🛡️ GitHub Safe-Settings

Policy-as-Code for GitHub Organizations
Centrally manage and enforce repository settings, branch protections, teams, and more across your entire GitHub organization.

Safe-settings – an app to manage policy-as-code and apply repository settings across an organization.

1. In safe-settings, all the settings are stored centrally in an admin repo within the organization. Unlike the GitHub Repository Settings App, the settings files cannot be in individual repositories.

   It is possible specify a custom repo instead of the admin repo with ADMIN_REPO. See Environment variables for more details.

2. The settings in the default branch are applied. If the settings are changed on a non-default branch and a PR is created to merge the changes, the app runs in a dry-run mode to evaluate and validate the changes. Checks pass or fail based on the dry-run results. The dry-run compares the PR's config against the base branch config, so the check run and PR comment report only the changes the PR itself introduces (see Dry-run PR comment).

3. In safe-settings the settings can have 2 types of targets:

   1. org - These settings are applied to the organization. Org-targeted settings are defined in .github/settings.yml. Currently, only rulesets are supported as org-targeted settings.
   2. repo - These settings are applied to repositories.

4. For the repo-targeted settings, there can be 3 levels at which the settings are managed:

   1. Org-level settings are defined in .github/settings.yml

      It is possible to override this behavior and specify a different filename for the settings.yml file with SETTINGS_FILE_PATH. Similarly, the .github directory can be overridden with CONFIG_PATH. See Environment variables for more details.

   2. Suborg level settings. A suborg is an arbitrary collection of repos belonging to projects, business units, or teams. The suborg settings reside in a yaml file for each suborg in the .github/suborgs folder.

      In safe-settings, suborgs could be groups of repos based on repo names, or teams which the repos have collaborators from, or custom property values set for the repos

   3. Repo level settings. They reside in a repo specific yaml in .github/repos folder

5. It is recommended to break the settings into org-level, suborg-level, and repo-level units. This will allow different teams to define and manage policies for their specific projects or business units. With CODEOWNERS, this will allow different people to be responsible for approving changes in different projects.

Note

The suborg and repo level settings directory structure cannot be customized.

🚀 Quick Start

1. Deploy Safe-Settings

Choose your preferred deployment method:

• 🌟 AWS Lambda (: Use the SafeSettings-Template for production-ready deployment with Docker containers, GitHub Actions CI/CD, and comprehensive testing
• 🐳 Docker: Deploy using Docker containers locally or in your infrastructure
• ☁️ Cloud Platforms: Deploy to Heroku, Glitch, or Kubernetes

👉 View all deployment options →

2. Create Admin Repository

Create an admin repository in your organization to store all configuration files:

# Create admin repo in your organization
gh repo create your-org/admin --private

3. Configure Settings Structure

Set up your configuration files in the admin repository:

admin/
├── .github/
│   ├── settings.yml          # Organization-wide settings
│   ├── suborgs/              # Sub-organization settings
│   │   ├── frontend-team.yml
│   │   └── backend-team.yml
│   └── repos/                # Repository-specific settings
│       ├── api-service.yml
│       └── web-app.yml

4. Install GitHub App

Install the Safe-Settings GitHub App in your organization with the required permissions.

👉 Complete setup guide →

📊 Visual Architecture

Configuration Hierarchy

graph TD
    A[Organization Settings<br/>.github/settings.yml] --> B[Sub-Organization Settings<br/>.github/suborgs/*.yml]
    B --> C[Repository Settings<br/>.github/repos/*.yml]
    
    style A fill:#e1f5fe,stroke:#01579b,stroke-width:2px,color:#000
    style B fill:#f3e5f5,stroke:#4a148c,stroke-width:2px,color:#000
    style C fill:#e8f5e8,stroke:#1b5e20,stroke-width:2px,color:#000

Loading

Precedence Order: Repository > Sub-Organization > Organization

Request Flow

sequenceDiagram
    participant GH as GitHub
    participant SS as Safe-Settings
    participant AR as Admin Repo
    participant TR as Target Repos
    
    Note over GH,TR: Webhook Event Processing
    
    GH->>+SS: Webhook Event<br/>(push, repo created, etc.)
    SS->>SS: Validate Event Source
    SS->>+AR: Fetch Configuration Files<br/>(.github/settings.yml, suborgs/, repos/)
    AR-->>-SS: Return Config Files
    
    SS->>SS: Merge Configurations<br/>(Org → Suborg → Repo)
    SS->>SS: Compare with Current<br/>GitHub Settings
    
    alt Configuration Changes Detected
        SS->>+TR: Apply Settings<br/>(Branch Protection, Teams, etc.)
        TR-->>-SS: Confirm Changes
        SS->>GH: Create Check Run<br/>(Success/Failure)
    else No Changes Needed
        SS->>GH: Create Check Run<br/>(No Changes)
    end
    
    SS-->>-GH: HTTP 200 Response
    
    Note over GH,TR: Pull Request Validation (Dry-Run Mode)
    
    GH->>+SS: PR Event<br/>(opened, synchronize)
    SS->>+AR: Fetch PR Changes<br/>(Modified Config Files)
    AR-->>-SS: Return Changed Configs
    
    SS->>SS: Validate Changes<br/>(Dry-Run Mode)
    SS->>SS: Run Custom Validators<br/>(if configured)
    
    alt Validation Passes
        SS->>GH: ✅ Check Success<br/>+ PR Comment (optional)
    else Validation Fails
        SS->>GH: ❌ Check Failure<br/>+ Error Details
    end
    
    SS-->>-GH: HTTP 200 Response
    
    Note over GH,TR: Scheduled Sync (Drift Prevention)
    
    SS->>SS: Cron Trigger<br/>(if configured)
    SS->>+AR: Fetch All Configurations
    AR-->>-SS: Return All Configs
    SS->>+TR: Sync All Repositories<br/>(Prevent Drift)
    TR-->>-SS: Confirm Sync
    SS->>GH: Create Check Run<br/>(Sync Results)

Loading

How it works

Safe-settings is designed to run as a service listening for webhook events or as a scheduled job running on some regular cadence. It can also be triggered through GitHub Actions. (See the How to use section for details on deploying and configuring.)

Events

The App listens to the following webhook events:

• push: If the settings are created or modified, that is, if push happens in the default branch of the admin repo and the file added or changed is .github/settings.yml or .github/repos/*.ymlor .github/suborgs/*.yml, then the settings would be applied either globally to all the repos, or specific repos. For each repo, the settings that are actually applied depend on the default settings for the org, overlaid with settings for the suborg that the repo belongs to, overlaid with the settings for that specific repo.

• repository.created: If a repository is created in the org, the settings for the repo - the default settings for the org, overlaid with settings for the suborg that the repo belongs to, overlaid with the settings for that specific repo - is applied.

• branch_protection_rule: If a branch protection rule is modified or deleted, safe-settings will sync the settings to prevent any unauthorized changes.

• repository.edited: For e.g. If the default branch is renamed, or if topics change, safe-settings will sync the settings, to prevent any unauthorized changes.

• repository.renamed: If a repository is renamed, the default behavior is safe-settings will ignore this (for backward-compatibility). If BLOCK_REPO_RENAME_BY_HUMAN env variable is set to true, safe-settings will revert the repo to the previous name unless it is renamed using a bot. If it is renamed using a bot, it will try to copy the existing <old-repo>.yml to <new-repo>.yml so that the repo config yml stays consistent. If a <new-repo.yml> file already exists, it doesn't create a new one.

• pull_request.opened, pull_request.reopened, check_suite.requested: If the settings are changed, but it is not in the default branch, and there is an existing PR, the code will validate the settings changes by running safe-settings in nop mode and update the PR with the dry-run status. The run loads the base-branch config and filters the results so only the changes the PR introduces are reported (see Dry-run PR comment).

• repository_ruleset: If the ruleset settings are modified in the UI manually, safe-settings will sync the settings to prevent any unauthorized changes.

• member_change_events: If a member is added or removed from a repository, safe-settings will sync the settings to prevent any unauthorized changes.

• member', team.added_to_repository, team.removed_from_repository, team.edited: safe-settings will sync the settings to prevent any unauthorized changes.

• custom_property_values: If new repository properties are set for a repository, safe-settings will run to so that if a sub-org config is defined by that property, it will be applied for the repo

• repository_dispatch (event_type: safe-settings-generate): Triggers the settings generator, which reads the current configuration of a repo/org/suborg and opens a PR against the admin repo with the generated YAML. See Generating settings from existing configuration.

Dry-run PR comment

When a config change is proposed in a PR (a non-default branch), safe-settings runs in nop (no-operation) dry-run mode and posts a comment summarizing what would change if the PR were merged. The results are filtered against the base branch config, so the comment reports only the changes the PR introduces — not the full diff against live GitHub settings.

The comment contains:

• A header with the run timestamp, the number of repos considered, and the number of repos affected.
• Breakdown of changes — a collapsible section, grouped by plugin/repo, showing field-level diffs. Each entry is marked as an addition, modification, or deletion, with the before/after values for modified fields. When there are no changes, it shows No changes to apply.
• Breakdown of errors — a collapsible section listing any errors by repo, or None when there are none. The check run is marked as failed when errors are present.
• Informational messages — a collapsible section listing non-error notices such as plugins skipped via disable_plugins or deletions suppressed by additive_plugins, so reviewers can see which settings were intentionally not applied.

For very large diffs the comment is split across multiple comments, and the check-run summary is truncated with a notice when it exceeds the size limit.

Suborg re-evaluation after repo-level changes

A repo's suborg membership can depend on state that is itself written by safe-settings:

• suborgteams — repos belong to a suborg because a given team is granted access
• suborgproperties — repos belong to a suborg because a custom property has a given value
• suborgrepos — repos belong to a suborg because their name matches a glob

When a repo-level change (a push to .github/repos/<repo>.yml, or a repository.created event for a brand-new repo) adds, removes, or changes a team or custom property, the repo may start or stop matching a suborg config. A new repo may also start matching a suborg because its name matches a suborgrepos glob.

To handle this, after applying a repo-yml change safe-settings re-evaluates the repo's suborg membership. If the matched suborg source set changed, it runs the repo through the apply pipeline a second time so newly matched suborg settings are applied and settings from a no-longer-matching suborg can be removed in the same sync.

For a repository that does not exist yet, team and custom-property membership lookups are deferred until after creation. A 404 from a membership endpoint is treated as an unmatched selector only when the repository itself also returns 404; other lookup failures remain errors. Name-based suborgrepos matching still applies before creation. This allows force_create validation and apply runs to work when earlier changes have already configured team- or property-based suborgs.

Scope: Re-evaluation runs only on the repo-yml change paths (Settings.sync and the per-repo loop of Settings.syncSelectedRepos). Global settings changes (syncAll) and suborg-yml changes (syncSubOrgs) already iterate all relevant repos and do not need it.

Loop prevention. Two guards prevent infinite re-evaluation:

1. Stability check (primary): Before applying changes, safe-settings snapshots the set of suborg source paths that match the repo. After applying, it refreshes the suborg cache and recomputes the set. If the set did not change, re-evaluation stops. If a source appeared or disappeared, the repo is processed once more.
2. Hard depth cap (safety net): Each repo is re-evaluated at most MAX_REEVALUATION_DEPTH = 1 time per sync. This resolves the dominant single-hop case (repo change → suborg membership changed → apply the corrected suborg overlay once) while preventing pathological chains (suborg A applies a team that activates suborg B that activates suborg C…). Chains beyond one hop are resolved on the next sync event, and a warning is logged when the cap is hit.

Trigger optimization. Re-evaluation is skipped entirely when the applied repo change did not affect teams, custom_properties, repository creation, or repository rename state — these are the repo-level changes that can affect suborg matching.

Use safe-settings to rename repos

If you rename a <repo.yml> that corresponds to a repo, safe-settings will rename the repo to the new name. This behavior will take effect whether the env variable BLOCK_REPO_RENAME_BY_HUMAN is set or not.

Restricting safe-settings to specific repos

To restrict which repositories safe-settings can manage, create a deployment-settings.yml file. This file controls the app's scope through the restrictedRepos configuration:

# Using include/exclude
restrictedRepos:
  include:
    - api
    - core-*    # Matches `core-api`, `core-service`, etc.
  exclude:
    - admin
    - .github
    - safe-settings
    - test-*    # Matches `test-repo`, etc.

# Or using simple array syntax for includes
restrictedRepos: 
  - admin
  - .github
  # ...

Note

Pattern matching uses glob expressions, e.g use * for wildcards.

When using include and exclude:

• If include is specified, will only run on repositories that match pattern(s)
• If exclude is specified, will run on all repositories except those matching pattern(s)
• If both are specified, will run only on included repositories that are'nt excluded

By default, if no configuration file is provided, safe-settings will excludes these repos: admin, .github and safe-settings.

See our deployment-settings.yml sample.

Custom rules

Admins setting up safe-settings can include custom rules that would be validated before applying a setting or overriding a broader scoped setting.

The code has to return true if validation is successful, or false if it isn't.

If the validation fails, the error attribute specified would be used to create the error message in the logs or in the PR checks.

The first use case is where a custom rule has to be applied for a setting on its own. For e.g. No collaborator should be given admin permissions.

For this type of validation, admins can provide custom code as configvalidators which validates the setting by itself.

For e.g. for the case above, it would look like:

configvalidators:
  - plugin: collaborators
    error: |
      `Admin role cannot be assigned to collaborators`
    script: |
      console.log(`baseConfig ${JSON.stringify(baseconfig)}`)
      return baseconfig.permission != 'admin'

For convenience this script has access to a variable, baseconfig, that contains the setting that is be applied.

The second use case is where custom rule has to be applied when a setting in the org or suborg level is being overridden. Such as, when default branch protection is being overridden.

For this type of validation, admins can provide custom code as overridevalidators. The script can access two variables, baseconfig and overrideconfig which represent the base setting and the setting that is overriding it.

A sample would look like:

overridevalidators:
  - plugin: branches
    error: |
      `Branch protection required_approving_review_count cannot be overidden to a lower value`
    script: |
      console.log(`baseConfig ${JSON.stringify(baseconfig)}`)
      console.log(`overrideConfig ${JSON.stringify(overrideconfig)}`)
      if (baseconfig.protection.required_pull_request_reviews.required_approving_review_count && overrideconfig.protection.required_pull_request_reviews.required_approving_review_count ) {
        return overrideconfig.protection.required_pull_request_reviews.required_approving_review_count >= baseconfig.protection.required_pull_request_reviews.required_approving_review_count
      }
      return true

A sample of deployment-settings file is found here.

Configuration schemas

The JSON schemas in schema/dereferenced/ support editor validation of org, suborg, and repository configuration. They use GitHub's OpenAPI 3.1 description for API version 2026-03-10, including native JSON Schema nullable types. Branch protection also accepts null, {}, [], or false to delete protection. Regenerate all three schemas after changing their sources in schema/ with npm run build:schema; do not edit the generated files directly.

Org-level rulesets use the organization API shape; repo-level rulesets use the repository API shape. Suborg ruleset_scope overrides the org default in settings.yml, falling back to repo when neither sets a scope. An org scope requires nonempty suborgproperties; otherwise it also falls back to repo. For org-scoped suborgs, safe-settings replaces repository targeting conditions with repository_property targeting derived from suborgproperties, so a ruleset may specify only its ref conditions. When a suborg omits its scope and has property filters, its standalone schema accepts both API shapes because it cannot determine the inherited org default.

Custom Status Checks

For branch protection rules and rulesets, you can allow for status checks to be defined outside of safe-settings together with your usual safe settings.

This can be defined at the org, sub-org, and repo level.

To configure this for branch protection rules, specify {{EXTERNALLY_DEFINED}} under the contexts keyword:

branches:
  - name: main
    protection:
      ...
      required_status_checks:
        contexts:
          - "{{EXTERNALLY_DEFINED}}"

For rulesets, specify {{EXTERNALLY_DEFINED}} under the required_status_checks keyword:

rulesets:
  - name: Status Checks
    ...
    rules:
      - type: required_status_checks
        parameters:
          required_status_checks:
            - context: "{{EXTERNALLY_DEFINED}}"

Notes:

• For the same branch that is covered by multi-level branch protection rules, contexts defined at the org level are merged into the sub-org and repo level contexts, while contexts defined at the sub-org level are merged into the repo level contexts.
• Rules from the sub-org level are merged into the repo level when their ruleset share the same name. Becareful not to define the same rule type in both levels as it will be rejected by GitHub.
• When {{EXTERNALLY_DEFINED}} is defined for a new branch protection rule or ruleset configuration, they will be deployed with no status checks.
• When an existing branch protection rule or ruleset configuration is amended with {{EXTERNALLY_DEFINED}}, the status checks in the existing rules in GitHub will remain as is.

⚠️ Warning: When {{EXTERNALLY_DEFINED}} is removed from an existing branch protection rule or ruleset configuration, the status checks in the existing rules in GitHub will revert to the checks that are defined in safe-settings. From this point onwards, all status checks configured through the GitHub UI will be reverted back to the safe-settings configuration.

Entries in the organization-level centralized_ruleset_bypass_actors replace matching actors in individual rulesets. For OrganizationAdmin and DeployKey, identity is the actor type alone: GitHub ignores actor_id, so omitted, null, and concrete IDs all refer to the same actor. The centralized entry's bypass_mode takes precedence without adding a duplicate actor.

Referencing ruleset bypass actors and reviewers by name

Rulesets normally require numeric ids for bypass_actors[].actor_id and for the team in required_reviewers[].reviewer.id. To avoid looking these ids up, you can reference them by name and safe-settings resolves them to the correct id before applying the ruleset:

• bypass_actors[].name — an alternative to actor_id. The value is resolved based on actor_type:
  • Team → team slug
  • User → username
  • Integration → GitHub App slug
  • RepositoryRole → role name. Built-in roles (read, triage, write, maintain, admin) are mapped automatically; any other name is looked up among the organization's custom repository roles.
• required_reviewers[].reviewer.slug — an alternative to reviewer.id, the slug of the reviewing team.

rulesets:
  - name: Main protection
    target: branch
    enforcement: active
    bypass_actors:
      - name: my-team         # resolved to actor_id
        actor_type: Team
        bypass_mode: always
      - name: admin           # built-in repository role
        actor_type: RepositoryRole
        bypass_mode: always
    rules:
      - type: pull_request
        parameters:
          required_approving_review_count: 1
          dismiss_stale_reviews_on_push: false
          require_code_owner_review: false
          require_last_push_approval: false
          required_review_thread_resolution: false
          required_reviewers:
            - minimum_approvals: 1
              file_patterns: ["*.js"]
              reviewer:
                slug: my-reviewers-team   # resolved to reviewer.id
                type: Team

Notes:

• This is fully backward compatible. Existing policies that use actor_id / reviewer.id continue to work unchanged, and numeric ids are never looked up.
• Provide either the name (name / slug) or the id (actor_id / reviewer.id) for a given entry, not both. Specifying both is an error.
• If a name cannot be resolved to an id, the ruleset sync fails with a clear error so the misconfiguration is surfaced rather than silently ignored.

Status checks inheritance across scopes

Refer to Status checks.

Performance

When there are 1000s of repos to be managed -- and there is a global settings change -- safe-settings will have to work efficiently and only make the necessary API calls.

The app also has to complete the work within an hour: the lifetime of the GitHub app token.

To address these constraints the following design decisions have been implemented:

1. Probot automatically handles rate and abuse limits.
2. Instead of loading all the repo contents from .github/repos/*, it will selectively load the specific repo file based on which repo settings has changed, or a subset of the repo files associated with suborg settings that has changed. The only time all the repo files will be loaded is if there is a global settings file change.
3. The PR check will only provide a summary of errors and changes. (Providing the details of changes for 1000s of repos will error out.)
4. To ensure it handles updates to GitHub intelligently, it will compare the changes with the settings in GitHub, and will call the API only if there are real changes.

Comparing changes with GitHub

To determine if there are real changes, the code will generate a detailed list of additions, modifications, and deletions compared to the settings in GitHub:

For e.g:

If the settings is:

{
  "branches": [
    {
      "name": "master",
      "protection": {
        "required_pull_request_reviews": {
          "required_approving_review_count": 2,
          "dismiss_stale_reviews": false,
          "require_code_owner_reviews": true,
          "dismissal_restrictions": {}
        },
        "required_status_checks": {
          "strict": true,
          "contexts": []
        },
        "enforce_admins": false
      }
    }
  ]
}

and the settings in GitHub is:

{
  "branches": [
    {
      "name": "master",
      "protection": {
        "url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection",
        "required_status_checks": {
          "url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/required_status_checks",
          "strict": true,
          "contexts": [],
          "contexts_url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/required_status_checks/contexts",
          "checks": []
        },
        "restrictions": {
          "url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/restrictions",
          "users_url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/restrictions/users",
          "teams_url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/restrictions/teams",
          "apps_url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/restrictions/apps",
          "users": [],
          "teams": [],
          "apps": []
        },
        "required_pull_request_reviews": {
          "url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/required_pull_request_reviews",
          "dismiss_stale_reviews": true,
          "require_code_owner_reviews": true,
          "required_approving_review_count": 2,
          "dismissal_restrictions": {
            "url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/dismissal_restrictions",
            "users_url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/dismissal_restrictions/users",
            "teams_url": "https://api.github.com/repos/decyjphr-org/test/branches/develop/protection/dismissal_restrictions/teams",
            "users": [],
            "teams": []
          }
        },
        "required_signatures": false,
        "enforce_admins": false,
        "required_linear_history": false,
        "allow_force_pushes": {
          "enabled": false
        },
        "allow_deletions": false,
        "required_conversation_resolution": false
      }
    }
  ]
}

the results of comparison would be:

{
      "additions": {},
      "modifications": {
        "branches": [
          {
            "protection": {
              "required_pull_request_reviews": {
                "dismiss_stale_reviews": false
              }
            },
            "name": "master"
          }
        ]
      },
      "deletions": {},
      "hasChanges": true
    }

Schedule

The App can be configured to apply the settings on a schedule. This could be a way to address configuration drift since webhooks are not always guaranteed to be delivered.

To periodically converge the settings to the configuration, set the CRON environment variable. See Environment variables for more details.

Pull Request Workflow

Safe-settings explicitly looks in the admin repo in the organization for the settings files. The admin repo could be a restricted repository with branch protections and CODEOWNERS

In that set up, when changes happen to the settings files and there is a PR for merging the changes back to the default branch in the admin repo, safe-settings will run checks – which will run in nop mode and produce a report of the changes that would happen, including the API calls and the payload.

For e.g. If we have override validators that will fail if org-level branch protections are overridden at the repo or suborg level with a lesser number of required approvers, here is an screenshot of what users will see in the PR.

Note

If you don't want the PR message to have these details, they can be turned off with CREATE_PR_COMMENT. See Environment variables for more details.

Here is a screenshot of what the users will see in the checkrun page:

Error handling

The app creates a Check at the end of its processing to indicate if there were any errors. The Check is called safe-settings and corresponds to the latest commit on the default branch of the admin repo.

Here is an example of a checkrun result:

And the checkrun page will look like this:

Disabling plugins (disable_plugins)

Any settings file (deployment-settings, org settings.yml, suborg, or repo) can contain a top-level disable_plugins list to turn off one or more safe-settings plugins for a given scope. Each entry is either:

• A plugin name string (shorthand for { plugin: <name>, target: all }), or
• An object { plugin: <name>, target: self | children | all } (default target: all).

Valid plugin names: repository, labels, collaborators, teams, milestones, branches, autolinks, validator, rulesets, environments, custom_properties, custom_repository_roles, variables, archive, app_installations.

Strip matrix (which source layers are removed before merge)

Declared at	target: self	target: children	target: all
deployment-settings	deployment	org + suborg + repo	deployment + org + suborg + repo
org settings.yml	org	suborg + repo	org + suborg + repo
suborgs/*.yml (matched)	suborg	repo	suborg + repo
repos/*.yml	repo	(no-op)	repo

When safe-settings builds the merged configuration for a repo, it strips the disabled plugin's keys from the indicated source layers before merging. For repo-level execution points (the repository and archive plugins) and org-level execution points (rulesets, custom_repository_roles), a disable that targets the corresponding layer also short-circuits the plugin run, and the skip is recorded as an INFO NopCommand in NOP mode (PR check run).

Cascade rules

• Union-only. Strips accumulate across layers; a lower-level config can add more strips but can never undo a strip declared above it.
• No re-enable. If disable_plugins: [labels] is set at the org layer, a repo cannot re-enable labels for itself.

Important limitation

Because strips operate on source layers, a lower-level disable cannot remove configuration contributed by a higher layer. For example, if branches is defined at the org layer and a suborg adds disable_plugins: [{plugin: branches, target: all}], the suborg's strip removes the branches key only from the suborg and repo layers — the org's branches config still merges in, and the branches plugin still runs.

To fully suppress a plugin for matched repos, declare the disable at (or above) the layer that contributes the configuration — typically the org layer with target: all, or at the deployment layer.

Examples

Org settings.yml — disable custom_repository_roles only at the org execution point (rulesets still run):

disable_plugins:
  - plugin: custom_repository_roles
    target: self

Org settings.yml — disable branches everywhere (shorthand):

disable_plugins:
  - branches

Suborg suborgs/team-x.yml — strip labels for matched repos (effective only if labels is not also defined at the org layer):

disable_plugins:
  - plugin: labels
    target: all

Preserving unmanaged custom properties

The legacy custom_properties list enforces declared values and clears undeclared values by setting them to null; GitHub does not delete the property definition. To preserve values owned by other automation while still clearing other undeclared values, use the object form:

custom_properties:
  include:
    - name: environment
      value: production
    - property_name: services
      value: [api, worker]
  exclude:
    - name: '^external-'
    - name: '^OWNER$'

include, exclude, or both may be specified. Exclusions are case-insensitive JavaScript regular expressions, not globs; use .*, not *, to match all names. Regex escapes and character classes retain their normal meaning. Included properties take precedence over exclusions, including an explicit value: null to clear a value. An exclude-only object preserves matching properties and clears unmatched properties. Protected properties are omitted from dry-run deletions and commands, not merely skipped at apply time.

Org, suborg and repo layers can mix the list and object forms. When any active layer uses the object form, includes merge by case-insensitive property name (name and property_name are aliases), and the more specific layer replaces the complete value, including a multi-value array. Exclusions accumulate across layers; an included property can override an inherited exclusion. Empty lists do not remove inherited entries or patterns. A custom_properties: null layer resets inherited custom-property configuration. All-list configurations retain their existing merge behavior.

Malformed objects, invalid include values, and invalid exclusion entries or regexes are recorded as per-repository configuration errors without aborting other repositories. No values are cleared for that repository; valid non-null included values can still be applied. Schema validation is an authoring aid, not a replacement for these runtime safeguards.

additive_plugins: [custom_properties] remains the simpler option to suppress all undeclared-value clears. It still allows declared values to be applied. disable_plugins strips the configured layers before includes and exclusions are merged, using its existing target semantics.

Smoke phase 20 verifies selective preservation, exact create/update/clear values, include precedence, exclude-only reconciliation, case-insensitive regex escapes, invalid-regex NOP failure, and real-Octokit dry-run PATCH commands. It creates a dedicated smoke-property-exclusions repo and smoke-exclusion-* property definitions, refuses pre-existing fixtures, and removes its owned resources afterward. Run with npm run smoke-test:phase -- 20 only in a dedicated test organization; the harness also runs its normal setup and teardown.

Additive plugins (additive_plugins)

additive_plugins is the complementary "soft mode" to disable_plugins. When a Diffable plugin is listed here, safe-settings will only add and update entries — it will never call remove(). Items that exist on GitHub but are absent from the YAML are preserved, effectively merging external changes with your policy rather than overwriting them.

Declare additive_plugins only in settings.yml (org level) to keep behaviour consistent across all repositories.

Supported plugins (all extend Diffable):

Plugin	Effect in additive mode
labels	Extra labels not in YAML are kept
collaborators	Extra collaborators not in YAML are kept
teams	Extra team permissions not in YAML are kept
milestones	Extra milestones not in YAML are kept
autolinks	Extra autolinks not in YAML are kept
environments	Extra environments not in YAML are kept
custom_properties	Extra property values not in YAML are kept
variables	Extra variables not in YAML are kept
rulesets	Extra rulesets not in YAML are kept
custom_repository_roles	Extra custom roles not in YAML are kept

Important

repository, archive, branches, and validator are not supported. Listing them in additive_plugins will produce a validation error.

NOP mode: when additive_plugins is active and the diff would produce deletions, an informational message — "Additive mode active: N deletion(s) suppressed by additive_plugins" — is included in the PR check-run comment so reviewers can see what is being preserved.

Example — never delete labels or collaborators that were added outside of safe-settings:

additive_plugins:
  - labels
  - collaborators

App installation management (app_installations)

Most safe-settings plugins target a repository. The app_installations plugin is different: its target is a GitHub App installation. It lets you declaratively manage which repositories a GitHub App can access (the app's repository_selection), using the same org → suborg → repo config hierarchy you already use for repository settings.

This is useful for controlling, as code, which repos apps such as Copilot, Dependabot, or your own internal apps are installed on across the org.

Prerequisites

• Safe-settings must be installed on the enterprise with the Enterprise organization installations permission (see the Enterprise organization installations API). Managing app installations requires an enterprise-level token; the regular org installation token is not sufficient. If safe-settings is not installed on the enterprise with this permission, app installation sync is reported as an error and skipped.
• The enterprise slug is read from the webhook event payload (payload.enterprise.slug); no extra environment variable is required.

How repository selection is resolved

The config layer where app_installations is declared determines which repos are selected for the app:

Layer	File	Repos selected for the app
Org	settings.yml	All repos in the org
Suborg	suborgs/*.yml	Repos matching the suborg's targeting (suborgrepos, suborgteams, suborgproperties)
Repo	repos/<repo>.yml	That specific repo

Important

An app listed at the org level (which implies all repos) takes precedence. Suborg/repo-level selections for that same app are ignored, and repos are never removed from it by incremental (suborg/repo) changes — it is reconciled only by the full (scheduled) sync.

Examples

Org-level settings.yml — give an app access to all repos in the org:

app_installations:
  - app_slug: my-internal-app

Suborg-level suborgs/backend.yml — give an app access to the repos targeted by this suborg (here, all repos with the Team=backend custom property):

suborgproperties:
  - Team: backend
app_installations:
  - app_slug: my-internal-app

Repo-level repos/my-repo.yml — add this specific repo to the app:

app_installations:
  - app_slug: my-internal-app

Removing an app from a suborg/repo config (or changing the suborg's targeting) removes the affected repos from that app on the next sync, unless another layer still selects them.

Sync behavior

• Incremental (delta) sync runs when a suborgs/*.yml or repos/*.yml file changes. Only the apps affected by the changed file are reconciled: the previous version of the file is compared with the new one to compute repos to add (repository_selection) and repos to remove (repository_unselection). Additions are applied before removals (422-safe swap), so a repo removed by one config and added by another ends up present.
• Full sync runs on the schedule (cron), on manual sync, and when settings.yml changes. It recomputes the full desired state for every managed app across all layers and reconciles it against the live installation state. This is the mechanism that corrects any configuration drift.
• Add/remove operations are automatically batched in chunks of 50 repos (the API limit).

Note

Drift on managed apps is reconciled by the full (cron) sync, not by webhooks. A GitHub App only receives installation repository events for its own installation, so safe-settings cannot detect — via webhooks — when a human changes another app's repository access. Keep the scheduled sync enabled for timely drift correction.

Disabling and additive mode

app_installations honors both disable_plugins and additive_plugins. In additive mode the plugin only adds repos to installations and never removes them.

The Settings Files

The settings files can be used to set the policies at the org, suborg or repo level.

The following can be configured:

• Repository settings - home page, url, visibility, has_issues, has_projects, wikis, etc.
• Default branch - naming and renaming
• Topics
• Custom properties
• Teams and permissions
• Collaborators and permissions
• Issue labels
• Milestones
• Branch protections - if the name of the branch is default in the settings, it is applied to the default branch of the repo.
• Autolinks
• Repository name validation using regex pattern
• Rulesets
• Environments - wait timer, required reviewers, prevent self review, protected branches deployment branch policy, custom deployment branch policy, variables, deployment protection rules
• App installations - which repositories a GitHub App installation can access (see App installation management)

See docs/sample-settings/settings.yml for a sample settings file.

Note

When using collaborators, teams or labels, you can control which repositories they apply to using include and exclude:

• If include is specified, settings will only apply to repositories that match those patterns
• If exclude is specified, settings will apply to all repositories except those matching the patterns
• If both are specified, exclude takes precedence over include but include patterns will still be respected

Pattern matching uses glob expressions, e.g use * for wildcards. For example:

teams:
  - name: Myteam-admins
    permission: admin
  - name: Myteam-developers
    permission: push
  - name: Other-team
    permission: push
    include:
      - '*-config'

Additional values

In addition to the values in the file above, the settings file can have some additional values:

1. force_create: This is set in the repo-level settings to force create the repo if the repo does not exist.
2. template: This is set in the repo-level settings, and is used with the force_create flag to use a specific repo template when creating the repo
3. suborgrepos: This is set in the suborg-level settings to define an array of repos. This field can also take a glob pattern to allow wild-card expression to specify repos in a suborg. For e.g. test* would include test, test1, testing, etc.
4. The suborgteams section contains a list of teams, and all the repos belonging to the teams would be part of the suborg

Environment variables

You can pass environment variables; the easiest way to do it is via a .env file.

1. CRON you can pass a cron input to run safe-settings at a regular schedule. This is based on node-cron. For eg.

# ┌────────────── second (optional)
# │ ┌──────────── minute
# │ │ ┌────────── hour
# │ │ │ ┌──────── day of month
# │ │ │ │ ┌────── month
# │ │ │ │ │ ┌──── day of week
# │ │ │ │ │ │
# │ │ │ │ │ │
# * * * * * *
CRON=* * * * * # Run every minute

1. Logging level can be set using LOG_LEVEL. For e.g.

LOG_LEVEL=trace

1. Configure the source repository using ADMIN_REPO (default is admin). For e.g.

ADMIN_REPO=safe-settings-config

1. Configure the config path using CONFIG_PATH (default is .github). For e.g.

CONFIG_PATH=.github

1. Configure the settings file path using SETTINGS_FILE_PATH (default is settings.yml). For e.g.

SETTINGS_FILE_PATH=settings.yml

1. Configure the deployment settings file path using DEPLOYMENT_CONFIG_FILE (default is deployment-settings.yml). For e.g.

DEPLOYMENT_CONFIG_FILE=deployment-settings.yml

1. Enable the pull request comment using CREATE_PR_COMMENT (default is true). For e.g.

CREATE_PR_COMMENT=true

1. Add an error count, affected-plugin list, check-run link (when available), and unchecked review-verification checkbox to each pull request comment page using PR_COMMENT_SUMMARY_ENABLED (default is false). Counts and plugins describe the entire operation, not just the current page. Errors count reported error entries after deduplication; affected plugins have rendered changes. GitHub App subjects do not count as affected repositories. Existing repos-considered/affected counts and detailed output remain unchanged when this flag is unset or false. Truncated opt-in comments close their collapsible sections before the checkbox, keeping it outside those sections within the 55,536-character limit. The checkbox is descriptive only; it is not automatically checked or enforced. CREATE_PR_COMMENT=false still suppresses all PR comments. For e.g.

PR_COMMENT_SUMMARY_ENABLED=true

1. Block repository renaming manually using BLOCK_REPO_RENAME_BY_HUMAN (default is false). For e.g.

BLOCK_REPO_RENAME_BY_HUMAN=true

1. Create the configured default_branch instead of renaming the current one using CREATE_DEFAULT_BRANCH (default is false). By default, when repository.default_branch is configured and differs from the repo's current default branch, safe-settings renames the current default branch to the configured name. When CREATE_DEFAULT_BRANCH=true, if the configured default_branch does not exist, safe-settings instead creates a new branch off the current default branch and sets it as the default, leaving the existing default branch untouched (no rename). If the configured branch already exists, it is simply set as the default. For e.g.

CREATE_DEFAULT_BRANCH=true

1. Scope CLI and cron full syncs to one account using GH_ORG. For e.g.

GH_ORG=my-org

A full sync (CRON or npm run full-sync) selects the installation whose account login matches GH_ORG case-insensitively and reads configuration from that account's <ADMIN_REPO>. If no installation matches, including when the app has no installations, the sync fails instead of syncing another account. This also applies to dry runs (FULL_SYNC_NOP=true); the CLI exits nonzero on a targeting error, and cron reports the execution as failed. When GH_ORG is unset, all repository-owning installations are synced sequentially. Set it to restrict full sync to one account; enterprise installations supply app-management context but are not repository sync targets. Webhook events still use their own installation context. GH_ORG is also used by the manifest flow to choose where the app is registered.

Runtime Settings

1. Besides the above settings files, the application can be bootstrapped with runtime settings.
2. The runtime settings are configured in deployment-settings.yml that is in the directory from where the GitHub app is running.
3. Currently the only setting that is possible are restrictedRepos: [... ] which allows you to configure a list of repos within your org that are excluded from the settings. If the deployment-settings.yml is not present, the following repos are added by default to the restrictedrepos list: 'admin', '.github', 'safe-settings'

Notes

1. Label color can also start with #, e.g. color: '#F341B2'. Make sure to wrap it with quotes!
2. Each top-level element under branch protection must be filled (eg: required_pull_request_reviews, required_status_checks, enforce_admins and restrictions). If you don't want to use one of them you must set it to null (see comments in the example above). Otherwise, none of the settings will be applied.
3. The precedence order is repository > suborg > org (.github/repos/.yml > .github/suborgs/.yml > .github/settings.yml

How to use

1. Create an admin repo (or an alternative of your choosing) within your organization. Remember to set ADMIN_REPO if you choose something other than admin. See Environment variables for more details.

2. Add the settings for the org, suborgs, and repos. Sample files can be found here.

3. Deploy and install the app. Alternatively, the GitHub Actions Guide describes how to run safe-settings with GitHub Actions.

Smoke Testing

The repository includes an end-to-end smoke test script (smoke-test.js) that validates safe-settings against a live GitHub organization. It starts the app, creates repos/configs via the API, and verifies that safe-settings correctly applies and enforces settings.

Prerequisites

• Node.js (same version used to run safe-settings)
• gh CLI — authenticated and available on PATH (used for drift-remediation tests only)
• A GitHub App installed on the target org with the required permissions
• A .env file in the project root (see below)

Authentication

The smoke test uses two authentication methods:

• GitHub App token (via APP_ID + PRIVATE_KEY) — used for the majority of tests: creating configs, merging PRs, validating repos, teams, rulesets, custom properties, etc.
• Fine-grained PAT (via GH_TOKEN) — used only in Phase 2 (team removal) and Phase 3 (rogue ruleset creation). These drift-remediation tests must appear as a human action because safe-settings ignores webhook events where sender.type is Bot.

Configuration

Add the following to your .env file:

Variable	Description	Required
GH_ORG	Target GitHub organization (e.g. my-org)	Yes
APP_ID	GitHub App ID	Yes
PRIVATE_KEY	GitHub App private key (use \n for newlines)	Yes
WEBHOOK_PROXY_URL	Smee.io proxy URL for webhooks	Yes
ADMIN_REPO	Admin repo name (default: admin)	No
CONFIG_PATH	Config path within admin repo (default: .github)	No
GH_TOKEN	Fine-grained PAT with org admin + repo permissions	Yes
SMOKE_VERBOSE	Set to 1 to show live safe-settings logs	No

Running

# Run all phases
npm run smoke-test
# or
node smoke-test.js

# Interactive mode — pause after each phase for manual validation
npm run smoke-test:interactive
# or
node smoke-test.js --interactive

# Run a single phase (with setup + teardown)
npm run smoke-test:phase -- 3
# or
node smoke-test.js --phase 3

# Run a range of phases
npm run smoke-test:phase -- 1-3
node smoke-test.js --phase 1-3

# Run specific comma-separated phases
npm run smoke-test:phase -- 1,3,5
node smoke-test.js --phase 1,3,5

# Bypass actor apply/NOP convergence (Phase 1 creates the required test repo)
node smoke-test.js --phase 1,19

# Explicit GH_ORG full-sync CLI dry run (setup + Phase 24 + teardown)
node smoke-test.js --phase 24

# Exact team slug NOP diffs and real permission updates (setup + Phase 26 + teardown)
node smoke-test.js --phase 26

# Opt-in PR comments with owned PR/check-run readback (setup + Phase 27 + teardown)
node smoke-test.js --phase 27

# Repository and environment variables, including pagination and no-op convergence
node smoke-test.js --phase 1,13,21

# Full-sync NOP with real Settings, restricted to the verified test-org installation
# Clear harness CRON; see the spawned-server configuration warning below.
CRON= node smoke-test.js --phase 22

# Mix range + interactive
npm run smoke-test:phase -- 1-3 interactive
node smoke-test.js --phase 1-3 --interactive

What it tests

The smoke test runs the following phases:

Phase	Description
Setup	Initializes the admin repo with an empty settings.yml, removes stale test repos, and starts safe-settings
Phase 1	Creates a repo config (test), validates NOP mode via check runs, merges, and verifies repo creation, teams, custom properties, and rulesets
Phase 2	Removes a team from the repo and verifies safe-settings re-adds it (drift remediation)
Phase 3	Creates a rogue ruleset and verifies safe-settings removes it (drift remediation)
Phase 4	Creates demo-repo-service1 with teams, topics, and branch protection
Phase 5	Creates a property-targeted suborg config, verifies suborg rulesets apply to two matching repos, then changes one repo's custom property and verifies the ruleset is removed only from the repo that no longer matches
Phase 6	Archives demo-repo-service1 and verifies the repo is archived
Phase 7	Creates demo-repo-service2 and verifies suborg rulesets are inherited
Phase 7b	Tests external group team sync
Phase 8	Creates org-level settings (custom repository roles + org rulesets) and verifies they are applied
Phase 10	Validates disable_plugins — ensures disabled plugins are skipped
Phase 11	Validates additive_plugins — verifies additive-mode plugin behaviour
Phase 12	Tests custom_properties plugin
Phase 13	Tests the variables plugin (create, update, remove variables)
Phase 19	Tests ignored OrganizationAdmin/DeployKey IDs, order-independent NOP convergence, real bypass-mode/role-ID changes, and no redundant updates (requires Phase 1)
Phase 21	Reads all pages of 101 repository variables and 100 environment variables, verifies unchanged NOP/apply writes nothing, and updates only the two boundary variables (requires Phases 1 and 13)
Phase 22	Runs real full-sync NOP against the verified GH_ORG installation with controlled enumeration/auth and read-only, org-scoped requests. Requires only Setup; multi-installation fanout and failure isolation are covered by unit/CLI tests, not this single-org smoke
Phase 23	Tests real config loading, empty YAML, directories, and missing-file/ref HTTP 404s on an owned temporary branch (requires only Setup)
Phase 24	Runs the real full-sync CLI with explicit GH_ORG and FULL_SYNC_NOP=true, verifying the selected installation and successful completion; multi-account, no-match, and cron failure scenarios use local mocked tests
Phase 25	Creates an owned smoke-archived-repo, archives it with real Settings, verifies listing-based skips make zero fixture requests, checks unknown-state and labels-only fallbacks, then verifies unarchive NOP/apply and convergence (requires only Setup; no webhooks)
Phase 26	Creates an owned smoke-team-slug repo and Smoke Team Slug 26 team, checks unchanged NOP/apply and exact permission-change diffs against the real API, then verifies one PUT and convergence while preserving inherited security-manager teams. Requires only Setup, refuses existing fixture names, and removes both fixtures even on failure. It does not require webhook forwarding.
Phase 27	Creates an owned private smoke-pr-comment-summary repo, PR and check runs; calls real reporting with controlled NOP rows and reads back posted comments. Covers unset/false/true summary flags, no-op and disabled comments, subject-aware counts, errors/warnings/info, pagination and the exact truncation limit with review footers. Removes the owned repo and all its fixtures even on failure. Requires only Setup, refuses existing fixture names, and needs no webhook forwarding. This checks generated markup, not visual rendering or reconciliation.
Teardown	Shuts down safe-settings, deletes test repos, teams, custom roles, and rulesets

Run config-loading and existing config-validation coverage with node smoke-test.js --phase 10,23 (Setup and Teardown run automatically). Phase 23 removes its temporary smoke-test-phase23 branch even on failure. HTTP 403/500 and network failures are injected only in local unit tests, which also verify original-error identity and full-sync/NOP reporting; the live smoke does not change permissions or provoke rate limits.

Run archived-repository coverage with node smoke-test.js --phase 25. Phase 25 refuses to reuse an existing fixture and removes only the repository it created, including on failure. It exercises real installation repository listing and repository/label plugins with fixture-only in-memory configuration; it does not test webhook delivery or multi-organization sync. Setup and Teardown still touch the shared fixtures described above, so inventory those resources first. The offline npm run test:archived suite verifies exact request counts, configuration precedence, NOP behavior, and safe PATCH payloads using the installed Octokit.

The harness rejects a nonempty CRON before authentication or setup. However, the spawned Probot CLI reloads .env and can override explicit environment values. This check does not guarantee that the server's CRON, webhook forwarding, or enterprise verification remains disabled. Phase 22's controlled installation boundary applies to its direct NOP call, not the spawned server; verify the server's effective configuration separately before running against a shared App.

GitHub limits each environment to 100 variables, so Phase 21 stays within that live limit. The phase records actual API next links and page sizes: GitHub can return fewer than the requested 100 items per page. Run npm run test:pagination on Node 22+ for real installed-Octokit loopback coverage of 101 repository and environment variables, server-capped and cursor pagination, empty pages, and API failures without live credentials.

The archived-repository, pagination, and team transport suites live under test/integration/transport/ and run automatically with npm run test:integration and npm run test:integration:ci, including in Node.js CI. They use Node's native test runner and real Octokit clients against local HTTP or controlled fetch fixtures, with no live GitHub credentials required. The test:pagination and test:archived commands remain available for focused runs.

Output

The script uses colored terminal output with pass (✅) / fail (❌) indicators and prints a summary at the end:

══════════════════════════════════════
  Results: 45 passed, 0 failed
══════════════════════════════════════

Generating settings from existing configuration

Safe-settings normally works "forward": you declare settings in YAML and it applies them to GitHub. The settings generator does the reverse — it reads the current state of a repo, an org, or a collection of repos (a suborg) and produces the corresponding safe-settings YAML (repos/<name>.yml, settings.yml, or suborgs/<name>.yml). This is useful for onboarding existing repositories/orgs onto safe-settings without hand-authoring config.

It can be invoked two ways:

• Standalone CLI (generate-settings.js) — writes the generated file to your local filesystem.
• App trigger via a repository_dispatch event — the running app generates the file and opens a pull request against the admin repo.

Source types

source_type	source_value	What is extracted	Output file
repo	repository name	All repo-level plugins (repository, labels, collaborators, teams, milestones, branches, autolinks, custom_properties, variables, environments, repo rulesets)	repos/<repo>.yml
org	org login	Org-level rulesets and custom repository roles only	settings.yml
custom-property	name=value (e.g. Team=backend)	Repo-level settings common to all repos carrying that custom property value (intersection)	suborgs/<name>_<value>.yml

Note on suborgs: for custom-property, the generator discovers every repo with the given custom property value, extracts each repo's config, and keeps only the settings that are identical across all of them. A suborgproperties selector is prepended automatically.

Overwrite behavior

By default (overwrite=false) the generator will not replace an existing file. If the target already exists it writes a <name>.sample.yml file next to it instead. Set overwrite=true to replace the file.

Standalone invocation

The CLI loads variables from a .env file in the project root (APP_ID, PRIVATE_KEY, and optionally GH_ORG/OWNER). Options can be passed as flags or environment variables.

# Generate repos/my-repo.yml from a single repository
node generate-settings.js \
  --source-type repo \
  --source-value my-repo \
  --owner my-org \
  --output-dir ./out

# Generate settings.yml from org-level rulesets + custom repository roles
node generate-settings.js --source-type org --source-value my-org --output-dir ./out

# Generate suborgs/Team_backend.yml from all repos with the custom property Team=backend
node generate-settings.js \
  --source-type custom-property \
  --source-value "Team=backend" \
  --owner my-org \
  --output-dir ./out

# Overwrite an existing file instead of writing a .sample.yml
node generate-settings.js --source-type repo --source-value my-repo --owner my-org --overwrite

# Using environment variables instead of flags
SOURCE_TYPE=repo SOURCE_VALUE=my-repo OWNER=my-org OUTPUT_DIR=./out node generate-settings.js

Flag	Env var	Description	Default
--source-type	SOURCE_TYPE	repo, org, or custom-property	(required)
--source-value	SOURCE_VALUE	repo name / org login / name=value	(required)
--property-name	SOURCE_PROPERTY_NAME	Custom property name (alternative to encoding it in --source-value)	—
--owner	OWNER / GITHUB_ORG / GH_ORG	Org login (selects the matching App installation)	first installation
--output-dir	OUTPUT_DIR	Directory to write generated files into	.
--overwrite	OVERWRITE=true	Replace existing files instead of writing .sample.yml	false

App invocation (repository_dispatch)

When the app is running, trigger generation by sending a repository_dispatch event (with event_type: safe-settings-generate) to the admin repo. The app generates the file and opens a PR against the admin repo's default branch.

# Generate a repo config and open a PR
gh api --method POST \
  /repos/my-org/admin/dispatches \
  -f event_type=safe-settings-generate \
  -F 'client_payload[source_type]=repo' \
  -F 'client_payload[source_value]=my-repo' \
  -F 'client_payload[overwrite]=false'

# Generate org-level settings.yml and open a PR
gh api --method POST \
  /repos/my-org/admin/dispatches \
  -f event_type=safe-settings-generate \
  -F 'client_payload[source_type]=org' \
  -F 'client_payload[source_value]=my-org'

# Generate a suborg config from a custom property
gh api --method POST \
  /repos/my-org/admin/dispatches \
  -f event_type=safe-settings-generate \
  -F 'client_payload[source_type]=custom-property' \
  -F 'client_payload[source_value]=Team=backend' \
  -F 'client_payload[overwrite]=false'

The client_payload fields are:

Field	Description	Required
source_type	repo, org, or custom-property	Yes
source_value	repo name / org login / name=value	Yes
property_name	Custom property name (alternative to encoding it in source_value)	No
overwrite	true to replace an existing file; otherwise a .sample.yml is created	No (default false)

Tip: Always review the generated PR before merging. Running safe-settings in NOP mode against the generated config should report no unexpected diffs.

Generated changes always go through a pull request

The app never commits generated configuration directly to the admin repo's default branch. Every repository_dispatch invocation produces a pull request that must be reviewed and merged before it takes effect. Concretely, for each request the app:

1. Creates a new branch off the admin repo's default branch (safe-settings-generate/<source_type>-<source_value>-<timestamp>).
2. Commits the generated YAML to that branch only.
3. Opens a pull request from that branch against the default branch.

This means it is safe to give developers write access to the admin repo so they can trigger generation: a repository_dispatch event can only create a branch and open a PR — it cannot change the live configuration on its own. The generated config does not reach the path safe-settings acts on until the PR is merged, so all changes are subject to your normal review process and any branch protection / required-reviews rules configured on the admin repo's default branch.

To enforce review, protect the admin repo's default branch (for example, require pull request reviews and disallow direct pushes). Because the generator only ever writes to a feature branch and opens a PR, those rules apply to every generated change.

License

safe-settings is licensed under the ISC license

safe-settings uses 3rd party libraries, each with their own license. These are found here.