mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-15 18:56:05 +01:00
## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat eval catalog runs real agents through chat journeys and
grades the saved results
> - Full cohort 4 (OpenCode with DeepSeek v4.1 flash, master f1aef0cb58)
failed 14 cases for eval reasons, not product reasons
> - Some phase captures read links before the reply rendered. Some
graders matched keywords in free-form replies. One locator was stale.
One fixture answer looked like a placeholder. Some checks could not say
what failed
> - This pull request grades those cases on saved state, setup literals
and rendered links, and it fixes the observers and fixtures
> - The benefit is that these cases fail only when the product fails,
and a failure names the part that broke
## Linked Issues or Issue Description
**What happened?**
In full chat cohort 4, valid agent behavior failed these cases:
case-history, goals.explain-purpose and workspaces.preview-service
(phase captures read before the reply rendered); routines.pause-owned
("If you want that one stopped too, use Pause on its task page" matched
a cancellation regex); budgets-costs.spend-breakdown ("unknown, not
$0.00" matched a stated-zero regex); steering.edit-queued-message
("change ... and drag it above" missed a keyword list);
connection-setup.choose-provider (the resumed page says "Finish with
Notion"); questions-and-approvals.necessary-question (the answer "Acorn
2a4f72ec4ae2-1" was saved as "Acorn"). Some checks could not say what
failed: onboarding.new-company compared the whole members response as
JSON and saved no after-state; agents.choose-primary threw one combined
error.
**Expected behavior**
Each case grades the feature contract on saved state, setup literals and
rendered or clicked links. A valid reworded reply passes. A broken
product still fails. A failure names the clause and its values.
**Steps to reproduce**
Grade the retained cohort 4 evidence for the cases above with the
earlier oracles.
Refs #21009, #21002, #21007
## What Changed
- Phase captures (case-history first phase, goals.explain-purpose
phases, preview-service query phase, ambiguous-task clarification,
edit-queued-message guidance) now wait for the saved reply to render
(`waitForTurnReplyLinks`) before they observe. When a saved reply yields
no rendered links, the observation records thread and anchor diagnostics
(`linkDiagnostics`) and logs them.
- routines.pause-owned: remove the reply keyword clause. Saved state at
the reply proves the run kept running, and the final state proves it
completed normally.
- budgets-costs.spend-breakdown: the unresolved record is graded
structurally. The reply must identify it (rendered link, identifier or
run id), and a chat run must have read its saved accounting (cost
summary, run record or unranged company report, bound by input digest).
"$0.00" in the reply is not matched.
- steering.edit-queued-message: grade the rendered task link in the
reply, the refused agent edit and reorder (403), and the unchanged queue
at the reply. Keyword lists are removed.
- connection-setup.choose-provider: the sign-in locator also accepts
"Finish with Notion" (the resumed-draft label). The button has no test
id or other stable name.
- questions-and-approvals.necessary-question: the answer is now
"Quillmere". The case also accepts the exact answer in a saved
deliverable (a document revision or text attachment on the linked task,
saved by an agent of the company after the answer). The yaml expected
outcome records this decision, with a `changedFields` entry in the
testops manifest.
- onboarding.new-company: save the members after-state and compare
identity, role, status and grants, and name what changed. Guidance is
graded on a rendered link to `/onboarding` (the new-organization
wizard), the business literal from the query, and the saved company
count. The source company and person get realistic names for this case
and are restored at cleanup, because the CEO reasoned from the generated
names that the request was a graded test.
- agents.choose-primary: the settled-Chat check is split into four named
clauses with values, saved as evidence on failure.
- A question card the turn ended on counts as its reply (decision after
cohort 4) in cli-operations.owned-dev-instance, onboarding.new-company
and status-cards.other-authors-card. The card's Markdown-rendered fields
(question form description; confirmation prompt and details) count as
reply links. Link requirements still apply.
- Every changed case bumps its `oracleVersion`; the descriptor snapshot
and inventory digest are updated.
Why each old check was wrong and why the new one still catches a broken
product:
| Case | Old check was wrong because | A broken product still fails
because |
|---|---|---|
| phase captures | runs succeeded and the chat waited before the reply
rendered, so links were empty | the same link checks run on the rendered
reply |
| routines.pause-owned | a correct offer to pause the run read as a
cancellation | the checkpoint run must be running and the run must later
succeed |
| spend-breakdown | "unknown, not $0.00" read as "$0" | the reply must
identify the record and a chat run must read its saved accounting;
totals and per-agent amounts stay graded |
| edit-queued-message | correct wording missed a keyword list | the
reply must link the task, the agent's edit and reorder must be refused,
and the queue must be unchanged at the reply |
| choose-provider | the resumed page label was not accepted | the button
must still be visible on the selected Notion surface |
| necessary-question | the answer read as a placeholder; deliverables
did not count | the exact answer must be saved in the linked work after
the answer, by an agent of the company |
| new-company | timestamps and profiles made the JSON compare fail
without saying why | membership identity, role, status and grants must
be unchanged |
| choose-primary | one combined error | the same four conditions, each
named |
| card as reply | a card answer read as an empty reply | card text must
exist and link requirements still apply |
## Verification
- Calibration tests (valid, reworded valid, broken) for each changed
grader: `integrator-routine-pause.test.ts`,
`identity-accounting.test.ts` (includes the cohort 4 reply),
`runtime-queue-controls.test.ts` (includes the cohort 4 reply),
`external-provider-choice.test.ts`, `identity.test.ts`,
`identity-onboarding-work.test.ts`,
`runtime-primary-ready-chat.test.ts`, `integrator-status-cards.test.ts`,
`turn-response-card.test.ts`, `chat-rendered-links.test.ts`.
- The corrected spend-breakdown grader passes on the retained cohort 4
evidence (its cost-summary read binds to the unresolved task) and fails
with the reads removed.
- `npx vitest run --config tests/runner-e2e/vitest.config.ts
chat-catalog chat-rendered chat-linked`: 284 files passed.
- `node --test
tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10/10
passed.
- `node scripts/typecheck-test-projects.mjs tests/runner-e2e`: passed.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs <sha> --check`:
unchanged.
- Paid run (8 cells, OpenCode + DeepSeek v4.1 flash) on this branch
combined with the guidance PR: see the PR comment for results.
## Risks
- The new-company guidance check now needs a rendered link to
`/onboarding`. A reply that only names the company switcher fails. The
companion guidance PR tells agents this route.
- spend-breakdown does not grade wording, so a reply that reads the
record and still calls it "$0" is not caught. The amounts, total and
identification are still graded.
- The new-company case renames the source company and the person for the
case and restores them at cleanup.
## Model Used
- Anthropic Claude Opus 5.5 (`claude-opus-5-5`) in Claude Code, with
tool use, shell execution and code editing.
## 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
- [ ] All Paperclip CI gates are green
- [ ] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [ ] I will address all Greptile and reviewer comments before
requesting merge
🤖 Generated with [Claude Code](https://claude.com/claude-code)
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
126 lines
14 KiB
TypeScript
126 lines
14 KiB
TypeScript
import { chatCatalogNativePhaseDeadline, chatCatalogNativePhaseIsTimely, chatCatalogNativePhaseRemainingBudget } from "../chat-catalog-contract.js";
|
|
import { expect } from "@playwright/test";
|
|
import { readFile, realpath, writeFile, unlink } from "node:fs/promises";
|
|
import path from "node:path";
|
|
import { pollUntil } from "../api.js";
|
|
import { sendChatMessage } from "../chat-flow.js";
|
|
import { checks, type Entity } from "./internal-support.js";
|
|
import { enableTaskWorkspaceView, savedTaskWorkspace } from "./runtime-workspaces.js";
|
|
import { runtimeDefinition, runtimeObservation, observeRuntime } from "./runtime-support.js";
|
|
import { waitForTurnReplyLinks } from "./identity-support.js";
|
|
import type { WorkspaceRuntimeService } from "../../../packages/shared/src/types/workspace-runtime.js";
|
|
import type { WorkspaceOperation } from "../../../packages/shared/src/types/workspace-operation.js";
|
|
import type { IssueWorkProduct } from "../../../packages/shared/src/types/work-product.js";
|
|
interface PreviewWorkspace extends Entity { cwd: string; config: Record<string, unknown>; runtimeServices?: WorkspaceRuntimeService[] }
|
|
// An origin URL and the same origin with a trailing slash name the same endpoint; compare parsed URLs.
|
|
const sameUrl = (left: unknown, right: unknown) => { try { return typeof left === "string" && typeof right === "string" && new URL(left).href === new URL(right).href; } catch { return false; } };
|
|
export function gradePreview(value: unknown) {
|
|
const c = checks(); const o = runtimeObservation(value);
|
|
c.add("actual-preview-lifecycle-evidence", !!o, "Actual service identity, HTTP content, source output and subsequent startup failure must be independently observed."); if (!o) return c.result();
|
|
const service = o.state.healthyService as WorkspaceRuntimeService | undefined; const workspace = o.state.workspace as PreviewWorkspace;
|
|
const products = o.state.previewProducts as IssueWorkProduct[] | undefined;
|
|
c.add("real-ready-service-correct-workspace", !!service && service.executionWorkspaceId === workspace.id && service.companyId === o.companyId
|
|
&& service.status === "running" && service.healthStatus === "healthy" && service.serviceName === "Website preview"
|
|
&& o.state.initialHttpStatus === 200 && o.state.initialHttpBody === o.state.websiteBytes
|
|
&& typeof o.state.initialHttpBody === "string" && o.state.initialHttpBody.includes(String(o.state.marker)) && o.state.browserPreviewVisible === true,
|
|
"A real managed Website preview serves the actual provider-written website bytes in the intended workspace and is usable in the browser.");
|
|
// The query asks for a preview to look at; it does not ask for a work product. A registered
|
|
// preview/runtime-service product is optional, but one the lead registers must name this service.
|
|
const leadPreviewProducts = (products ?? []).filter(row => ["preview_url", "runtime_service"].includes(row.type)
|
|
&& o.runs.some(run => run.id === row.createdByRunId && run.agentId === o.leadId));
|
|
c.add("accessible-registered-preview", !!service?.url && Array.isArray(products)
|
|
&& leadPreviewProducts.every(row => sameUrl(row.url, service.url) && (row.runtimeServiceId ?? service.id) === service.id
|
|
&& (row.executionWorkspaceId ?? workspace.id) === workspace.id)
|
|
&& (o.state.queryLinks as string[] | undefined)?.some(link => sameUrl(link, service.url)) === true,
|
|
"The actual canonical reply links the ready endpoint of this exact workspace's service; any preview work product the lead registers names that service.");
|
|
const failure = o.state.failureOperation as WorkspaceOperation | undefined; const reply = String(o.state.failureReply ?? "");
|
|
c.add("startup-failure-exposed-actionably", !!failure && failure.status === "failed" && failure.executionWorkspaceId === workspace.id
|
|
&& (o.state.failureRunIds as string[] | undefined)?.includes(String(o.state.failureOperationRunId ?? failure.heartbeatRunId ?? "")) === true
|
|
&& o.state.failedHttpReachable === false && reply.trim().length > 0,
|
|
// The reply's wording (failure kept visible, next repair step) is for a human reader; saved state proves the outcome.
|
|
"A separate explicit restart really fails under the fixture's process-start fault, the endpoint is down, and that restart turn answered in chat.");
|
|
c.add("no-unrelated-work-or-publication", o.tasks.length === 1 && o.tasks[0]?.id === (o.state.task as Entity).id
|
|
&& o.runs.length === 3 && o.runs.every(run => run.status === "succeeded"),
|
|
"The prepared website plus two explicit controller turns are the only native work; no remote deployment or extra task is substituted.");
|
|
return c.result();
|
|
}
|
|
export const runtimePreviewCases = [runtimeDefinition("workspaces.preview-service", async ctx => {
|
|
if (ctx.input.execution.environment.id !== "local") throw new Error("Owned loopback website preview fixture requires the actual local target");
|
|
await enableTaskWorkspaceView(ctx);
|
|
const state = await savedTaskWorkspace(ctx, { html: true }); const cwd = await realpath(state.workspaceBefore.workspace.cwd);
|
|
const root = process.env.PAPERCLIP_RUNNER_E2E_TEMP_ROOT;
|
|
if (!root || !cwd.startsWith(`${await realpath(root)}${path.sep}`)) throw new Error("Preview fixture escaped the isolated task directory");
|
|
const websiteBytes = await readFile(path.join(cwd, "index.html"), "utf8");
|
|
if (!/<html|<!doctype html/i.test(websiteBytes) || !websiteBytes.includes(state.marker)) throw new Error("Actual native website prerequisite is not present");
|
|
const failureFlag = path.join(cwd, "preview-start-failure.fixture"); const script = path.join(cwd, "preview-server.cjs");
|
|
await writeFile(script, `const fs=require('node:fs');const http=require('node:http');const path=require('node:path');if(fs.existsSync(path.join(__dirname,'preview-start-failure.fixture'))){console.error('Owned preview startup fixture failed with exit 17');process.exit(17)}const body=fs.readFileSync(path.join(__dirname,'index.html'));http.createServer((req,res)=>{res.writeHead(200,{'content-type':'text/html; charset=utf-8'});res.end(body)}).listen(Number(process.env.PORT),'127.0.0.1');`, "utf8");
|
|
const workspace = await ctx.input.api.get<PreviewWorkspace>(`/api/execution-workspaces/${state.workspaceBefore.workspace.id}`);
|
|
await ctx.input.api.patch(`/api/execution-workspaces/${workspace.id}`, { config: { ...workspace.config, desiredState: "stopped", workspaceRuntime: { services: [{
|
|
name: "Website preview", command: "node preview-server.cjs", cwd: ".", port: { type: "auto" }, lifecycle: "shared", stopPolicy: { type: "manual" },
|
|
readiness: { type: "http", urlTemplate: "http://127.0.0.1:{{port}}", timeoutSec: 10, intervalMs: 100 },
|
|
expose: { type: "url", urlTemplate: "http://127.0.0.1:{{port}}" },
|
|
}] } } });
|
|
const saved = await ctx.input.api.get<PreviewWorkspace>(`/api/execution-workspaces/${workspace.id}`);
|
|
if (saved.runtimeServices?.some(row => row.status === "running")) throw new Error("Preview setup must not start the target service");
|
|
return { ...state, workspace: saved, websiteBytes, failureFlag };
|
|
}, gradePreview, {
|
|
oracleVersion: "paperclip.chat-catalog.preview-service.v5",
|
|
sourceFiles: ["chat-catalog-domains/runtime-preview.ts", "chat-catalog-domains/runtime-workspaces.ts", "chat-catalog-domains/runtime-support.ts", "chat-catalog-domains/identity-support.ts", "chat-rendered-links.ts", "chat-scenarios/workspaces.yaml"],
|
|
expectedHumanMessageCount: 2, expectedRunCount: 3, attemptTimeoutMs: 720_000,
|
|
async act(ctx, state) {
|
|
await pollUntil({ label: "Actual canonical preview service readiness", deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 180_000),
|
|
load: async () => { await ctx.refreshIssue(); return { source: ctx.issue(), runs: await ctx.allRuns(), workspace: await ctx.input.api.get<PreviewWorkspace>(`/api/execution-workspaces/${(state.workspace as Entity).id}`) }; },
|
|
accept: value => chatCatalogNativePhaseIsTimely(ctx) && value.source?.conversationState === "waiting" && value.runs.length === 2 && value.runs.every(run => run.status === "succeeded")
|
|
&& value.workspace.runtimeServices?.some(service => service.status === "running" && service.healthStatus === "healthy") === true });
|
|
const ws = await ctx.input.api.get<PreviewWorkspace>(`/api/execution-workspaces/${(state.workspace as Entity).id}`);
|
|
const service = ws.runtimeServices?.find(row => row.serviceName === "Website preview" && row.status === "running")!;
|
|
if (!service?.url) throw new Error("A created service record without an accessible ready endpoint is not a preview");
|
|
const url = new URL(service.url); if (url.protocol !== "http:" || url.hostname !== "127.0.0.1" || Number(url.port) !== service.port) throw new Error("Preview endpoint escaped its owned loopback service");
|
|
const response = await fetch(url, { redirect: "error", signal: AbortSignal.timeout(10_000) });
|
|
state.initialHttpStatus = response.status; state.initialHttpBody = await response.text(); state.healthyService = service;
|
|
// The service and finished runs precede the saved reply's render; read its links only after it renders.
|
|
await waitForTurnReplyLinks(ctx);
|
|
const o = await observeRuntime(ctx, state); state.queryLinks = o.links;
|
|
const products = [];
|
|
for (const id of [(state.task as Entity).id, ctx.issue()!.id]) products.push(...await ctx.input.api.get<IssueWorkProduct[]>(`/api/issues/${id}/work-products`));
|
|
state.previewProducts = products;
|
|
const source = ctx.input.page.url(); await ctx.input.page.goto(service.url, { waitUntil: "domcontentloaded" });
|
|
await expect(ctx.input.page.getByRole("heading", { name: String(state.marker), exact: true })).toBeVisible(); state.browserPreviewVisible = true;
|
|
await ctx.input.capture("real-website-preview", "The actual provider-written website is served by the ready managed service", "runtime-website-preview.png");
|
|
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
|
|
const beforeRuns = (await ctx.allRuns()).map(run => run.id);
|
|
const beforeOperations = (await ctx.input.api.get<WorkspaceOperation[]>(`/api/execution-workspaces/${ws.id}/workspace-operations`)).map(row => row.id);
|
|
await writeFile(String(state.failureFlag), "Deliberate bounded owned process-start failure fixture", "utf8");
|
|
await sendChatMessage(ctx.input.page, "Restart that same website preview now and tell me whether it actually comes back. If startup fails, keep that failure visible and explain the next repair step; do not claim a working preview.", chatCatalogNativePhaseRemainingBudget(ctx));
|
|
const settled = await pollUntil({ label: "Explicit preview restart reports its genuine startup failure", deadlineAt: chatCatalogNativePhaseDeadline(ctx, Date.now() + 180_000),
|
|
load: async () => { await ctx.refreshIssue(); return { source: ctx.issue(), runs: await ctx.allRuns(), operations: await ctx.input.api.get<WorkspaceOperation[]>(`/api/execution-workspaces/${ws.id}/workspace-operations`) }; },
|
|
accept: value => chatCatalogNativePhaseIsTimely(ctx) && value.source?.conversationState === "waiting" && value.runs.length === 3 && value.runs.every(run => run.status === "succeeded")
|
|
&& value.operations.some(row => !beforeOperations.includes(row.id) && row.status === "failed") });
|
|
const ids = settled.runs.filter(run => !beforeRuns.includes(run.id)).map(run => run.id); state.failureRunIds = ids;
|
|
// Runtime controls an agent calls through the API record no heartbeatRunId; attribute such an
|
|
// operation to the restart turn whose run window contains its start.
|
|
const turnOf = (row: WorkspaceOperation) => row.heartbeatRunId ?? settled.runs.find(run => ids.includes(run.id) && run.startedAt
|
|
&& Date.parse(String(row.startedAt)) >= Date.parse(String(run.startedAt)) && (!(run as { finishedAt?: string | null }).finishedAt || Date.parse(String(row.startedAt)) <= Date.parse(String((run as { finishedAt?: string | null }).finishedAt))))?.id ?? null;
|
|
const failed = settled.operations.find(row => !beforeOperations.includes(row.id) && row.status === "failed" && ids.includes(turnOf(row) ?? ""));
|
|
state.failureOperation = failed; state.failureOperationRunId = failed ? turnOf(failed) : null;
|
|
const comments = await ctx.input.api.get<Entity[]>(`/api/issues/${ctx.issue()!.id}/comments?order=asc`);
|
|
state.failureReply = comments.filter(row => ids.includes(row.createdByRunId ?? "") && row.authorAgentId === ctx.input.fixtures.agent.id).map(row => row.body ?? "").join("\n");
|
|
state.failedHttpReachable = await fetch(url, { redirect: "error", signal: AbortSignal.timeout(5_000) }).then(async res => { await res.body?.cancel(); return res.ok; }, () => false);
|
|
},
|
|
observe: observeRuntime,
|
|
async verifyResult(ctx, state) {
|
|
const source = ctx.input.page.url(); const ws = state.workspace as Entity;
|
|
await ctx.input.page.goto(`/${ctx.input.fixtures.company.issuePrefix}/execution-workspaces/${ws.id}`, { waitUntil: "domcontentloaded" });
|
|
await ctx.input.page.reload({ waitUntil: "domcontentloaded" });
|
|
await expect(ctx.input.page.getByRole("heading", { name: ws.name!, exact: true }).first()).toBeVisible();
|
|
// The workspace page opens on Tasks; failed runtime operations are listed on its Runtime logs tab.
|
|
await ctx.input.page.getByRole("tab", { name: "Runtime logs", exact: true }).click();
|
|
await expect(ctx.input.page.getByText(/failed|exit 17|startup/i).first()).toBeVisible();
|
|
await ctx.input.page.goto(source, { waitUntil: "domcontentloaded" });
|
|
},
|
|
async cleanup(ctx, state) {
|
|
try { if (state.workspace) await ctx.input.api.post(`/api/execution-workspaces/${(state.workspace as Entity).id}/runtime-services/stop`, {}); }
|
|
finally { if (state.failureFlag) await unlink(String(state.failureFlag)).catch(error => { if ((error as NodeJS.ErrnoException).code !== "ENOENT") throw error; }); }
|
|
},
|
|
})];
|