PaperClipAI/tests/runner-e2e/EVERYDAY-WORKFLOWS.md

Everyday Paperclip workflow evals

Main-server outage qualification

The remote native Codex and Claude recover-controller cells default to SIGKILL, an observed 125-second offline interval and retirement of the old controller cache. They use natural ownership timing. The wrapper emits controller-outage.json; the grader requires distinct server processes, confirmed HTTP loss, cache retirement and the requested interval, in addition to same-run/provider-turn and finalization proof. Partial fault receipts remain in the attempt's private control/ directory on failure. A successful task without the fault receipt cannot qualify.

Use PAPERCLIP_RUNNER_E2E_CONTROLLER_STOP_SIGNAL=SIGTERM for graceful shutdown, PAPERCLIP_RUNNER_E2E_CONTROLLER_OUTAGE_MS=1000 for an explicit one-second experiment, or PAPERCLIP_RUNNER_E2E_CONTROLLER_OUTAGE_MS=300000 for the five-minute contract. These settings are recorded; compare only matching configurations. The configured duration is a minimum hold after confirmed HTTP loss. Shutdown and startup add to total recovery time; they do not extend the runner's seven-minute grace or the workflow's attempt deadline. recover-controller permits no automatic retries.

PAPERCLIP_RUNNER_E2E_CONTROLLER_OUTAGE_MS=300000 pnpm test:e2e:runner -- --id everyday-workflows.runner-codex.daytona.recover-controller --max-automatic-retries 0

The credential-free wrapper smoke starts the actual main server and a separate disposable Postgres, kills/replaces the server, checks that Postgres never restarted, reads existing data, accepts a new HTTP mutation, and confirms cleanup:

node cli/node_modules/tsx/dist/cli.mjs tests/runner-e2e/controller-replacement-smoke.ts

It prints a private JSON evidence path. It uses no model and does not qualify active-agent continuity, queued launches, or an A-to-B version upgrade. See the qualification plan for the separate gates and outstanding work. Hosted qualification still requires a valid pinned Daytona image and the existing credential/budget admission.

Workflow suite

This manual suite tests useful work through the production browser, public API, native runner, and normal agent instructions. It complements the tightly scripted runner contract fixtures. It does not add a scheduled or default paid run: --all and generic profile selectors exclude it. Select the suite or an exact execution ID explicitly.

Stories and assertions

Story	Cases	Required evidence
Build a small project and revise it	build-revise	Download both ZIPs through the UI; independently execute the delivered CLI and import its function; test the revision; retrieve the original bytes again.
Delegate and incorporate late feedback	delegate-feedback	One child assigned to Riley; send feedback while the child runs; find it in the child history and independently test --max-length in the delivered ZIP. The worker must not execute on the parent.
Hire a teammate and use them again	hire-reuse	One Morgan QA reporting to the lead, native runner and the same encrypted connection bindings, real child execution, then a second usable delivery from that same agent.
Decide on an installed service action	service-approve, service-decline	Assign an authenticated local MCP fixture with Ask first; match its tool action and connection ID; no provider call before approval; exactly one after approval and a verified document; none after decline.
Decline a new connection	connection-decline	Start without service connections; match a Notion connection intent; click Not now; verify the saved rejection, no new connection or repeated request, and an explanation followed by Done.
Request an email address	agentmail-setup	Enable Chat connectors, ask for an email address, and require a durable AgentMail card for the requesting agent and user. Reload and verify one password field, the direct API-key link, and no access selectors or modal. Decline and verify no connection or repeated request.
Continue work after a controller restart	recover-controller	Observe saved source, persist a user message, restart the isolated controller, and independently test the delivered result. Native Daytona cases keep the controller offline for 125 seconds before restarting.
Stop work and change direction	stop-redirect	Click Stop, send one new request, reload, observe exactly one stored user message and the new answer, and reach Done.
Create and edit a company skill	create-skill-studio	Create one skill through the runner, verify its persisted library entry and activity-feed card, open Skill Studio, save an edit, and verify the edit after returning. Local Codex, local ACPX Claude, and warm Daytona cells are explicit.

For normal completion, all story tasks must reach Done, with no active run, pending completion confirmation, or scheduled recovery. Runs must prove native identity and native terminal contracts. A workspace-contention cancellation is not provider execution only when the persisted pre-dispatch record explicitly says providerWorkStarted: false and no process/session/runner identity exists. Other unexplained cancellations remain failures. Twelve total run records bound each story, including contention and recovery.

Arbitrary runner-process termination is not part of the model scorecard. The historical recover-runner, recover-runner-safe, and recover-runner-uncertain attempts remain available as diagnostics, with their original grades and costs. The first two did not establish a safe restart boundary, and the uncertainty case measures a deterministic safety rule. None supports ranking models. See controlled recovery tests.

Staging-soak reliability group

Seven explicit-only cases turn the staging soak's ad-hoc browser scripts into repeatable runs. Each case comes from a real staging failure. They live in the everyday-workflows suite, but neither --all, --suite everyday-workflows, a profile/environment selector, nor a scheduled campaign selects them. Only the reliability group, an explicit --case, or an exact --id does (explicitGroup on the task fixture enforces this in selectors.ts).

pnpm test:e2e:runner -- --list --suite everyday-workflows --group reliability
pnpm test:e2e:runner -- --suite everyday-workflows --group reliability --max-parallel 2
pnpm test:e2e:runner -- --suite everyday-workflows --case queue-steer
pnpm test:e2e:runner -- --id everyday-workflows.runner-codex.daytona.interrupt-redirect

Cadence: run the whole group before a release and after a risky runner, recovery, queue, tree-control or restart change. Do not run it per PR. It is paid: ten Daytona cells, single attempt each, with a 1,000-cent company and lead-agent hard stop. Missing provider or Daytona credentials, or a mutable Daytona image, stop the launch before any provider spend.

Product rule (every case): work interrupted by a restart, Stop, reassignment or lost runner must resume automatically, or offer a Retry that works on first use. The harness watches each task's executionBlocker, interruption and execution projection while it waits. It uses an offered Retry at most once (never while an automatic retry is scheduled) and requires a new run from it. These fail: a "send a message to continue" blocker (resume.no-send-message-blocker), a dead or repeated Retry (resume.retry-works-first-use), a silent stall past the case bound (resume.no-silent-stall), and duplicate execution. Evidence is saved to reliability-observations.json.

Message delivery is attributed through each run's wake context (wakeCommentIds), not its saved prompt: the prompt carries the whole conversation history, so later turns legitimately repeat earlier messages. A message delivered by more than one successful run is duplicate execution. --group slow selects only long-silent-command.

Case	Soak signature	Journey	Key checks
interrupt-redirect	S47 (fixes #19370, #19355)	While a slow step loop runs, send a follow-up and click its queued Interrupt. Then reopen with a second slow loop, click Stop, and immediately send another message while the stopped run is still saving (the workspace_busy retry path).	Interrupted run ends operator_interrupted; exactly one continuation carries the message and starts within 180 s (the soak saw 13–90 s once fixed, 4 minutes or never before); no second user message; the Stop-settlement message is consumed once within 180 s; Done.
reassign-midrun	S48 (fix #19354)	While the lead runs a resumable step loop, reassign the task to Riley through the public issue API.	Request is 2xx within 10 s (no 409 or 20 s hang); old run ends issue_reassigned; Riley's run starts after the request and succeeds; no overlapping executions on the task; steps.txt holds step-1..step-8 exactly once, so work continued from saved files.
three-turn-revise	S49	Create items.md, then two reopen-comment revisions on the same workspace: add two items, then convert to a table. Runs on the per-turn Daytona lease (the bug hit gpt-5.6-sol with an ephemeral lease on turn 3).	Each turn succeeds; no reconciliation or send-a-message blocker; the file is a five-row Number/Item table.
long-silent-command	S52/S52b (fixes #19407, #19419)	The agent runs one 13-minute foreground loop (sleep 60 × 13, appending to silent.txt) that prints nothing, then reports. Tagged slow; 50-minute Daytona deadline.	Run succeeds after the command (not shortened) and is terminal within 8 minutes after it; silent.txt holds tick-1..tick-13; Stop stays available in API and browser samples throughout; any failure carries a named cause.
tree-pause-release	S50/S50b (fixes #19394, #19421)	The lead delegates two slow children to Riley. While a child is queued or has a parked workspace_busy retry, pause the parent's tree with POST /api/issues/:id/tree-holds. Hold past the retry's due time (at least 75 s), then release without wakeAgents.	The hold interrupted work; no child run starts during the hold; each interrupted task gets exactly one issue_tree_resumed run; all tasks reach Done.
queue-steer	H2	During an active run, queue three follow-ups through the public comment API; edit one, discard one, and steer one (retrying only steering_temporarily_unavailable, for up to two minutes).	Edited text reaches the agent and the original never does; the discarded text never reaches the agent; the steered message is acknowledged by the active turn and never replayed; the remaining message is handled by exactly one run.
queued-across-restart	S55/U9 (fix #19431)	Queue a follow-up through the composer while a run is active, then restart the controller with the existing restart machinery.	The follow-up dispatches only after the restart, automatically within six minutes, exactly once; every observed queued wait names a cause (server_restarting, task_drain, …) rather than a bare "Queued"; any offered Retry works on first use.

Matrix (10 cells). Each case runs once on native Codex (runner-codex, gpt-5.6-sol) on Daytona: the warm environment, except three-turn-revise on the per-turn lease. long-silent-command also runs on the released ACPX facades Grok (runner-acpx-grok), Kimi (runner-acpx-kimi) and OpenRouter/Pi (runner-acpx-pi), which hung 3/3 before the fix. They run as ordinary released profiles, as in the rolling-deployment matrix; a profile whose ACPX qualification returns to pending drops out of the group automatically. Their credentials (XAI_API_KEY, KIMI_API_KEY, OPENROUTER_API_KEY) are only required when those cells are selected.

Coverage boundary. The graders and selection are calibrated by everyday-reliability.test.ts against passing and plausible failing observations. No paid cell has run yet; a first run of the group is the acceptance step. The original soak browser scripts are not in this repository, so these cases encode the reported signatures, not a replay of those scripts. long-silent-command checks the completion and Stop paths; it does not inject a runner connection loss, so the "fails within a bounded time with a named cause" contract is checked only if a loss happens naturally.

Matrix and running

The local matrix has fourteen cases on native Codex gpt-5.6-sol, native ACPX Claude claude-sonnet-5, and native Codex gpt-5.4-mini: 42 cells. The two core profiles also declare build/revise, delegation, controller-restart, and skill-creation cases on Daytona: eight cells. OpenCode adds only local hiring/reuse and delegation, for 52 cells total. Hiring/reuse and delegation have one attempt and a 1,000-cent company and lead-agent hard stop. Remote runner-process killing is not supported. For remote controller restart, a verified first download supplies the persistence checkpoint; the controller is interrupted during a subsequent revision with another queued requirement.

The native Codex and ACPX Daytona restart cells require a completed native tool receipt while that revision is still running. They keep the controller offline for 125 seconds after its health listener stops, then start its replacement. Before and after restart each pins the same run, normalized session, Runner identity, PID/start time, and active sandbox lease. The original receipt prefix must remain unchanged. A run that uses this native Codex remote cell also loses the old controller runtime cache between stopping and starting the server. The harness moves it into private evidence outside product lookup paths and requires an acknowledgement of that fault. Postgres and the instance encryption key survive. This checks deployment recovery without relying on the old controller journal or local warm-session cache. A run that ends before either active observation leaves the fault boundary unexercised; it cannot pass from the final ZIP alone. The cell no longer exempts an original Codex remote run that fails with process_lost and is replaced.

Completion requires one original provider turn and control-plane terminal, the same native receipt prefix, no failed native tool receipts, and consistent results for every stable completed tool identity. Reattached activity can redeliver an identical start or tool receipt; changed content or a second submission fails. It also requires a committed run and successful workspace_finalize receipt, and no pending workspace operation or recovery. Existing independent ZIP and preserved-download checks still apply. These receipt checks do not prove that arbitrary external effects with different tool IDs cannot repeat.

remote-restart-before.json, remote-restart-after.json and remote-restart-final.json retain receipt identities and hashes. They exclude provider text, tool arguments, command output, and raw provider sandbox IDs. The dedicated explicit-only daytona-restart-continuity definition 7 additionally replaces the controller after completion, then posts its unchanged saved-project follow-up through the normal authenticated task-comment API. Exactly one matching persisted user comment, consumed wake, same native/provider identity, new completed tool and useful output are required. Its initial browser journey remains; post-C readback is API-only and does not qualify browser reconnection or rendered status. The definition 6 browser timeout remains failed with an unknown browser cause. The suite definition version and harness digest change with this oracle. Its deterministic calibration is not a claim that a live Daytona run passed.

pnpm test:e2e:runner -- --id everyday-workflows.runner-codex.daytona.recover-controller --max-automatic-retries 0

pnpm test:e2e:runner -- --list --suite everyday-workflows
pnpm test:e2e:runner -- --suite everyday-workflows --environment local --max-parallel 2
pnpm test:e2e:runner -- --id everyday-workflows.runner-codex-mini.local.build-revise
pnpm test:e2e:runner -- --suite everyday-workflows --environment daytona --max-parallel 2

Before project stories or the Python calibration tests, start Docker on the harness host and fetch the pinned oracle image. CI prepares and verifies this same pinned image before the paid project-story cells; artifact checks run on the harness host. The workflow verifies the exact repository digest after the pull. This is required for local and Daytona stories.

docker pull python@sha256:9d2e5553305c7c7b0097999bb17187c69b921ccd6bc9d40e4bb5ebe652c00285
python3 tests/runner-e2e/everyday-artifact.py --preflight

The harness checks this prerequisite before it creates the task. It does not pull an image during a model attempt or fall back to host execution.

Use the credential and immutable Daytona image setup in README.md. Provider calls cost money. Each cell owns an isolated instance and project. There are no real third-party mutations in the service fixture; it exercises production connection, transport, tool approval, and document delivery paths.

Deterministic checks and calibration

pnpm test:e2e:runner:typecheck
pnpm test:e2e:runner:unit
python3 -m unittest discover -s tests/runner-e2e -p test_everyday_artifact.py

The independent oracle rejects wrong output, ignored late feedback, trailing separator bugs, invalid argument acceptance, duplicate source modules, archive path traversal, and symlinks. Passing agent-authored tests cannot override it. Lifecycle calibration rejects legacy execution, missing runner identity, unexpected crashes, workers on the parent, and answers left in review.

ZIP evaluation runs delivered Python in a Docker container with a read-only project mount and root filesystem, no network, a non-root user, no Linux capabilities, and bounded CPU, memory, process count, output, and duration. Only the extracted delivery enters the container. The container is removed after grading. Calibration includes attempts to read a host file and reach a host loopback service.

Evalbook evidence and qualification

Each packaged attempt retains snapshots/everyday-workflow.json, downloaded ZIPs, assertions, actual task comments and run records, timing, accounting, source provenance, and screenshots. The story records a digest of its harness sources. Infrastructure failures and failed attempts must remain inspectable.

Import packaged results with paperclip-evals/evals/everyday-workflows/import_results.py. It uses the canonical Runner Evalbook generator and the built Runner Lab viewer. It does not invent provider transcripts, tool counts, model observations, or cost estimates. The selected model is checked against persisted native execution inputs; that is distinct from provider-side model identity verification.

Initial live results are diagnostic. They are not a reliability estimate or a model ranking. Before promotion, freeze both source revisions and harness digest, run at least three independent local repetitions, qualify the eight remote cells against a verified image, and review every failure. Keep model quality, lifecycle correctness, infrastructure availability, and latency separate.

Revised evaluation contract (14 September, second campaign)

That campaign used 32 cells: the original local stories plus two local Codex text-only safe-replacement probes, and the unchanged six remote cells. The old recover-runner results remain historical; recover-runner-uncertain is a new case that expects a visible Blocked safety stop, preserved source and queued input, and no unverified provider replay. Its Retry control is inspected, not claimed to restore work. Successful manual recovery remains unqualified.

recover-runner-safe interrupts a text-only Codex turn and queues new direction. A pass requires the server's durable verified_safe_replacement evidence and the new answer. No safety proof is injected or fabricated. If that premise cannot be verified in a live probe, report it as an unqualified recovery boundary, not an established product defect. Claude has no catalog cell for this Codex-specific replacement proof. Deterministic native-safe-replacement tests cover its proof and admission gates independently of model behavior.

Delegation now submits feedback through the existing child task composer and records the delivered comment ID. The child must consume the message and deliver the revised program. This does not require a lead to relay a parent comment. The separate issue-update-comment-wakeup route tests exercise exact supported mention routing, including access, dependency, identity, and duplicate-wake gates.

The approval case provisions an authenticated local service through the public API, with a random server-held credential that never enters the agent environment or browser trace. Approval/decline interactions still use the browser. Provider captures distinguish rejected unauthenticated requests from accepted calls. The old public-endpoint attempts remain boundary evidence, not an isolation promise.

Stop now waits for the owned runner to exit, records project file hashes, and checks them again after the new response. This proves stability over that interval, not indefinite monitoring. Hiring and declined-access policy changes are deferred by user decision; their old results must not be presented as new campaign runs.

Decline correction (14 September, third campaign)

That campaign used 35 cells (29 local, six remote). service-decline tests rejection of a protected action on an already installed service; its former "connection request" title was misleading. connection-decline separately tests Not now on new Notion setup. Both permit a brief explanation as the complete fallback, so Done is expected after that explanation. Neither test requires completion after refusing work that is still required.

The installed-service decline fixture now uses the same server-held credential as approval. The harness requires one pending interaction, validates its kind and connection/provider identity before clicking, and waits for the exact interaction's saved decision. Wrong interactions fail decision-request-matches-story with a screenshot; they are not evidence of an ignored decline. Both decline stories check a new explanation after the decision and reject repeated requests.

Historical attempts remain unchanged. This campaign resumes the previously deferred decline cases; hiring remains deferred. Notion setup is declined in the UI, so this test neither authenticates to nor reads real Notion data.

Decision screenshots are included in the evidence package. Before capturing the final screen, the harness waits for the thread and latest persisted agent comment to render, then scrolls that comment into view. A Done header alone is not proof that the final response was visible.

Recovery scope correction (14 September)

This correction reduced the catalog to 30 cells: 24 local and six remote. Forced runner crash probes are retired from paid selection. Their original attempt IDs remain in Evalbook's Diagnostics history and Latest pages; they are excluded from the main matrix without changing grades or deleting evidence. Reported spend still includes all attempts.

recover-controller and stop-redirect retain concrete supported journeys: restart the controller while preserving the runner, or use Stop and submit a new direction. Their assertions verify pending input, saved work, and the next usable result. Neither claims recovery from an arbitrary provider-process crash.

A future user-facing crash-recovery case needs a reproducible recoverable fault, an identified supported recovery action, and evidence through the final usable result. A missing test premise must be reported as unexercised, not a model failure. Do not introduce a new paid case just to replace a retired row.

Skill creation (16 September)

create-skill-studio adds five cells: three local profiles and the two core profiles on Daytona. The current catalog has 35 cells: 27 local and eight remote. The test opens the created skill from its task-feed card, checks the canonical skill identity in Studio, saves an edit, and returns to the same skill in the task sidebar. A model's authored document heading is not used as the identity check.

External-provider fallback

Three explicit local cases cover aggregator routing with the normal production agent guidance. They add nine local cells. The AgentMail setup case adds three local cells; the suite now has 50 cells total.

The AgentMail case uses real model discovery and production interaction/UI paths, but declines before sending credentials to AgentMail. Its independent grader is calibrated against missing, duplicate, misaddressed, hidden, and malformed cards. The email integration suite separately proves credential/inbox creation, access defaults, assignment checks, and completion using a fixture provider. Neither test qualifies live AgentMail delivery. Run a bounded local model cell with:

pnpm test:e2e:runner -- --id everyday-workflows.runner-codex-mini.local.agentmail-setup --max-automatic-retries 0

• provider-native: Jira is supported natively and by aggregators. Require the Jira connection card directly, decline it in the browser, and verify no provider question, connection creation, or repeated request.
• provider-decline: HubSpot has no built-in connector in this fixture. Require Composio, Arcade, Zapier, and None in that order with external-service disclosure. Restart the controller, reload the question, choose None through the UI, and verify one saved answer, no connection changes, and no fabricated result.
• provider-second: Install a deterministic Arcade gateway with a read-only HubSpot action through public APIs. Choose Arcade in the browser after restart. Require no calls before selection, exactly one call afterwards, the independently generated contact marker in the agent response, and no duplicate connection.

These use the existing native profiles and 12-minute local attempt deadline; expected provider runs are two per case. There are no real third-party mutations. The fixture server is closed and the harness cleans its disposable instance. Provider-choice screenshots, persisted interactions, gateway call counts, source revision, harness digest, and existing usage/cost evidence accompany each attempt. connection-routing-evidence.test.ts calibrates the grader against undisclosed routing, incorrect ordering, early calls, duplicate questions, and fabricated reads. Run a single everyday-workflows.runner-codex-mini.local.provider-decline cell first; do not treat these fixtures as live provider compatibility tests.

Connection-guidance observation boundaries

The explicit native-connection-guidance suite keeps one attempt per cell. For decline cases, Done and a succeeded run are insufficient: settlement waits within the existing cell deadline for a saved reply attributed by run ID to the lead agent's final successful run on this task. Readiness checks storage and identity, not favorable wording. A missing reply times out; an incorrect reply settles and fails the independent explanation check. Both decline explanation checks use the same bounded matcher, including unavailable/rejected access and “wasn't able to pull” wording. This is textual evidence, not proof of cognition.

An executed approval or a recorded tool-action rejection with wake_assignee may precede its continuation run. Only a response bound to the same task, company, agent, and completed source run can defer the stranded-blocker check, and only within the original deadline. A failed execution, unrelated card, or already consumed response cannot extend that wait.

connection-guidance-decline-grade.json and the workflow's declineGradeEvidence retain the exact cloned assertion input and original checks. Later final-state or cleanup observations cannot overwrite that input. Earlier campaign grades remain unchanged when the evaluator is corrected.

Upgrade qualification receipts

PAPERCLIP_RUNNER_E2E_UPGRADE_ARTIFACTS selects frozen installed A/B artifacts and an alternating A/B/A sequence. The remote controller-recovery workflow executes every requested replacement on the original active run. If that run ends before a later fault can be injected, that boundary is unexercised; starting a replacement task does not qualify it.

The supervisor reads existing Postgres ownership and Runner-owned checkpoint evidence. It records HTTP readiness, natural lease takeover, a new authenticated connection, and ingestion past the stopped controller's saved event cursor. Final native and workspace receipts establish backlog settlement. No lease timestamps are edited. Authentication and processing must be observed within the original 420-second grace, including artifact verification and startup.

For a credential-free startup/migration experiment, run controller-replacement-smoke.ts with the manifest and an explicit disposable PAPERCLIP_TEST_DATABASE_URL. The driver creates and drops a unique empty database; it does not change the supplied database. This smoke qualifies zero agent turns. The qualification record is the acceptance ledger for exact artifacts, attempts, and remaining gates.