## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Paperclip Cloud deploys verified images from merged source commits. > - Cloud readiness waits for every release verification check. > - Runner verification currently runs long TypeScript tests before Rust checks. > - These checks can run on independent runners with their own build directories. > - This PR runs them in parallel while preserving all checks and the shared dependency cache. ## Linked Issues or Issue Description Refs #13194. Related prior work: #13142 and #13259. A search found no duplicate parallel release-check change. **What existing behavior does this improve?** Time from merge to Cloud source verification and deployment readiness. **Current behavior** Recent successful runs take roughly 13 minutes from merge to deployable. In run 34705914878, Runner verification took 11m23s. Protocol tests finished before Rust tests and API authority checks started. **Proposed behavior** Run protocol and Rust verification in two matrix jobs. Cloud readiness still requires both jobs to pass. **Reason and benefit** Remove the serial dependency between independent checks. Expected improvement is about 2–3 minutes on a typical cached run, until the image build or server tests become the longest job. This is an estimate; post-merge timing will confirm it. **Breaking changes** Individual release Runner job names gain a lane suffix. Cloud source and readiness marker names stay the same. PR runner routing is unchanged. ## What Changed - Split release Runner checks into protocol and Rust lanes. Keep every constituent of `check:all` exactly once. - Restore the existing Rust dependency cache in both lanes. Allow only the Rust lane to save it after warming both build profiles. - Add coverage and cache authorization regressions. Document the parallel verification and single cache writer. ## Verification - Passed 477 workflow and source-verification tests with `node --test .github/scripts/tests/*.test.mjs scripts/cloud-source-verification.test.mjs scripts/__tests__/release-verify-workflow.test.mjs`. - Passed `actionlint`, `git diff --check`, and the private AWS routing regression suite. - Passed local `pnpm -r typecheck` and the standalone `check:runner && check:api-authority` lane, including all 1,671 API tests before the protocol lane had built TypeScript output. - The broad local protocol run under Node 25 had four failures. The two affected files passed under CI's Node 24.19.0: 67 passed, 6 platform skips. - Local `pnpm test:run` aborted when disk space ran out; local `pnpm build` could not run afterward. These are local verification limits. [Linux CI run 34710421424](https://github.com/paperclipai/paperclip/actions/runs/34710421424) passed full typecheck, all grouped tests, native verification, build, release dry run, and browser checks. Native protocol CI passed 1,986 tests, plus 1,671 API tests and the Rust suites. - Latest-head Greptile is 5/5 with no open findings. All 33 current-head checks are successful or intentionally skipped. ## Risks - Uses one additional short-lived verification runner per release verification. The existing AWS exact-master restriction remains in place. - The Rust lane warms debug dependencies so its cache save also serves protocol tests. Both lanes always rebuild workspace code. - A workflow regression could omit a check. The new coverage test compares the matrix checks directly with `check:all`; Cloud readiness depends on the complete reusable workflow. ## Model Used OpenAI GPT-6 through Codex, with reasoning, repository tools, and code execution. The exact serving model ID and context window are not exposed by this environment. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with version and capability details) - [x] I have checked ROADMAP.md and confirmed this PR does not duplicate planned core work - [x] I have searched GitHub for duplicate or related PRs and linked them above - [x] I have either (a) linked existing issues with `Fixes: #` / `Closes #` / `Refs #` OR (b) described the issue in-PR following the relevant issue template - [x] I have not referenced internal/instance-local Paperclip issues or links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip` URLs) - [x] My branch name describes the change (e.g. `docs/...`, `fix/...`) and contains no internal Paperclip ticket id or instance-derived details - [x] I have run tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation to reflect my changes - [x] I have considered and documented any risks above - [x] All Paperclip CI gates are green - [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups - [x] I will address all Greptile and reviewer comments before requesting merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
14 KiB
Release Automation Setup
This document covers the GitHub and npm setup required for the current Paperclip release model:
- automatic canaries from
master - manual stable promotion from a chosen source ref
- npm trusted publishing via GitHub OIDC
- protected release infrastructure in a public repository
Repo-side files that depend on this setup:
.github/workflows/release.yml.github/CODEOWNERS
Note:
- the release workflows intentionally use
pnpm install --no-frozen-lockfile - this matches the repo's current policy where
pnpm-lock.yamlis refreshed by GitHub automation after manifest changes land onmaster - the publish jobs then restore
pnpm-lock.yamlbefore runningscripts/release.sh, so the release script still sees a clean worktree
1. Merge the Repo Changes First
Before touching GitHub or npm settings, merge the release automation code so the referenced workflow filenames already exist on the default branch.
Required files:
.github/workflows/release.yml.github/CODEOWNERS
2. Configure npm Trusted Publishing
Do this for every public package that Paperclip publishes.
At minimum that includes:
paperclipai@paperclipai/server@paperclipai/ui- public packages under
packages/
2.1. In npm, open each package settings page
For each package:
- open npm as an owner of the package
- go to the package settings / publishing access area
- add a trusted publisher for the GitHub repository
paperclipai/paperclip
2.2. Add one trusted publisher entry per package
npm currently allows one trusted publisher configuration per package.
Configure:
- workflow:
.github/workflows/release.yml
Repository:
paperclipai/paperclip
Environment name:
- leave the npm trusted-publisher environment field blank
Why:
- the single
release.ymlworkflow handles both canary and stable publishing - GitHub environments
npm-canaryandnpm-stablestill enforce different approval rules on the GitHub side
2.2.1. Newly added public packages need a bootstrap phase
Trusted publishing is configured on the npm package itself, not at the repo scope. That means a brand-new public package must not be auto-enrolled into CI publishing until its npm package exists and its trusted publisher has been configured.
Repo policy:
- add every non-private package to
scripts/release-package-manifest.json - set
"publishFromCi": trueonly when CI is expected to publish that package - if the package is not ready for CI publishing yet, keep
"publishFromCi": false - complete the package bootstrap before merging any PR that changes a release-enabled new package
Bootstrap sequence for a new package:
- publish the package once from a trusted maintainer machine using normal npm auth
- open that package on npm and add the
paperclipai/papercliptrusted publisher for.github/workflows/release.yml - rerun or dry-run the release flow as needed to confirm CI publishing now works
- only then enable
"publishFromCi": true
PR CI enforces this by checking changed release-enabled package manifests against npm. That keeps master canary publishing healthy while preserving the no-long-lived-token model for normal CI releases.
2.3. Verify trusted publishing before removing old auth
After the workflows are live:
- run a canary publish
- confirm npm publish succeeds without any
NPM_TOKEN - run a stable dry-run
- run one real stable publish
Only after that should you remove old token-based access.
3. Remove Legacy npm Tokens
After trusted publishing works:
- revoke any repository or organization
NPM_TOKENsecrets used for publish - revoke any personal automation token that used to publish Paperclip
- if npm offers a package-level setting to restrict publishing to trusted publishers, enable it
Goal:
- no long-lived npm publishing token should remain in GitHub Actions
4. Create GitHub Environments
Create three environments in the GitHub repository:
npm-canarynpm-betanpm-stable
Path:
- GitHub repository
SettingsEnvironmentsNew environment
5. Configure npm-canary
Recommended settings for npm-canary:
- environment name:
npm-canary - required reviewers: none
- wait timer: none
- deployment branches and tags:
- selected branches only
- allow
master
Reasoning:
- every push to
mastershould be able to publish a canary automatically - no human approval should be required for canaries
The scheduled nightly lane also publishes under npm-canary: it is the same
trust level (fully automated, no human gate), its runs execute on master so
the branch rule is satisfied, and reusing the environment means the nightly
lane required no new environments and no npm trusted-publisher changes
(publishing still happens from release.yml, see section 2.2).
5.1. Configure npm-beta
Recommended settings for npm-beta:
- environment name:
npm-beta - required reviewers: at least one maintainer
- prevent self-review: enabled when your team size allows it
- wait timer: none
- deployment branches and tags:
- selected branches only
- allow
master
Reasoning:
- beta promotions are deliberate human decisions; the required reviewer on this environment is the promotion gate
- create this environment before the first
channel: betadispatch. If the workflow runs first, GitHub auto-creates the environment with no protection rules, and that first beta would publish without approval
Like nightly, beta publishing lives in release.yml, so no npm
trusted-publisher changes are needed (see section 2.2).
6. Configure npm-stable
Recommended settings for npm-stable:
- environment name:
npm-stable - required reviewers: at least one maintainer other than the person triggering the workflow when possible
- prevent self-review: enabled
- admin bypass: disabled if your team can tolerate it
- wait timer: optional
- deployment branches and tags:
- selected branches only
- allow
master
Reasoning:
- stable publishes should require an explicit human approval gate
- the workflow is manual, but the environment should still be the real control point
7. Protect master
Open the branch protection settings for master.
Recommended rules:
- require pull requests before merging
- require status checks to pass before merging
- require review from code owners
- dismiss stale approvals when new commits are pushed
- restrict who can push directly to
master
At minimum, make sure workflow and release script changes cannot land without review.
8. Enforce CODEOWNERS Review
This repo now includes .github/CODEOWNERS, but GitHub only enforces it if branch protection requires code owner reviews.
In branch protection for master, enable:
Require review from Code Owners
Then verify the owner entries are correct for your actual maintainer set.
Current file:
.github/CODEOWNERS
If @cryppadotta is not the right reviewer identity in the public repo, change it before enabling enforcement.
9. Protect Release Infrastructure Specifically
These files should always trigger code owner review:
.github/workflows/release.ymlscripts/release.shscripts/release-lib.shscripts/release-package-map.mjsscripts/create-github-release.shscripts/rollback-latest.shdoc/RELEASING.mddoc/PUBLISHING.md
If you want stronger controls, add a repository ruleset that explicitly blocks direct pushes to:
.github/workflows/**scripts/release*
10. Do Not Store a Claude Token in GitHub Actions
Do not add a personal Claude or Anthropic token for automatic changelog generation.
Recommended policy:
- stable changelog generation happens locally from a trusted maintainer machine
- canaries never generate changelogs
This keeps LLM spending intentional and avoids a high-value token sitting in Actions.
11. Verify the Canary Workflow
After setup:
- merge a harmless commit to
master - open the
Releaseworkflow run triggered by that push - confirm it passes verification
- confirm publish succeeds under the
npm-canaryenvironment - confirm npm now shows a new
canaryrelease - confirm a git tag named
canary/vYYYY.MDD.P-canary.Nwas pushed
Install-path check:
npm install --prefix "$(mktemp -d)" paperclipai@canary --no-audit --no-fund
The release script runs this clean-prefix install after publishing every workspace
package dependency-first and publishing paperclipai last. A package that is not
yet registry-visible stops the train before the channel entrypoint can advance.
12. Verify the Stable Workflow
After at least one good canary exists:
- resolve the target stable version with
./scripts/release.sh stable --date YYYY-MM-DD --print-version - prepare
releases/vYYYY.MDD.P.mdon the source commit you want to promote - open
Actions->Release - run it with:
source_ref: the tested commit SHA or canary tag source commitstable_date: leave blank or set the intended UTC date like2026-03-18do not enter a version like2026.318.0; the workflow computes that from the datedry_run:true
- confirm the dry-run succeeds
- rerun with
dry_run: false - approve the
npm-stableenvironment when prompted - confirm npm
latestpoints to the new stable version - confirm git tag
vYYYY.MDD.Pexists - confirm the GitHub Release was created
Implementation note:
- the GitHub Actions stable workflow calls
create-github-release.shwithPUBLISH_REMOTE=origin - local maintainer usage can still pass
PUBLISH_REMOTE=public-ghexplicitly when needed
13. Suggested Maintainer Policy
Use this policy going forward:
- canaries are automatic and cheap
- stables are manual and approved
- only stables get public notes and announcements
- release notes are committed before stable publish
- rollback uses
npm dist-tag, not unpublish
14. Troubleshooting
Trusted publishing fails with an auth error
Check:
- the workflow filename on GitHub exactly matches the filename configured in npm
- the package has the trusted publisher entry for the correct repository
- the job has
id-token: write - the job is running from the expected repository, not a fork
Stable workflow runs but never asks for approval
Check:
- the
publishjob uses environmentnpm-stable - the environment actually has required reviewers configured
- the workflow is running in the canonical repository, not a fork
CODEOWNERS does not trigger
Check:
.github/CODEOWNERSis on the default branch- branch protection on
masterrequires code owner review - the owner identities in the file are valid reviewers with repository access
Related Docs
Runner verification dependency cache
release-verify.yml runs Verify Paperclip Runner on two independent runners.
The protocol lane runs check:eval-kernel and check:protocol. The Rust lane
runs check:runner and check:api-authority. Together they retain every check
in check:all; both lanes must pass before Cloud source verification or
readiness can succeed. A failed lane does not cancel the other lane.
Both lanes restore Cargo dependencies with the pinned Rust Cache action. The
compiler comes from the Runner package's rust-toolchain.toml before the action
computes its key. Compiler and Cargo metadata changes select a new cache. The
existing release-runner-v1 shared key avoids separate copies for these lanes.
Only the Rust lane saves this cache. After verification it also runs build:rust
to warm the debug dependencies used by the protocol lane; its own tests already
warm release dependencies. The cache writer is shorter than the protocol lane.
Workspace crates and installed Cargo binaries are excluded. Every run rebuilds workspace code and runs all assigned checks, including on a cache hit. Only an own-repository master-push run verifying that push's exact SHA can restore the cache, and only a successful Rust lane saves it. PR, tag, and manual candidate verification compile without this cache. A miss or eviction costs compilation time but does not change the checks. To discard old dependency caches, increment the shared-key version and let the next successful master verification warm it.
The trust boundary is the protected master branch, not the cache-key text. GitHub does not let master restore caches created by a child branch, sibling branch, tag, or PR merge ref. Both permitted restore scopes (current branch and default branch) are master here. A workflow with authority to execute arbitrary code on master can affect verification directly and is already trusted. The cache contains dependency build artifacts, not credentials or workspace output. See GitHub cache access restrictions.
Chat integration test shards
Release verification runs the large chat integration file on three independent
runners. Five other server shards cover every remaining general server file.
The ordinary local test command and trusted PR workflow keep their complete
general-server group. Each chat case shuts down its services, pauses its own
still-active endpoints, and retires its active/waiting conversations after
assertions. This keeps workers in later cases from claiming earlier
fixtures in the shared test database. Application assertions stay unchanged.
Each chat job collects active tests with Vitest, groups cases by source line, and balances those groups by case count. Parameterized cases and loop-generated cases on one line stay together. The job re-collects with the exact line filters it will execute and fails if the selected case identities differ. Hooks and test execution remain sequential inside each runner with its own temporary home.
Run one shard locally with:
pnpm test:run:general -- --group general-chat --shard-index 0 --shard-count 3
Use indexes 0, 1, and 2 to run the complete chat suite. The CLI validates that each shard has work and that collection includes usable source locations. A Vitest collection or filtering change fails verification instead of dropping tests. Splitting adds three release-verification jobs and repeats collection and fixture setup; it does not make a single test faster.
The file-duration manifest also records the native Codex Runner integration suite's measured import and execution cost, so the existing file balancer accounts for it in both ordinary PR and release verification.