mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-10-22 16:51:28 +02:00
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - The first task helps a new user define and approve useful work. > - That workflow needs reusable instructions and tests against the production experience. > - Native Codex and Claude must load the assigned skill, including after resume. > - Maintainers need recorded conversations and precise failed checks to judge regressions. > - This pull request adds the first-task skill and a suite in the shared Runner E2E harness. > - It keeps behavior results separate from informational quality scores and incomplete recordings. ## Linked Issues or Issue Description **What existing behavior does this improve?** The first onboarding task and the Runner E2E report used to review it. **Current behavior** Onboarding embeds its policy in a hidden brief. Native Codex drops the skill-instructions setting at the Rust boundary. The shared E2E harness has no onboarding suite or full conversation view. **Proposed behavior** Assign and invoke `/first-task` for the onboarding task. Send selected Codex skills as structured protocol inputs. Run twelve scenarios across legacy Codex, legacy Claude, native Codex, and native ACPX Claude. Include all 48 cells in full campaigns. Show recorded chat, question and approval cards, exact checks, instructions, and billing in the shared dashboard. **Reason and benefit** Measure the real onboarding experience before changing prompts. Distinguish infrastructure failures, behavior failures, and unexercised journey steps. **Breaking changes** No database migration or production API change. First-task instructions now live in an assigned skill. The user-edited persona is preserved; the skill includes the maintainer-approved proposal-mode mapping and saved-plan requirement. Related: #11043 is earlier onboarding work. #13422 already fixes native Claude model pinning, context delivery, and read permissions on master; this branch includes those fixes through its base. The new Claude recovery test supplements them. ## What Changed - Extract and assign the first-task skill while retaining the production greeting and opening question. - Carry the Codex skill-instructions flag through thread start and resume. Resolve explicit task skill references only against assigned skills and send native skill inputs. - Invoke an unambiguously selected assigned skill through Claude ACPX’s native slash-command parser on initial and resumed turns, retaining the entire task/wake envelope as its argument. Do not carry that invocation into ordinary tasks. - Restore the saved single-task proposal modes: confirmation card, or saved plan with revision-targeted checkbox approval. Explicit plan requests also require a saved plan. - Add first-response and complete-journey cases with fixed user facts, acceptance checkpoints, durable outcome checks, and accounting for child runs. - Fail the eval when choice questions have fewer than two real options. Recognize planning documents without treating them as completed work. - Add optional, bounded quality judging as explicit post-processing. - Render full conversations and static interaction cards in the shared report. Conversations start folded. Show original and regraded results and incomplete journeys distinctly. - Keep credential-persistence scanning outside the first-task behavioral suite; retain public evidence redaction. - Refresh generated capability references after the API-reference edits. - Correct shared native question guidance and tool schemas: choices need at least two meaningful options; open-ended questions use canonical text fields with the required compatibility payload. Verify both formats through real tool-authority persistence. - Disable announcements automatically for every isolated Runner E2E process and label the gallery environment/provider/target explicitly. - Remove CI races in the GitHub connection browser test and native session recovery test by waiting for the actual async work before asserting its results. ## Verification - `pnpm exec vitest run server/src/services/onboarding-first-task-assets.test.ts server/src/__tests__/issue-onboarding-first-task-routes.test.ts`: 19 passed. - `pnpm --dir packages/paperclip-runner exec vitest run src/drivers/acpx/runtime-host.test.ts src/drivers/acpx/native-skill-prompt.test.ts src/cli/acpx-runtime-sidecar.test.ts`: 70 passed. Native command forwarding and the 1 MiB input boundary both failed before their fixes and passed afterward. Coverage includes changed skills on reopen, approval context, and an ordinary subsequent task. - Runner E2E unit suite: 306 passed. Harness typecheck passed. The 64 first-task fixture and grader tests also pass. - Full repository typecheck and build passed locally. Server typecheck and Runner build passed again after the native-command change. - Full GitHub Actions CI passed on `23e56447b`: all server/workspace/browser shards, Runner verification, typecheck/release registry, build, canary, policy, and Docker checks. Greptile reviewed this exact head at 5/5 with no unresolved threads. The earlier broad local run had database startup/timing failures that passed isolated retries; the complete remote suite is green. - Merge verification against current master: 312 harness tests and 13 native recovery tests passed. Regenerated semantic contracts and fixture hashes pass their consistency check. Full local typecheck and build also passed on the stacked queue branch. After merging the latest master and preserving the GitHub setup timing regression in the split browser suite, both focused GitHub browser tests passed. Three CI timing/startup flakes passed local verification and one remote retry; all latest-head checks are green. - Real pinned Claude SDK and Claude ACP JSON-RPC probes against a local mock API confirmed that `/skill-name` expands the assigned skill body before the model request and retains the task arguments. A prose mention does not. The probes made no paid model calls. The ACP probe used the current first-task skill body and retained the wake arguments. - [Full 48-case campaign and report](https://pages.paperclip.ing/runner-e2e-first-task-35053063880/): 44 passed after three interrupted Codex cases completed in targeted reruns. Original results, regrades, and all 51 executions remain in the report provenance. - [Claude campaign after the shared-question fix](https://pages.paperclip.ing/runner-e2e-first-task-claude-35099525201/): 10/12 passed with zero single-option failures. All 12 recorded the current assigned skill and corrected guidance. The failures exposed skipped skill invocation and a missing saved plan. This PR adds native command invocation and explicit saved-plan instructions; the subsequent report below still shows behavior failures. - [Fresh 12-case Claude report](https://pages.paperclip.ing/runner-e2e-first-task-claude-35102737804/) at `78452129e`: 10/12 pass after correcting two false proposal-matcher failures. The recordings said “Here is the task I will create and run/complete” in approval cards; the old matcher missed that word order. Regression tests failed before the fix and pass after it. Original results and offline regrade provenance remain linked. No agent rerun was needed. Zero single-option-question failures; two behavior failures remain: direct work before acceptance on a plain first message, and an explicit plan request without a saved plan. Neither check was relaxed. The follow-up `82087ac7e` fixes command-prefix size accounting; `94aefb1f3` fixes only that proposal matcher. - Report browser checks confirm folded conversations, rendered cards, explicit Local/Daytona labels, and no page errors. The published-object audit scanned 1,306 text files across 2,154 objects with no credential-format findings or prohibited files. Image pixels and unknown token formats are outside that scan. ## Risks - Model behavior is nondeterministic. One campaign is evidence, not a guarantee. The two remaining Claude behavior failures are visible in the report and require further product work; this PR does not claim all onboarding scenarios pass. - The suite checks persisted Paperclip effects. It cannot prove the absence of arbitrary external effects. - Historical recordings can miss later journey steps. These remain incomplete, never passes. - Native profiles switch runtime after the production onboarding wizard because it does not yet expose a native option. - Quality scores are informational and cannot override behavioral failures. ## Model Used OpenAI Codex, GPT-6, with reasoning, repository tools, and code execution. The exact deployed model identifier and context-window size are not exposed in this session. ## 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>
242 lines
13 KiB
Markdown
242 lines
13 KiB
Markdown
# Runner E2E security for a public repository
|
|
|
|
This suite can spend provider money, expose four API credentials to isolated
|
|
test processes, publish a container, retain private visual evidence, and write
|
|
public structured evidence. Treat changes to the workflow, harness, fixture
|
|
prompts, evidence packager, and publisher as security-sensitive production
|
|
changes.
|
|
|
|
## GitHub authorization
|
|
|
|
Set `RUNNER_E2E_ALLOWED_ACTOR_IDS` to a non-empty JSON array of numeric GitHub
|
|
user IDs. Keep the list equal to the owners of `.github/**` in
|
|
`.github/CODEOWNERS`. For example, use `[123456,789012]`. Resolve each ID from
|
|
the authenticated CLI and verify the login before adding it:
|
|
|
|
```bash
|
|
gh api users/LOGIN --jq '{login,id}'
|
|
```
|
|
|
|
The paid workflows reject manual dispatches when the workflow definition does
|
|
not come from the default branch. A trusted dispatcher may name any branch in
|
|
`paperclipai/paperclip` as the code under test. The authorization job resolves
|
|
that branch through the GitHub API and passes only its immutable commit SHA to a
|
|
credential-free target-lock job. That job checks out the commit, regenerates
|
|
`pnpm-lock.yaml` once with lifecycle scripts disabled and lockfile-only mode,
|
|
then uploads the file under a run-attempt-scoped artifact ID. Catalog, image,
|
|
shared-build, provider-pack, and paid test jobs download that exact artifact by
|
|
ID, verify its recorded SHA-256, and restore it before setup or a frozen
|
|
dependency install. The lock resolver receives no provider credentials and
|
|
must never run repository lifecycle scripts. The shared-build and provider-pack
|
|
jobs also receive no provider credentials and disable dependency lifecycle
|
|
scripts; they package outputs with SHA-256 sidecars that consumers verify
|
|
before extraction. The paid test job installs with lifecycle scripts disabled,
|
|
and materializes the exact pinned OpenCode executable from its lockfile-verified
|
|
optional package without invoking package lifecycle code. Provider secrets are
|
|
scoped only to the final test step rather than dependency setup. Report sanitization and AWS
|
|
history publication explicitly use the trusted workflow commit and do not
|
|
consume the target lockfile. Never run the workflow definition from the target
|
|
branch.
|
|
|
|
The workflows verify both the original actor and triggering actor for every
|
|
scheduled or manual attempt, including human reruns. Every
|
|
secret-bearing job repeats this check as its first step so GitHub's partial-job
|
|
rerun feature cannot bypass a successful predecessor authorization job. The
|
|
legacy manually dispatched E2E workflow uses the same gate. Numeric IDs are
|
|
stable across username changes and prevent lookalike-name authorization.
|
|
|
|
The full-stack and live campaigns have one Sunday UTC schedule each and also
|
|
support explicit manual dispatch. Their legacy-named nightly repository
|
|
variables remain independent kill switches. Neither paid workflow accepts
|
|
pull-request, push, workflow-run, or reusable-workflow triggers.
|
|
|
|
Protect the default branch, require review for workflow/harness paths, restrict
|
|
workflow dispatch permission, and restrict repository variable/environment
|
|
administration to the same trusted maintainers. Configure the organization to
|
|
allow only approved GitHub Actions. A malicious change merged into the default
|
|
branch executes with the same authority as the suite.
|
|
|
|
Every external action in the paid workflow is pinned to a full commit SHA. Keep
|
|
the adjacent major-version comment for update tooling, and resolve and review a
|
|
new immutable SHA before upgrading an action. The credential-free security test
|
|
rejects mutable tag or branch references.
|
|
|
|
## Secrets and protected environments
|
|
|
|
Create `runner-e2e-paid`, restrict deployments to the default branch, and put
|
|
only `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, `OPENROUTER_API_KEY`, and
|
|
`DAYTONA_API_KEY` in it. Do not duplicate these credentials as repository- or
|
|
organization-level Actions secrets: environment scoping is the boundary that
|
|
prevents branch or pull-request jobs from requesting them. Require approval
|
|
from an account in `RUNNER_E2E_ALLOWED_ACTOR_IDS` for this environment and
|
|
disable administrator bypass. The authorize, target-lock, catalog, image,
|
|
report, history, and Pages jobs receive none of these secrets.
|
|
Each full-stack matrix cell receives only its selected profile credential, plus
|
|
Daytona only for Daytona cells. Secret-bearing and OIDC jobs use frozen installs
|
|
without a shared dependency cache.
|
|
On disposable GitHub Linux runners with Ubuntu's unprivileged-user-namespace
|
|
restriction, the authorized default-branch workflow provisions an AppArmor profile before provider credentials are exposed. The profile is attached to the exact
|
|
lockfile-pinned Codex executable. It grants `userns` so Codex can construct its
|
|
filesystem sandbox; it does not disable the kernel restriction or Codex's
|
|
workspace policy. Setup fails before invoking a model if the noninteractive
|
|
profile load fails. Target-controlled tests only probe the existing sandbox and never invoke sudo or load host policy. This host-only profile disappears with the ephemeral runner.
|
|
See [Ubuntu's namespace restriction documentation](https://documentation.ubuntu.com/security/security-features/privilege-restriction/apparmor/).
|
|
Local developer machines are never modified by this setup. Legacy Codex fixtures
|
|
disable optional shell-environment snapshots to avoid persisting credentials;
|
|
other suites retain the persisted-state scanner. The `first-task` suite omits
|
|
private home/workspace credential-persistence scanning so its evaluation focuses
|
|
on onboarding behavior. Artifact redaction and publication scanning remain in
|
|
force for every suite.
|
|
The Paperclip server process also receives none; the browser posts each value
|
|
once to the encrypted company secret API and agents/environments retain only
|
|
secret references.
|
|
|
|
Create `runner-e2e-history`, also default-branch-only, for the OIDC publishing
|
|
job. It contains no long-lived AWS key. Required reviewers may be added when a
|
|
human approval on every nightly publication is acceptable; otherwise rely on
|
|
the actor gate, environment branch restriction, and protected default branch.
|
|
|
|
## Runner fleet isolation
|
|
|
|
When `RUNNER_E2E_AWS_ENABLED=true`, paid matrix cells use the exact RunsOn fleet
|
|
selector `runs-on/fleet=paperclip-public-pr-x64/env=public-ci`, matching the AWS
|
|
fleet selected by `pr-trusted.yml` only after its stable numeric-ID trust gate.
|
|
Any other or missing toggle value falls back to the GitHub-hosted
|
|
`ubuntu-latest` runner and its lower concurrency ceiling. The workflow chooses
|
|
between those two reviewed literal labels; it never evaluates a configured
|
|
runner label.
|
|
|
|
Keep both runner targets restricted to `paperclipai/paperclip` and workflows
|
|
that independently authorize trusted source revisions. Never let a fork or
|
|
untrusted pull-request workflow target them. The RunsOn fleet must launch a
|
|
fresh ephemeral instance for every job, prohibit persistent runner reuse, and
|
|
disable interactive SSH/debug access unless a separate incident procedure
|
|
explicitly authorizes it.
|
|
|
|
Changing the runner does not widen who can authorize secret access. The paid
|
|
workflow still has only schedule and manual triggers, requires its trusted
|
|
definition to come from the protected default branch, requires allowlisted
|
|
stable actor IDs before checkout, and repeats that authorization as the first
|
|
matrix step. Provider credentials come only from the protected
|
|
`runner-e2e-paid` environment. The fleet selector is an exact workflow literal;
|
|
the only repository-controlled routing input is its boolean rollout switch, so
|
|
configuration cannot redirect a secret-bearing job to an arbitrary runner.
|
|
|
|
The optional target branch is code, not workflow authority. A CODEOWNER who
|
|
dispatches a target branch explicitly authorizes that branch's selected test
|
|
process to receive the cell's scoped provider credential. The workflow resolves
|
|
the target only inside the same repository, pins one SHA for the campaign, and
|
|
checks it out only after authorization. Target-controlled code cannot replace
|
|
the report sanitizer or the AWS history publisher. Fork refs and
|
|
target-controlled workflow definitions do not enter this path.
|
|
|
|
## AWS OIDC and S3
|
|
|
|
The AWS role trust policy should accept only GitHub's OIDC audience and the
|
|
publishing environment subject:
|
|
|
|
```json
|
|
{
|
|
"Version": "2012-10-17",
|
|
"Statement": [
|
|
{
|
|
"Effect": "Allow",
|
|
"Principal": {
|
|
"Federated": "arn:aws:iam::ACCOUNT_ID:oidc-provider/token.actions.githubusercontent.com"
|
|
},
|
|
"Action": "sts:AssumeRoleWithWebIdentity",
|
|
"Condition": {
|
|
"StringEquals": {
|
|
"token.actions.githubusercontent.com:aud": "sts.amazonaws.com",
|
|
"token.actions.githubusercontent.com:sub": "repo:paperclipai/paperclip:environment:runner-e2e-history"
|
|
}
|
|
}
|
|
}
|
|
]
|
|
}
|
|
```
|
|
|
|
Grant only List on the bucket prefix and Get/Put on its objects. Do not grant
|
|
Delete, ACL, bucket-policy, or wildcard-resource permissions:
|
|
|
|
```json
|
|
{
|
|
"Version": "2012-10-17",
|
|
"Statement": [
|
|
{
|
|
"Effect": "Allow",
|
|
"Action": "s3:ListBucket",
|
|
"Resource": "arn:aws:s3:::BUCKET",
|
|
"Condition": {
|
|
"StringLike": { "s3:prefix": ["runner-e2e", "runner-e2e/*"] }
|
|
}
|
|
},
|
|
{
|
|
"Effect": "Allow",
|
|
"Action": ["s3:GetObject", "s3:PutObject"],
|
|
"Resource": "arn:aws:s3:::BUCKET/runner-e2e/*"
|
|
}
|
|
]
|
|
}
|
|
```
|
|
|
|
Enable S3 versioning, default encryption, and Block Public Access. Disable
|
|
object ACLs. CloudFront receives read-only access through Origin Access Control;
|
|
the bucket itself stays private. Log S3 data writes and alert on attempts to
|
|
write outside the prefix or assume the role with a different subject.
|
|
|
|
Campaign prefixes are content-digested and immutable. The publisher refuses a
|
|
different digest at an existing campaign key. Only the compact history and
|
|
latest pointers are mutable, and S3 versioning makes those updates recoverable.
|
|
|
|
## Public evidence boundary
|
|
|
|
CloudFront and GitHub Pages are public. Fixture identifiers, timing, token
|
|
usage, costs, normalized results, allowlisted inert structured per-attempt
|
|
evidence, and trusted runner PNG screenshots are expected public data. Each
|
|
public screenshot must carry the explicit `public-runner-fixture` marker in
|
|
the normalized result. This includes a `failure.png` capture. Screenshot paths
|
|
must be safe PNG basenames and must be tied to the exact normalized execution
|
|
ID and attempt. The runner capture helper accepts only the exact issue route
|
|
for the live fixture that the harness created. Other issue routes, credential
|
|
pages, setup pages, and administration pages fail closed. The
|
|
CloudFront-backed S3 history also publishes one
|
|
synthetic campaign-summary PNG generated by trusted publisher code solely from
|
|
fixed catalog labels and sanitized numeric/status fields. Video, archives,
|
|
generated Playwright/blob/HTML report trees, SVG or other active content,
|
|
credentials, Paperclip homes, databases, workspaces, master keys,
|
|
raw/unredacted logs, unmarked images, and unallowlisted files are not public.
|
|
Allowlisted `.log` copies must pass the existing exact-value/key-shape scan and
|
|
redaction boundary.
|
|
|
|
The packaged evidence uploaded as a 30-day GitHub Actions artifact has a
|
|
different, broader boundary. Text is exact-value and key-shape scanned and
|
|
redacted. PNG and WebM are raw-byte scanned; SVG is rejected during packaging
|
|
because it is active content. Raster pixels cannot be exhaustively
|
|
secret-scanned by bytes, so fixture authors must treat every marked capture as
|
|
public and must never extend the allowed task route to credentials, secrets,
|
|
private user data, or other non-public content. Adding or changing a marked
|
|
capture requires review of the visible page state. Videos remain
|
|
access-controlled.
|
|
|
|
Before permanent publication, the campaign publisher creates a separate S3
|
|
stage and retains only allowlisted `.json`, `.log`, `.md`, and `.txt` evidence,
|
|
result PNGs with the explicit `public-runner-fixture` marker.
|
|
It then launches publisher-only Chromium with networking blocked to render one
|
|
`public-images/campaign-summary.png`. That fixed-path PNG is capped at 12 MiB
|
|
and its signature is validated. Per-attempt XML is excluded because browsers
|
|
can process XML/XSLT;
|
|
the only public XML is the root `junit.xml`, which the report aggregator
|
|
constructs from fixed markup and XML-escaped fields. Videos, archives,
|
|
raw/unallowlisted logs, SVG, undeclared images, generated reports, and symlinks
|
|
fail closed or are removed before the immutable manifest is calculated.
|
|
|
|
GitHub Pages is built from a second stage without the synthetic summary PNG but
|
|
with the same trusted-fixture screenshot allowlist. A leak detected by the
|
|
existing packager scan fails the cell and withholds the unsafe file.
|
|
|
|
Rotate the affected credential immediately if a secret-scanning failure or
|
|
unexpected public object is observed. Preserve the access-controlled Actions
|
|
artifact and S3 object versions for incident analysis; do not weaken scanning
|
|
to make a campaign publish.
|