Phase A moved the CFN execution role's boundary-gated IAM statements into the attached seahaven-cfn-exec-iam-management managed policy, with the escalation fixed and three Deny backstops, while deliberately leaving the old inline iam-role-management-boundary-gated policy in place so that deploy removed nothing. That is now redundant and this removes it. Effective permissions are unchanged, proven statically before deploying: of the 6 Allow statements being removed, 5 are byte-identical to the managed policy's. The only difference is Sid IAMPutPermissionsBoundary, where the inline copy also listed iam:DeleteRolePermissionsBoundary -- the action DenyBoundaryTampering explicitly denies, so that Allow was already inert. Frees the scarce budget: inline usage drops from 10,006 to 8,261 of IAM's 10,240-byte per-role limit, leaving 1,979 bytes of headroom on a role that previously had 234.
17 KiB
.github
Organization-level GitHub configuration for Sea Haven Industries.
What's in here
Reusable Workflows
.github/workflows/ci-python-sam.yaml — Reusable CI workflow for Python / SAM repos. Runs ruff check + ruff format --check, optional pytest, and optional sam validate --lint. Also usable for Python CDK repos by disabling SAM validate.
.github/workflows/ci-typescript-cdk.yaml — Reusable CI workflow for TypeScript / CDK repos. Runs npm ci + optional tsc --noEmit, optional ESLint, optional Jest, and optional cdk synth. Also supports Node.js SAM repos via an optional sam validate step.
.github/workflows/ci-typescript-frontend.yaml — Reusable CI workflow for bundled TypeScript front-end apps (Vite / React / Vue SPAs). Runs a Sea Haven standards gate (required npm scripts present, no AI-tool footers / hook bypasses / hardcoded secrets in added lines), then format:check, lint, build, vitest unit tests, and an optional Playwright browser smoke. Emits the single ci / ci status context — keep the caller job id ci.
.github/workflows/cd-sam.yaml — Reusable CD workflow for SAM repos. Runs sam build + sam deploy with OIDC credentials and a CloudFormation execution role. Triggers via workflow_call from per-repo deploy.yaml on push to main.
.github/workflows/cd-cdk.yaml — Reusable CD workflow for CDK repos (TypeScript and Python). Runs cdk deploy --all with OIDC credentials. Supports optional Python setup for Python CDK repos and QEMU emulation for cross-platform Docker builds.
.github/workflows/ci-python-app.yaml — Reusable CI for non-SAM Python apps (ruff check + format, optional pytest; no SAM validate).
.github/workflows/ci-dotnet.yaml — Reusable CI for .NET solutions (dotnet build, optional dotnet test; SDK version and solution path as inputs).
.github/workflows/ci-static.yaml — Reusable CI for static sites (e.g. Eleventy builds for seahaven-site).
.github/workflows/cd-mobile-ios.yaml — Reusable CD for iOS apps via Fastlane to TestFlight (Node + Ruby setup inputs).
.github/workflows/cd-dotnet-eb.yaml — Reusable CD for .NET apps on AWS Elastic Beanstalk. Publishes the project, packages a bundle, uploads it, creates an application version, and updates an existing environment with OIDC credentials — it never creates an environment. Serialised per environment via a concurrency group, and the post-deploy check fails the job if EB rolls the deploy back. The caller owns branch-to-environment mapping.
.github/workflows/callable-labeler.yaml — Org-wide PR auto-labeler. Label rules live inline here (single source of truth) — consumer repos need only a thin caller with contents: read, pull-requests: write, and issues: write; no per-repo labeler.yml.
.github/workflows/callable-dependency-review.yaml — Dependency review on PRs, failing on high severity. Requires Dependency Graph.
.github/workflows/ci.yaml — Self-CI for this repo: actionlint (checksum-verified install) over all workflow files, emitting the required ci / ci status context.
.github/workflows/compliance-audit.yaml — DEPRECATED (2026-06-10). The weekly scheduled org-wide audit has been retired: the schedule was removed and the workflow is disabled in the Actions tab (manual workflow_dispatch only, kept for historical reference). Repo compliance is now handled by the Claude Code App on pull requests and the engineering handbook directly. Safe to delete in a future cleanup.
Workflow templates (workflow-templates/)
Starter workflows offered on the org's Actions → New workflow page: ci-python, ci-node, cdk-deploy, sam-deploy, dotnet-eb-deploy, dependency-review, labeler, triage. Each is a thin caller of the corresponding reusable workflow above (triage is standalone). Every template has a paired properties.json (name, description, icon, filePatterns for auto-suggestion). Replace any REPLACE-ME placeholders before enabling.
Action pinning policy
Third-party action refs across the org follow a tiered policy:
- High-trust / high-blast-radius third-party actions are SHA-pinned with a trailing version comment (e.g.
actions/labelerincallable-labeler.yaml), and binary installs are checksum-verified (actionlint inci.yaml). Dependabot keeps the SHA current via its trailing-comment mechanism. - Common first-party actions (
actions/checkout,actions/dependency-review-action,actions/github-script) are pinned to a major tag (@v7,@v5, …) and kept current by Dependabot version updates gated by CI. - Org reusable workflows are referenced at
@main(uses: Sea-Haven-Industries/.github/.github/workflows/…@main). This is deliberate: caller and callable share one trust domain, and pinning callers to a SHA would freeze every consumer against central fixes. Templates inworkflow-templates/follow the same@mainconvention.
PR Reviews
PR reviews are handled by the official Claude Code GitHub App (installed org-wide, enabled as a required check in the org ruleset) — there is no review workflow in this repo. The earlier custom claude-code-review.yaml reusable workflow and its per-repo wrapper were retired on 2026-05-13 when the App took over.
Scripts
scripts/rollout-review-workflow.sh — Legacy / superseded. One-time script that pushed the old PR-review wrapper workflow to all org repos. Obsolete since reviews moved to the official Claude Code App (2026-05-13); retained only for historical reference.
AWS deploy roles & IAM (oidc-deploy-roles.yaml)
oidc-deploy-roles.yaml is a bootstrap CloudFormation stack (github-oidc-deploy-roles, us-east-1, account 328440206208) that owns the IAM the CI/CD workflows assume. It contains:
- The GitHub Actions OIDC provider (conditional — already exists in the account).
- One OIDC deploy role per repo (
githubdeploy-<repo>), assumed by that repo'sdeploy.yamlvia OIDC and passed in asAWS_DEPLOY_ROLE_ARN. CDK repos use these to assume thecdk-hnb659fds-*bootstrap roles; SAM repos use these to runsam deploy. - The shared SAM CloudFormation execution role
github-cfn-execution-role(SamCfnExecutionRole) — passed ascfn-role-arnby every SAMdeploy.yaml(see §5). CloudFormation assumes it to provision the SAM stacks' resources. - The
seahaven-lambda-execution-boundarymanaged policy. - The
seahaven-cfn-exec-iam-managementmanaged policy (SamCfnIamManagementPolicy), attached togithub-cfn-execution-role. It holds that role's boundary-gated IAM statements plus the Deny backstops that keep the permissions boundary from being detached, rewritten, or applied to the deploy substrate's own roles. It lives in a managed policy rather than inline because the role's inline policies sit at 10,006 of IAM's hard 10,240-byte per-role limit; attached managed policies have a separate 6,144-byte budget.
Constraint for future maintainers.
github-cfn-execution-roleis explicitly denied from mutating the deploy substrate's own principals — itself, anygithubdeploy-*role, and anyseahaven-*managed policy. Those are owned by this stack and deployed manually with administrator credentials, so nothing legitimate needs that path. If you ever add automation that manages one of them, it must not run throughgithub-cfn-execution-roleor it will fail withAccessDenied.
Where the exec role's IAM statements live. All of them are in the attached
seahaven-cfn-exec-iam-managementmanaged policy — there is no inline copy. The role's inline policies were previously at 10,006 of the 10,240-byte limit, leaving no room to add anything; consolidating into the managed policy brought that to 8,261 bytes (1,979 free). If you need to add a permission to this role, prefer the managed policy: the inline budget is the scarce one.
⚠️ This stack has no CD pipeline — it is deployed manually. (It defines the very roles the pipelines use, so it can't deploy itself.)
# Review IAM changes FIRST (IAM changes also require the cross-family review per the handbook):
aws cloudformation deploy \
--region us-east-1 \
--stack-name github-oidc-deploy-roles \
--template-file oidc-deploy-roles.yaml \
--capabilities CAPABILITY_NAMED_IAM \
--s3-bucket cdk-hnb659fds-assets-328440206208-us-east-1 \
--no-execute-changeset
# inspect the printed change-set, then drop --no-execute-changeset to apply.
--s3-bucket is required — the template is larger than the 51,200-byte inline limit.
github-cfn-execution-role is scoped (no *FullAccess)
The execution role carries no blanket *FullAccess/IAMFullAccess — only per-service inline policies. Its iam:CreateRole / iam:AttachRolePolicy / iam:PutRolePolicy are conditioned on iam:PermissionsBoundary == seahaven-lambda-execution-boundary, so it can only create roles that carry the boundary (it cannot mint an unconstrained admin role). Adding a new AWS service to a SAM stack means adding that service's provisioning actions to this role, or the deploy fails.
seahaven-lambda-execution-boundary is the Lambda runtime ceiling
Every SAM-created Lambda execution role gets this boundary attached — SAM stacks set it on Globals.Function:
Globals:
Function:
PermissionsBoundary: arn:aws:iam::328440206208:policy/seahaven-lambda-execution-boundary
A function's effective permissions are the intersection of its own role policy and this boundary. A new runtime permission must also be added to the boundary, or it is silently denied at runtime (the deploy still succeeds — the failure only shows when the function runs). SAM does not support a custom Path on auto-generated function roles, so the boundary condition (not a role path) is the escalation guard.
Order of operations when changing the exec role or boundary
- Deploy the boundary change first.
- Redeploy the SAM stacks so their roles pick it up (while the exec role still permits it).
- Then tighten the exec role.
Wrong order breaks every SAM deploy. (History: INFRA-103 established the boundary, INFRA-97 scoped the role.) CDK repos are unaffected — they deploy via cdk-hnb659fds-* roles, not this execution role.
This ordering rule is about changing the boundary or the conditions that gate it. It does not apply to changes that only add permissions to the exec role.
Permissions boundaries can no longer be removed by CloudFormation
github-cfn-execution-role is explicitly denied iam:DeleteRolePermissionsBoundary. It can set the boundary (that is what SAM needs) but never remove one. Two consequences worth knowing before debugging a stuck stack:
- Removing
PermissionsBoundaryfrom an existing role fails by design. CloudFormation issuesDeleteRolePermissionsBoundaryfor that edit, getsAccessDenied, and the stack update rolls back. Removing the boundary from a SAM function is a security regression, so failing loudly is intended. - Rollback of an update that adds a boundary to an existing role would also fail, landing the stack in
UPDATE_ROLLBACK_FAILED. This is currently unreachable — all 26 IAM roles across the five SAM stacks already carry the boundary (verified 2026-07-27), so no update can add one. It becomes reachable again only if a role is created without the boundary and given one later.
Recovery in either case is an administrator action, not a pipeline retry: clear the wedged stack with aws cloudformation continue-update-rollback --stack-name <stack> --resources-to-skip <RoleLogicalId>, or replace the role by renaming its logical id. Note cd-sam's pre-flight hard-fails on *ROLLBACK_COMPLETE, so that repo's deploys stay blocked until it is cleared.
Setup
1. Create a GitHub App
- Go to Organization Settings > Developer settings > GitHub Apps > New GitHub App
- Name it
claude-code-ci(or similar) - Set Homepage URL to your org URL
- Disable Webhook (uncheck "Active")
- Set these Repository permissions:
- Contents: Read and write
- Issues: Read and write
- Metadata: Read-only
- Pull requests: Read and write
- Set Where can this app be installed? to "Only on this account"
- Click Create GitHub App
- Note the App ID from the app's settings page
- Under Private keys, click Generate a private key — save the
.pemfile
2. Install the App
- From the app's settings page, click Install App
- Select
Sea-Haven-Industries - Choose All repositories
3. Add org-level secrets
Go to Organization Settings > Secrets and variables > Actions and add:
| Secret | Value |
|---|---|
ANTHROPIC_API_KEY |
Your Claude API key |
CLAUDE_CI_APP_ID |
The App ID from step 1 |
CLAUDE_CI_APP_PRIVATE_KEY |
The full contents of the .pem file from step 1 |
4. Add CI to a repo
Create .github/workflows/ci.yaml in the target repo. Examples:
Python SAM repo (e.g., afterhours-shift-manager, expense-approval-bot):
name: CI
on:
pull_request:
branches: [main]
jobs:
ci:
uses: Sea-Haven-Industries/.github/.github/workflows/ci-python-sam.yaml@main
TypeScript CDK repo (e.g., seahaven-door-unlock-api, seahaven-slack-bot):
name: CI
on:
pull_request:
branches: [main]
jobs:
ci:
uses: Sea-Haven-Industries/.github/.github/workflows/ci-typescript-cdk.yaml@main
Node.js SAM repo (e.g., payments-dashboard):
name: CI
on:
pull_request:
branches: [main]
jobs:
ci:
uses: Sea-Haven-Industries/.github/.github/workflows/ci-typescript-cdk.yaml@main
with:
run-typecheck: false
run-cdk-synth: false
run-sam-validate: true
Mixed stack (e.g., exec-aide — TypeScript CDK + Python Lambdas):
name: CI
on:
pull_request:
branches: [main]
jobs:
python:
uses: Sea-Haven-Industries/.github/.github/workflows/ci-python-sam.yaml@main
with:
source-dirs: "src"
run-sam-validate: false
typescript:
uses: Sea-Haven-Industries/.github/.github/workflows/ci-typescript-cdk.yaml@main
5. Add CD to a repo
Create .github/workflows/deploy.yaml in the target repo. Requires AWS_DEPLOY_ROLE_ARN repo secret.
SAM repo (e.g., afterhours-shift-manager):
name: Deploy
on:
push:
branches: [main]
jobs:
deploy:
uses: Sea-Haven-Industries/.github/.github/workflows/cd-sam.yaml@main
with:
stack-name: afterhours-shift-manager
cfn-role-arn: arn:aws:iam::328440206208:role/github-cfn-execution-role
secrets:
deploy-role-arn: ${{ secrets.AWS_DEPLOY_ROLE_ARN }}
The
cfn-role-arn(github-cfn-execution-role) is scoped and boundary-gated — adding a new AWS service or a new Lambda runtime permission to a SAM stack may require updating that role and/orseahaven-lambda-execution-boundaryfirst. See AWS deploy roles & IAM.
TypeScript CDK repo (e.g., seahaven-door-unlock-api):
name: Deploy
on:
push:
branches: [main]
jobs:
deploy:
uses: Sea-Haven-Industries/.github/.github/workflows/cd-cdk.yaml@main
secrets:
deploy-role-arn: ${{ secrets.AWS_DEPLOY_ROLE_ARN }}
Python CDK repo (e.g., po-ingest):
name: Deploy
on:
push:
branches: [main]
jobs:
deploy:
uses: Sea-Haven-Industries/.github/.github/workflows/cd-cdk.yaml@main
with:
python-version: "3.12"
cdk-dir: cdk
secrets:
deploy-role-arn: ${{ secrets.AWS_DEPLOY_ROLE_ARN }}
CDK repo with arm64 Docker builds (e.g., exec-aide):
name: Deploy
on:
push:
branches: [main]
jobs:
deploy:
uses: Sea-Haven-Industries/.github/.github/workflows/cd-cdk.yaml@main
with:
enable-qemu: true
secrets:
deploy-role-arn: ${{ secrets.AWS_DEPLOY_ROLE_ARN }}
.NET app on Elastic Beanstalk (e.g., shoc-backend):
name: Deploy
on:
push:
branches: [dev]
jobs:
deploy:
uses: Sea-Haven-Industries/.github/.github/workflows/cd-dotnet-eb.yaml@main
with:
project: Api.SeaHavenIndustries/Api.SeaHavenIndustries.csproj
eb-application: shoc-backend
eb-environment: shoc-backend-dev
procfile-command: dotnet Api.SeaHavenIndustries.dll
secrets:
deploy-role-arn: ${{ secrets.AWS_DEPLOY_ROLE_ARN }}
The environment must already exist — this workflow deploys a new application version to it and never creates one. Branch-to-environment mapping belongs in the caller: add one job per branch (e.g.
dev→…-dev,main→…-staging) rather than parameterising the reusable by branch.procfile-commandgenerates the Procfile the Amazon Linux .NET platform needs; omit it only if the repo commits its own Procfile into the publish output. Deploys are serialised per environment, and the job fails if Elastic Beanstalk rolls the version back.
Enable optional steps as repos adopt them:
| Input | Default | Turn on when... |
|---|---|---|
run-tests |
false |
Repo has pytest tests or Jest tests |
run-lint |
false |
Repo has an ESLint config |
run-typecheck |
true |
Repo has tsconfig.json |
run-cdk-synth |
true |
Repo is CDK-based |
run-sam-validate |
true (Python) / false (TS) |
Repo has a SAM template |
6. PR reviews
PR reviews run via the official Claude Code GitHub App — install it on the org and enable it as a required check in the ruleset. No per-repo workflow or rollout is needed; the legacy rollout-review-workflow.sh is retained only for historical reference.