compare environment drift across dotenv files
  graph    Render a mermaid provenance graph of the environment
  gitlab   Analyze a GitLab CI configuration's variables
  circleci Analyze a CircleCI configuration's environment variables
  actions  Analyze a GitHub Actions workflow's environment variables
  lsp      Start the LSP server (for editor integration)
  completions  Generate a shell completion script
  init      Generate an envorigin.toml rules template
  dotenv    Audit standalone dotenv files without a Compose context
  help     Print this message or the help of the given subcommand(s)

Options:
  -h, --help     Print help
  -V, --version  Print version

Common flags for scan/explain/audit/graph:

Flag	Meaning
-f, --file <PATH>	Compose file to analyze; omit to auto-detect (compose → actions → gitlab → circleci → dotenv)
PATH (positional)	a config file, or a directory to detect in (envorigin scan ./dir)
--env-file <PATH>	Compose interpolation file; repeatable, later files win
--project-directory <PATH>	override the Compose project directory
--host-env-file <PATH>	overlay the shell from a dotenv file (reproducible diagnosis)
--no-docker-check	skip comparing with docker compose config
--show-values	reveal plaintext values (default: redacted)
--format json	machine-readable output

scan

$ envorigin scan

EnvOrigin 1.12.0
compose: /app/compose.yaml
docker verification: verified
interpolation files: /app/.env

service web (2 present, 4 tracked)
  D                        absent  —                        ← service environment (compose.yaml:8)
  FROM_ENV                 set     <redacted sha256:b3aa92d6> ← service environment (compose.yaml:6)
  MODE                     set     <redacted sha256:ab8e18ef> ← service environment (compose.yaml:5)

docker verification reports whether your understanding matched docker compose config --format json:

• verified — EnvOrigin's model agrees with Docker (run it on a real project and it should normally say this; if it doesn't, that's a bug — please report)
• unavailable — Docker CLI not installed; analysis is unverified
• failed — Docker ran but disagreed, with a docker-divergence diagnostic on each affected variable
• skipped — --no-docker-check was passed

explain

$ envorigin explain T -s web --show-values

T for service web
state: present
value: "from_project_env"
winner: service env_file (env/web.env:4)
derived from:
  - default .env (.env:2)

T=${S} in env/web.env interpolated S from .env — the derived from block is EnvOrigin's answer to "but why is it that value?" for interpolated variables. explain --debug prints the full resolution trace: every definition of the variable in the interpolation context with its precedence and order.

actions

GitHub Actions resolves a variable from up to five declaration layers per scope, and one runtime layer you can't see statically:

Layer (low → high)	Writes a value	Notes
Workflow-level env:	every job and step
Job-level env:	every step in the job	overrides workflow env
Job-level env: file:	every step in the job	overrides job env
Step-level env:	that step	overrides job layers
Step-level env: file:	that step	overrides step env
GITHUB_ENV writes	subsequent steps	runtime only — flagged, not resolved

$ envorigin actions explain SHARED -j build -s Configure --show-values

SHARED in workflow .github/workflows/ci.yml
state: set
value: "step_value"
winner: step env (.github/workflows/ci.yml:19)
candidates:
  - [shadowed] workflow env (.github/workflows/ci.yml:4) = "workflow_value"
  - [shadowed] job env (.github/workflows/ci.yml:11) = "job_value"
  - [winner] step env (.github/workflows/ci.yml:19) = "step_value"

--debug adds a layer-by-layer resolution trace (which definition wins at each precedence level):

$ envorigin actions explain SHARED -j build -s Configure --debug

resolution trace for SHARED (workflow env < job env < step env < env file < inputs; later wins):
  #1   shadowed  workflow env (.github/workflows/ci.yml:4)   "workflow_value"
  #2   shadowed  job env (.github/workflows/ci.yml:11)       "job_value"
  #3   winner    step env (.github/workflows/ci.yml:19)      "step_value"
  3 definition(s)

actions also tracks ${{ ... }} expressions — the reference list answers "what did this value read from?" for ${{ env.X }} (resolved to its source), ${{ secrets.X }} / ${{ vars.X }} / ${{ github.* }} (marked external context, since their values live in repository/org settings), and ${{ matrix.X }} (marked external — a strategy variable). GITHUB_ENV writes like echo "RESULT=ok" >> "$GITHUB_ENV" produce an Info diagnostic: the variable exists at runtime but cannot be resolved statically.

Flags: -f/--file (default .github/workflows/ci.yml), -j/--job, -s/--step (name, id, or index), --project-directory, --show-values, --format json.

audit / actions audit / gitlab audit / circleci audit

A checkable health report for a whole project. envorigin audit auto-detects the project format; the prefixed variants pin one. It aggregates every diagnostic the analyzer produces, then adds checks a diagnostic model cannot express:

Check	Severity	What it reports
sensitive-value	error	a variable whose name matches TOKEN/SECRET/PASSWORD/KEY/… is set to a concrete value
undefined-interpolation-variable	warning	$VAR / ${VAR} references a variable that resolves to empty
shadowed-env-line	warning	a definition loses to a higher-precedence layer with a different value — dead code
secret-manager-reference	info	a sensitive variable references an external secret (Vault, AWS Secrets Manager/SSM, templating) — its value cannot be resolved statically
empty-value	info	a variable resolves to an empty string — usually an accidental stub
credential-in-url	error	any variable embeds credentials in a URL (scheme://user:pass@host) — the most common leak, usually under names like DATABASE_URL
private-key-in-value	error	any variable embeds a PEM private key (-----BEGIN ... PRIVATE KEY-----)
known-secret-format	error	any variable contains a known credential shape (AWS AKIA…, GitHub ghp_…, Stripe sk_live_…, Slack, OpenAI) — detected by value, not name
unused-interpolation-variable	info	an interpolation-file variable no service consumes (sensitive ones also get the placeholder/value checks — a plaintext secret is flagged even when unused)
env-file-missing	info	a short-syntax env_file: points at a missing file — a notice, mirroring docker compose config (long-syntax required: true still errors)
dotenv-trailing-content, docker checks, …	as diagnosed	existing analyzer diagnostics

Secret managers (Vault, AWS Secrets Manager/SSM)

EnvOrigin never resolves external secrets — it detects that a value is an external reference and reports it as secret-manager-reference (info), so pipelines don't treat a placeholder as a real credential:

Value shape	Detected as
vault:secret/data/db#password	HashiCorp Vault
arn:aws:secretsmanager:region:…:secret:name	AWS Secrets Manager
arn:aws:ssm:region:…:parameter/name / sm://…	AWS SSM
{{ secret … }} / {{ secrets.… }}	secret templating

The recommended pattern is to keep the reference in the config and store the real value out-of-band:

# docker-compose.yml — the reference is intentional, no secret here
services:
  web:
    environment:
      DB_PASSWORD: vault:secret/data/db#password
      AWS_CREDENTIALS: arn:aws:secretsmanager:us-east-1:123456789012:secret:prod/credentials

envorigin audit reports both as info [secret-manager-reference] (no pipeline failure by default). Rotate any concrete value the audit flags as sensitive-value into one of these references, then re-run envorigin audit --fail-on error to confirm.

$ envorigin audit

EnvOrigin 1.12.0 — audit
compose: /app/compose.yaml

2 error(s), 4 warning(s), 4 info

error [sensitive-value]: API_TOKEN is set to a concrete value; verify it is not a real credential (/app/compose.yaml:7)
warning [shadowed-env-line]: SHADOWED is shadowed by a higher-precedence layer and can be removed (/app/env/web.env:2)
info [unused-interpolation-variable]: ORPHAN in the interpolation file is not consumed by any service (/app/.env:2)

All audit commands support --format json (issues array) and --format github (GitHub Actions workflow commands — problems appear annotated on the offending file lines in pull requests). --ignore <code> (repeatable) exempts issue codes — use it to onboard legacy projects and remove ignores as problems get fixed.

--fail-on <level> (default error) turns the audit into a CI gate: --fail-on warning exits 1 when any warning or error is found. Values are never printed — the audit can run on repositories holding real credentials.

Rules (envorigin.toml)

envorigin init writes a commented rules template (all six rule types) into the current directory; it never overwrites an existing file.

Turn team conventions into enforced audit checks. Any audit command (audit, actions audit, gitlab audit, circleci audit) loads ./envorigin.toml automatically, or a file via --config:

# envorigin.toml
required = ["DATABASE_URL", "LOG_LEVEL"]   # must resolve somewhere
prefix = "APP_"                            # every user variable must follow
forbidden = ["CI"]                         # must not be defined

[patterns]
DATABASE_URL = "^postgres(ql)?://"         # value format validation
LOG_LEVEL = "^(debug|info|warn|error)$"

[allowed]
DB_ENGINE = ["postgres", "mysql"]      # enum whitelist

[max_length]
API_KEY = 64                            # length cap

Check	Severity	Trigger
required-variable-missing	error	a required name resolves nowhere
naming-prefix	warning	a user-defined variable violates the prefix
forbidden-variable	error	a forbidden name is defined
pattern-mismatch	error	a variable's value fails its [patterns] regex
disallowed-value	error	a variable's value is not in its [allowed] set
value-too-long	error	a variable exceeds its [max_length] cap
unknown-rule-variable	warning	a [patterns]/[allowed]/[max_length] key matches no variable — usually a typo, and the rule would silently never fire

The same checks apply to all four backends; --fail-on gates CI exactly as for the built-in checks. Rule keys for required variables are exempt from unknown-rule-variable: declaring a convention for a variable the project must add is the intended setup.

EnvOrigin audits its own workflow on every push: the Audit own workflow (dogfood) step in .github/workflows/ci.yml runs envorigin actions audit --fail-on warning on the repository's CI file.

dotenv audit

Audit standalone .env files with no Compose context — every entry gets the full sensitive-value / placeholder / secret-manager / URL-credential / private-key / known-format checks, plus --fail-on, --format, and envorigin.toml rules (--config):

$ envorigin dotenv audit .env

2 error(s), 1 warning(s), 0 info
warning [sensitive-placeholder]: DB_PASSWORD is set to a placeholder-looking value ... (.env:1)
error [known-secret-format]: AWS_KEY contains what looks like a AWS access key ... (.env:2)
error [credential-in-url]: URL embeds credentials in a URL ... (.env:3)

diff

Compare dotenv files (below) or two Compose projects' final resolved environments:

$ envorigin diff --project-a ./local --project-b ./prod

drift (1):
  web.DB_HOST: ./local = "localhost" | ./prod = "db.prod.example.com"
only in ./local (1):
  web.ONLY_A = "a_value"

Both modes support --format json (machine-readable; sensitive values stay redacted unless --show-values) and --fail-on-drift, which exits 1 when any variable carries different values — a CI gate that catches environments drifting apart:

# ci/drift.yml — fail the pipeline when local and prod diverge
steps:
  - run: envorigin diff --project-a ./local --project-b ./prod --fail-on-drift --format json

diff (dotenv files)

Compare dotenv files for environment drift — the same variable with different values across local, CI, and staging:

$ envorigin diff .env ci.env

EnvOrigin 1.12.0 — diff
.env vs ci.env

drift (2):
  API_TOKEN: .env = <redacted sha256:61be8cd3> | ci.env = <redacted sha256:9b723f99>
  DB_HOST: .env = "localhost" | ci.env = "db.prod.example.com"

only in .env (1):
  ONLY_LOCAL = "local_value"

Sensitive-looking variables are redacted unless --show-values is passed.

graph

A mermaid provenance graph — pipe it into mermaid-cli or any mermaid renderer:

$ envorigin graph | mmdc -o provenance.svg

scripts/validate-graphs.sh re-validates every backend's graph output against the official mermaid parser (no browser needed) — the syntax regression guard that keeps the generator honest.

Solid edges point at the winning source, dashed edges at shadowed candidates (dashed with a derived label at interpolation dependencies). actions graph renders the workflow's job/step/variable/source layout; gitlab graph and circleci graph render the same provenance for their backends. Output has been validated against the mermaid renderer for all four.

gitlab

Layered resolution for .gitlab-ci.yml variables:: included files (include: local) < file-global < job-level. $VAR / ${VAR} interpolation is tracked per reference, and — unlike Compose — definitions may reference each other in any order (GitLab semantics), so values are resolved in a second pass over all definitions. include: remote / include: template are reported as external and not fetched. Statically-known predefined variables (CI_*, GITLAB_*, RUNNER_*) are labelled as GitLab predefined. Multi-document files (--- separated, e.g. a spec: document followed by the pipeline) are merged in order; CI/CD component inputs v2 ($[[ inputs.X ]]) surface as external references.

$ envorigin gitlab explain BUILD_ID -j build --show-values

BUILD_ID in .gitlab-ci.yml
state: set
value: "-demo"
winner: job variables (.gitlab-ci.yml:20)
references:
  - $CI_PIPELINE_ID → GitLab predefined (GitLab predefined)
  - $APP_NAME → global variables (.gitlab-ci.yml:7)

circleci

Layered resolution for .circleci/config.yml: executor environment: < job environment: / env: list. << parameters.X >> references resolve to the job's parameters: declaration (default shown); << pipeline.X >> and context: are organization/API state and are reported as external. CIRCLE_* predefined variables are labelled; values interpolate shell-style.

$ envorigin circleci explain TARGET -j build --show-values

TARGET in .circleci/config.yml
state: set
value: "<< parameters.target >>"
winner: job environment (.circleci/config.yml:21)
references:
  - parameters.target → job parameter (.circleci/config.yml:15)

lsp — editor integration

envorigin lsp speaks the Language Server Protocol over stdio; point any LSP-capable editor (Neovim, Emacs, JetBrains, …) at it:

• Hover a variable definition line to see its winning source and value (redacted by default).
• Go to definition jumps from a variable to the file and line that won.
• Live diagnostics mirror the audit findings — undefined interpolation references, shadowed dead-code lines, sensitive values — as you edit.

Documents are routed to the matching backend by file name (compose.y*ml, .github/workflows/*.yml, .gitlab-ci.yml, .circleci/config.yml, .env — dotenv files get the live security diagnostics). Unsaved buffer edits are analyzed in memory — hover and diagnostics track the buffer, while relative references still resolve against the real file location.

Design

• src/compose.rs — Compose model (env_file specs, environment forms) and per-variable candidate resolution in Compose's own order.
• src/interpolation.rs — a Compose-flavored interpolator: $VAR, ${VAR}, ${VAR:-word}, ${VAR:?msg}, ${VAR:+alt}, $$ escaping, with per-reference source tracking.
• src/dotenv.rs — dotenv parsing: quoting, export prefixes, comments, unset entries, trailing-content warnings.
• src/docker.rs — spawns docker compose config --format json and reconciles Docker's canonical model against the local model.
• src/actions.rs — workflow model: layered env resolution (workflow → job → job file → step → step file), ${{ }} reference tracking, built-in default variables, and GITHUB_ENV runtime detection.
• src/output.rs — human and JSON renderers, default redaction with SHA-256 fingerprints.

Semantics notes:

• COMPOSE_ENV_FILES is expanded (path-separator split, relative to the project directory, replacing the default .env lookup); missing entries produce a compose-env-files-missing warning and are skipped. Explicit --env-file arguments take precedence over it, mirroring Docker Compose.
• COMPOSE_FILE is not followed; it produces a compose-file-redirect warning telling you to pass --file explicitly.
• actions is static analysis: GITHUB_ENV writes are flagged as runtime-only, and GitHub does not document the precedence of a file: key mixed with regular keys in one env: block — EnvOrigin applies the file layer after the map (same order as Compose env_file).

Performance

A synthetic project with 800 variables across 4 services plus 4 env files (821-line compose) scans in ~0.3 s and audits in ~2.4 s (debug build; release builds are several times faster) — linear in input size, no caching needed.

Testing

cargo test

145 tests: unit tests for the dotenv parser, the interpolator, Compose normalization, the Docker canonical extraction, the Actions layer resolution, the audit checks, and project-type auto-detection; plus end-to-end CLI tests against tests/fixtures/{basic,precedence,raw,env-files,actions,actions-inputs,audit} that assert redaction, derivation tracking, the full shadowing chain, COMPOSE_ENV_FILES expansion, format: raw semantics, expression-reference resolution, and audit exit codes. CI runs fmt + clippy -D warnings + tests on Ubuntu, and audits its own workflow (dogfood).

Roadmap

Phase 1 — doctor mode (in progress)

• audit with sensitive-value, shadowed-dead-code, unused-variable checks
• --fail-on CI gate + dogfooding EnvOrigin's own workflow
• envorigin diff — compare dotenv files for environment drift (sensitive values redacted by default)
• real-repository validation: audited all 39 examples in docker/awesome-compose (every file parses and resolves; the findings fixed the placeholder and unused-variable checks). The GitLab and CircleCI backends were validated against real production configs: GitLab's own Auto-DevOps template, and leiningen's JSON-format CircleCI config.

Phase 2 — visualization & IDE (in progress)

• envorigin graph / envorigin actions graph — mermaid provenance graph (solid = winner, dashed = shadowed, dashed-derived = dependency); output validated against the mermaid renderer
• LSP server + VS Code extension (hover, go-to-definition, live diagnostics)
• real CI config validation: rust-lang/cargo's 372-line workflow and astral-sh/uv's benchmark workflow (with workflow_dispatch inputs) parse, scan, and audit cleanly; unnamed steps display 1-based numbers
• GitLab CI variables support (include < global < job; $VAR reference tracking; predefined variables)
• CircleCI env support (executor < job, parameters, contexts external)

Phase 3 — ecosystem

• crates.io release (cargo install envorigin) + Homebrew tap
• rules engine: envorigin.toml for team conventions (required vars, name prefixes, value validation)
• prebuilt release binaries for linux/macOS/Windows (no toolchain needed)
• GitHub Action wrapper (FIERsity/envorigin-action, dogfooded in this repo's own CI)
• integration notes for secret managers (Vault, AWS SSM)

Shell completions

envorigin completions bash  > ~/.bash_completion.d/envorigin
envorigin completions zsh   > ~/.zfunc/_envorigin
envorigin completions fish  > ~/.config/fish/completions/envorigin.fish

CI integration

audit is built to gate pipelines: --format json for parsing, --fail-on <level> for the exit code. Wire it into any of the three CI platforms EnvOrigin understands:

GitHub Actions

The one-step wrapper (no Rust toolchain; downloads the prebuilt binary, auto-detects the config, annotates findings on the PR):

- uses: FIERsity/envorigin-action@v1
  with:
    fail-on: warning

Or the CLI directly:

- name: Install envorigin
  run: cargo install envorigin
- name: Audit environment config
  run: envorigin audit --fail-on warning

GitLab CI

audit:
  script:
    - cargo install envorigin
    - envorigin gitlab audit --fail-on warning
  allow_failure: false

CircleCI

jobs:
  audit:
    docker:
      - image: cimg/rust:1.86
    steps:
      - checkout
      - run: cargo install envorigin
      - run: envorigin circleci audit --fail-on warning

Each audit command targets its own platform's config file by default (compose.yaml, .github/workflows/ci.yml, .gitlab-ci.yml, .circleci/config.yml) and applies envorigin.toml rules automatically.

Real-world example

Regression validation against live open-source repositories is one command — scripts/validate-real-repos.sh clones outline, cargo, uv, and two GitLab-hosted repos (glab, fdroid) shallow, sweeps 51 workflow files for panics and noise, checks explain --debug traces on multi-layer workflows, validates the GitLab backend against real multi-document and double-merge-key configs, and pins the known-good findings (outline's CI embeds test credentials and must keep flagging them).

Running the released binary against outline's docker-compose.yml (production knowledge-base app) with its .env:

$ envorigin audit -f docker-compose.yml --project-directory .

0 error(s), 1 warning(s), 87 info

warning [sensitive-placeholder]: POSTGRES_PASSWORD is set to a placeholder-looking value; replace it before deploying (docker-compose.yml:13)
info [unused-interpolation-variable]: NODE_ENV in the interpolation file is not consumed by any service (.env:1)
info [unused-interpolation-variable]: URL in the interpolation file is not consumed by any service (.env:22)
info [unused-interpolation-variable]: PORT in the interpolation file is not consumed by any service (.env:26)
info [unused-interpolation-variable]: COLLABORATION_URL in the interpolation file is not consumed by any service (.env:30)

The audit reads the same way a maintainer would: the 312-line .env is almost entirely runtime configuration for the application, and only a handful of variables ever reach the Compose services — a real, immediately actionable answer to "why doesn't my .env change anything?".

$ envorigin explain POSTGRES_PASSWORD -s postgres

POSTGRES_PASSWORD for service postgres
state: present
value: <redacted sha256:d74ff0ee>
winner: service environment (docker-compose.yml:13)
candidates:
  - [winner] service environment (docker-compose.yml:13) = <redacted sha256:d74ff0ee>

Release

./scripts/release.sh <version> tags, creates the GitHub release, publishes to crates.io, and updates the Homebrew formula in one run.

License

MIT