Commit Graph
7 Commits
Author SHA1 Message Date
DottaandClaude Opus 5.5 e62fe18c72 fix(evals): wait for the primary save, select the offered Launch choice, and pace the keyboard reorder (#21396)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat eval catalog checks real chat journeys against saved
product state
> - Full cohort 12 (OpenCode with
`openrouter/deepseek/deepseek-v4.1-flash-1017`) failed three cases where
the agent did the correct thing
> - In each case the harness, not the product, broke the journey: it
left a page too early, could not answer a choice card, or pressed keys
faster than the drag library listens
> - A harness that breaks the journey cannot measure the product
> - This pull request makes the harness act like a person in these three
journeys, counts a verified card choice as a human action, and keeps
each check able to fail a broken product
> - The benefit is that these three cases measure the product again

## Linked Issues or Issue Description

**What happened?**
In full cohort 12, three cases failed for harness reasons:

- `agents.choose-primary`: the CEO answered correctly. The harness
clicked the crown and confirmed. Then it waited for "My primary agent"
with `.first()`. The sidebar already showed the CEO's crown, so this
wait passed at once. The harness then navigated away before the browser
sent `PUT /primary-agent/me`. The server log has no PUT. The read-back
still showed the CEO.
- `tasks.ambiguous-task`: the CEO changed nothing and asked a correct
question card. The card linked both tasks and offered "COM-1 (Launch
project)", "COM-2 (Experiments project)" and "Both tasks". The harness
could only answer with a text box or an "Other" option. It threw "The
linked question has no supported explicit clarification input".
- `steering.edit-queued-message`: the person's edit saved (PATCH 200).
The keyboard reorder (Space, ArrowUp, Space) sent no `PUT
/queued-comments/order` in 2 gestures. Cohort 9 had the same failure.

**Expected behavior**
Each journey completes as a person does it, and the oracle grades the
saved result.

**Steps to reproduce**

1. Use master `f7ddab8e94`.
2. Run the Runner Full-Stack E2E workflow with `all=false` and the three
case ids `chat-features.runner-opencode.local.<case>`.
3. Read each case's `playwright.log` and `server.log`.

**Additional context**
- Primary agent: the crown fills at once (optimistic update) and the
save starts on confirm. The mutation lives in the app-level
`PrimaryAgentProvider`, so in-app navigation does not cancel it. In the
failed run the confirm and the full-page `goto` happened within the same
second, and the PUT never reached the server. Only a full page load in
the first milliseconds can lose the save. A person cannot do that in
practice, so this is not a product bug. No UI change.
- Keyboard reorder: dnd-kit's `KeyboardSensor` starts the drag on Space,
adds its keydown listener in a `setTimeout`, and
`sortableKeyboardCoordinates` ignores arrows until the rows are measured
on a later render. Keys pressed back to back can arrive before this. In
the production Storybook story (headless Chromium), back-to-back
`locator.press` moved the item in 6 of 8 tries. Keys about 150 ms apart
moved it 5 of 5 times, also after an Edit and Save. With the drag-start
wait and a paint between keys it moved 16 of 16 times, also at 6x CPU
throttle. The keyboard reorder works for people, so this is not a
product bug. No UI change.

## What Changed

- `runtime-primary.ts`: wait for the browser's own `PUT
/primary-agent/me` that names Morgan, require 2xx, and poll the
read-back until Morgan is primary before leaving the page. Look for the
"My primary agent" crown only in the header that holds Morgan's avatar.
Add `isPrimarySaveRequest`. Oracle `runtime-primary.v6`.
- `chat-linked-question.ts` / `internal-tasks.ts`: add
`chosenTaskOption`. It finds the one option of the bound saved question
that names the Launch task's identifier (a setup literal) and no other
candidate's. It uses the option id for the rendered DOM id and the
payload's submit label. It does not read option wording. The harness
selects that option, checks `aria-checked`, submits, and waits for the
`respond` POST. The old text and "Other" paths stay for other cards. The
checks that nothing changes before the answer and that only COM-1
changes after it are unchanged.
- `chat-catalog-contract.ts` / `internal-tasks.ts`: a chosen option
saves no comment, so the flow counted only one human action and failed
`catalog-original-human-query`. `tasks.ambiguous-task` now joins the
cases that may count a bound UI answer. The case declares
`expectedHumanQuestionAnswer` only after the harness answered through
the choice, and names the chosen question and option. The flow then
verifies the answered card independently: answered by this person,
exactly the Launch option and no typed text, one audit event, and one
continuation run after it. Oracle `internal.ambiguous-task.v8`.
- `chat-catalog-contract.ts`: `chatCatalogHumanQuestionPolicies` takes
the answering person and mirrors the server's user rule in
`evaluateIssueThreadInteractionResolverAudience`. `anyone` and
`human_only` admit the person. `not_creator` admits a person who did not
create the card. With no resolver (a pending card) `not_creator` stays
rejected, so other callers keep their behavior. For the ambiguous-task
choice, an unaddressed card is accepted, as the server allows.
- `runtime-queue.ts`: `keyboardMoveQueuedMessageUp` waits for the
handle's `aria-pressed="true"` after Space. It lets the page paint
before ArrowUp and before the closing Space. It records whether each
gesture started a drag. Oracle `edit-queued-message.structural.v7`.
- No scenario YAML changed, so the TestOps manifest needs no entry.

## Verification

- `npx vitest run --config tests/runner-e2e/vitest.config.ts` for
`chat-linked-question.test.ts`, `chat-catalog-consistency.test.ts`,
`runtime-primary.test.ts`, `runtime-queue.test.ts`, `internal.test.ts`,
`external-company-secret.test.ts`, `external-personal-secret.test.ts`,
`turn-response-card.test.ts`, `runtime-task-coordination.test.ts`: pass.
- Red/green: the two new keyboard tests fail on the old
`keyboardMoveQueuedMessageUp` and pass on the new one. The
resolver-policy test fails on the old `chatCatalogHumanQuestionPolicies`
(`not_creator` answered by a person who is not the creator) and passes
on the new one; `not_creator` answered by its creator and a pending card
stay rejected. The choice-accounting test fails on the old allow list.
- `chosenTaskOption` tests use the cohort 12 card shape, a reworded
card, and cards with no single choice (both, duplicate, neither,
`COM-10`, a different card).
- `npx tsc -p tests/runner-e2e/tsconfig.json --noEmit`: pass. `node
--test tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10
pass.
- Paid narrow run 1 (commit `33399f8370`):
https://github.com/paperclipai/paperclip/actions/runs/43446445525.
`agents.choose-primary` passed (browser `PUT /primary-agent/me` 200).
`steering.edit-queued-message` passed (browser `PUT
/queued-comments/order` 200). `tasks.ambiguous-task` selected COM-1
(`respond` 200, only COM-1 became high) but failed
`catalog-original-human-query`, because the flow did not count the
choice. Later commits fix that.
- Paid narrow run 2 (`c90c2aa26a`) was cancelled by run 3 before its
case finished.
- Paid narrow run 3 (`tasks.ambiguous-task` only, final verifier
`fee3bbd453`):
https://github.com/paperclipai/paperclip/actions/runs/43452010284

## Risks

- Low risk. Test harness only.
- The ambiguous-task harness now answers through a structured choice
when one exists. If the agent asks a card with no option that names only
the Launch task, the old text and "Other" paths still apply.
- `chatCatalogHumanQuestionPolicies` accepts `not_creator` only when a
resolver is passed. Callers that check pending cards pass none, so their
behavior is unchanged.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with
tool use and code execution.

## 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
- [x] 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>
2026-11-13 09:56:55 -06:00
DottaandClaude Opus 5.5 d16becac04 test(evals): fix cohort-10 secrets entry, steering graders, private-task settle and queue brief (#21296)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat eval catalog checks real chat journeys against saved
product state, not reply wording
> - Full cohort 10 failed six cases where the product worked but the
eval was out of date or raced
> - #21225 redesigned Settings → Secrets, and #21243 changed when a
queued comment counts as delivered
> - The private secret steps, two steering graders, one final run check
and one fixture did not follow
> - This pull request updates those eval layers and keeps each one able
to fail a broken product
> - The benefit is that these six cases measure the product again
instead of the stale harness

## Linked Issues or Issue Description

Refs #21225
Refs #21243

**What happened?**
In full cohort 10 (OpenCode, local), six cases failed for eval reasons:

- `secrets.configure-company-secret` and `secrets.personal-secret`: the
private browser step waited for the old "Create secret" dialog and its
Provider field, and for the old personal form text. #21225 renamed the
dialog to "New secret", removed Provider, moved Key under "More
options", and moved the definition name into the "Set your value for …"
heading.
- `steering.steer-active` and `steering.discard-queued-message`:
Morgan's running turn read the queued comments through task history and
cited them in its accepted result. Under #21243 that is delivery, so no
continuation ran. The graders still required one more run and timed out.
- `tasks.private-task`: the final run check read the ledger while the
CEO's `chat_task_completed` follow-up was still running.
- `steering.queue-follow-up`: the brief was only a reference string, so
the continuation asked a question instead of building.

**Expected behavior**
Each case grades the product's current contract on saved state, and
still fails a broken product.

**Steps to reproduce**
Run the six cases at `origin/master` with the OpenCode local profile.
See the cohort-10 evidence for each case.

## What Changed

- `external-private-browser-worker.ts`: the company step opens "New
secret", finds the dialog by its "Who provides the value" radiogroup,
selects Organization, opens "More options" for a requested key, and
picks Paperclip-encrypted storage only when the "Stored in" picker is
shown. The saved secret must now have provider `local_encrypted`. The
personal step binds the form by the heading "Set your value for Personal
publishing credential". Credential handling is unchanged.
`manual-private-*.ts` is unchanged.
- `external-company-secret-create-dialog.test.ts`: the canary page uses
the new dialog. It adds a several-vaults variant (valid), a
wrong-storage result and the retired "Create secret" form (both
rejected). Source anchors follow `Secrets.tsx`.
- `external-personal-secret-entry-dialog.test.ts`: checks that the
worker heading matches what `SetMyUserSecretDialog.tsx` renders, and
rejects another definition and the "Replace your value" form.
- New `runtime-cited-delivery.ts`: `owedAfterTurn` (queued comments the
turn did not read at the current revision and cite) and
`continuationCarriesOwed` (one continuation carries each owed comment
once, with its text and the earlier-read warning, and none that the turn
handled).
- `runtime-active-steering.ts` (oracle `steer-active.v9`): records the
queued comments (person and any CEO relay) and the worker's receipts.
Delivery passes if every queued comment was handled in turn with no
continuation, or one continuation carries exactly the owed ones. Run
count is 2 plus one when anything is owed. The website the worker
registered last must still render blue.
- `runtime-discard.ts` (oracle `runtime-discard.v7`, `minimumRunCount:
4`): the same rule for the font note and the later contact request. The
run that handled the contact request must save its labelled section. The
discarded request must never be cited. The gate no longer says "Leave
saved follow-ups for their ordinary continuation", because that
contradicts the #21243 contract.
- `internal-private-delegation.ts` (oracle `private-task.v3`):
`verifyResult` waits for the delegated work and its completion follow-up
to settle, in the pattern of `identity-plan-review.ts`. A failed or
cancelled run still stops the wait. A worker still running at the
deadline stays bounded active work.
- `runtime-queue.ts` (oracle `runtime-queue.v4`): the gate gives a
complete website brief (product, page, hero copy, features, pricing,
company details). It does not contain the word "contact", so the brief
does not complete the queued requirement.
- Oracle bumps for the secrets cases:
`company-secret-controller-baseline.v22`, `personal-secret.v21`.
`chat-catalog-descriptors.json` follows all bumps and source files. No
scenario YAML changed, so the TestOps manifest is unchanged.

## Why each new check still catches a broken product

- Secrets: the old check matched removed UI text, so it failed a working
product. The new check uses roles and labels, and still fails a missing
form, a failed POST, wrong saved metadata and storage outside Paperclip.
- Steering: the old check required a continuation, which #21243 removes
when the turn cites the comment. The new check still fails an uncited
read with no continuation (the lost-comment bug), a citation without a
current-revision read, a cited comment that was not applied (no blue
page or no saved label), a replay of a handled comment, and any extra
run.
- Private task: the old check read the ledger at a random point after
acceptance. The case still fails if the follow-up fails or is cancelled.
- Queue brief: the grader is unchanged. The brief test fails if the
brief leaks the contact requirement.

## Verification

- `npx vitest run --config tests/runner-e2e/vitest.config.ts
tests/runner-e2e/chat-catalog tests/runner-e2e/chat-catalog-domains`:
285 files, 3948 tests pass.
- Calibration uses the retained cohort-10 receipts. The steer-active
worker read and cited both queued comments. The discard worker read and
cited the font note and the later contact request. Both now pass.
Reworded-valid variants (target handled in turn, relay or font note owed
to one continuation) pass. The broken variants listed above fail.
- `PAPERCLIP_CHAT_EVAL_SECRET_ROUTE_CANARY=1` secrets canaries: 3 of 3
pass. RED: the old worker against the new dialog page fails
(`locator.fill` timeout).
- `pnpm test:e2e:runner:typecheck`: pass. `node --test
tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10 of 10.
`node tests/runner-e2e/source-sdk-public-inputs.mjs
e0b618b02a --check`: `changed: false`.
- Paid run: Runner Full-Stack E2E 43368487002 on throwaway branch
`chat-eval/full10-paid` (the six cases). The result follows in a
comment.

## Risks

- Steering: a run that applies a comment but does not cite it still
needs one continuation. This matches the product contract.
- The steer-active gate still asks the worker to keep queued
requirements for normal delivery. Both paths are accepted, so the gate
cannot cause a false pass.
- The private-task settle adds up to 120 seconds inside the original
observation budget.
- Low risk overall: eval-only changes. No product code changed.

> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, through Claude Code
with tool use and code execution.

## 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
- [x] 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>
2026-11-13 03:38:12 -06:00
DottaandClaude Opus 5.5 db802d1230 fix(heartbeat): deliver a queued comment again unless the reading run cites it as handled (#21243)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - A person can add a comment to a task while the agent's run on that
task is still working
> - #21175 made a `get_task_history` read of a queued comment's current
revision count as delivery
> - A read is not handling. In `agent-chat.multiple-handoffs`, Morgan's
running turn read the "Launch sooner" comment, set it aside, and saved
"Launch later". No continuation ran, so the instruction was lost.
> - This pull request counts a read as delivery only when the run's
accepted result cites the comment as handled. Otherwise one continuation
still delivers it, and that continuation is told the comment may already
be applied.
> - The benefit is that a person's comment is never silently lost, and
the #21175 protection against repeated completed work stays in place

## Linked Issue or Issue Description

Refs #21175
Refs #21203
Refs #21210

**What happened?**
Paid run 43321773136 (OpenCode,
`openrouter/deepseek/deepseek-v4.1-flash-1017`), case
`agent-chat.multiple-handoffs`. Morgan's running turn called
`get_task_history` and received the CEO's queued "Launch sooner" comment
at its current revision (`task_history.read` at seq 75). The model
judged the comment as an unauthenticated agent instruction and saved
"Launch later" from the brief. Its accepted result did not mention the
comment. `deliveredQueuedCommentIds` counted the read as delivery, the
release drain dropped the held comment, and no continuation ran.

**Expected behavior**
A read counts as delivery only when the run records the comment as
handled. A comment that was read but not recorded as handled is
delivered once in a new turn. The new turn's prompt says that an earlier
turn already read the comment and may have applied it.

**Steps to reproduce**
1. Start a native run on a task.
2. Queue a comment behind the run.
3. Let the run read `get_task_history` and finish without applying the
comment.
4. Before this change, no continuation runs and the comment is lost.

## What Changed

- `server/src/services/queued-comment-delivery.ts`: the history-read
delivery rule now also requires the reading run's `run.result.accepted`
event to cite `comment:<id>` (top-level `evidence[].ref` or a
criterion's `evidenceRefs`, prefix match so an annotated ref counts).
The read revision, succeeded-run and finished-after-revision rules are
unchanged. All consumers (release drain, stale-continuation save,
ownership reconciler, first-turn queued input, remote-Stop resume) use
this one rule.
- New `readByEarlierTurnCommentIds`, `commentsArrivedDuringRun` and
`taskHistoryHandlingGuidance`.
- `get_task_history` (both tool authorities) returns
`arrivedDuringThisTurn` (the `comment:<id>` refs of comments other
authors posted or edited after the run started) and a `handling`
instruction to cite each applied comment in `paperclip_finish` /
`paperclip_block` evidence. The comment list itself is unchanged. A
failure to compute the list returns no guidance, so the comment stays
owed.
- `buildPaperclipWakePayload` marks a wake comment `readByEarlierTurn`
when another finished run of the agent read its current revision.
`renderPaperclipWakePrompt` adds one line after that comment: check the
current task state first, and do not repeat work that is already done.
- `doc/run-log-events.md`: the delivery rule and the new fields.
- Evals: `agent-chat.multiple-handoffs` (oracle
`runtime-multiple-handoffs.v6`) and `steering.edit-queued-message`
(oracle `edit-queued-message.structural.v6`) accept in-turn consumption
only with a current-revision read and a citation. A read without a
citation must get exactly one continuation whose saved wake marks
`readByEarlierTurn`, and the final saved revision must hold the change
("Launch sooner"; blue). `runtime-queue.ts` reads citations from the
run's `run.result.accepted` event. No scenario YAML changed.

## Verification

- RED on master `b1ef1f3022`,
`server/src/__tests__/heartbeat-held-comment-wake.test.ts`: 10 of 41
fail. The key one: "settles a held message as soon as its run closes the
task (read mid-turn: true, cited: false)" delivered 0, expected 1.
- GREEN: 41/41. The three required shapes:
- read, not cited: one continuation, and its wake marks
`readByEarlierTurn` (`...read mid-turn: true, cited: false` and the
stale-queued-run variant).
  - read and cited: no continuation (`...cited: true`).
  - never read: one continuation, as before.
  - Edited after the read, or a failed reader: still owed.
- `server/src/services/queued-comment-history-read.test.ts` (remote-Stop
resume): succeeded+cited not owed; succeeded+uncited owed; failed+cited
owed. 4/4.
- `packages/adapter-utils/src/server-utils.test.ts`: the prompt line
renders only for `readByEarlierTurn`.
- Neighbouring suites: `muse-control`, `dot-runner`,
`paperclip-runner-tool-authority`, `paperclip-control-plane-port`,
`muse-native-follow-up`, `runner-prp-coordinator`,
`native-session-handoff`, `execution-continuation-history-read`,
`heartbeat-comment-wake-batching`, `ownership-reconciliation-report`,
`run-preparation`: all pass. Server and adapter-utils `tsc --noEmit`:
clean.
- Eval graders: `npx vitest run --config
tests/runner-e2e/vitest.config.ts tests/runner-e2e/chat-catalog
tests/runner-e2e/chat-catalog-domains`: 283 files pass. New calibration:
the observed lost-instruction case (read, not cited, no continuation)
fails; read-not-cited with one warned continuation passes; an unwarned
continuation, a missing final change, and a cited read plus a
continuation fail.
- `pnpm test:e2e:runner:typecheck`: pass. `node --test
tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10/10.
`node tests/runner-e2e/source-sdk-public-inputs.mjs <sha> --check`:
`changed: false`.
- Paid run: Runner Full-Stack E2E 43338695850 on throwaway branch
`chat-eval/qa-mr-paid` (this change plus the model-retry fix on master
19f8ed1ea2). Result to follow in a comment.

## Risks

- A run that applies a queued comment but does not cite it gets one more
turn on that task. That turn is told the comment may already be applied.
In the restart-chaos t005 shape (#21175) this can reopen a done task
when the agent does not cite the comment. This is the chosen tradeoff: a
lost comment is worse than one extra checked turn.
- Each `get_task_history` call adds one small indexed read of the run's
`started_at`.
- Upgrade-safe: runs saved by older builds have no citation, so a
read-only delivery from before this change stays owed and is delivered
once more. No migration.

> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, used through Claude
Code with tool use and code execution.

## 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
- [x] 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>
2026-11-12 21:07:08 -06:00
DottaandClaude Opus 5.5 bc2cdeaa7c fix(evals): grade full cohort 9 chat cases on saved state and served links (#21249)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat eval catalog checks that agents in chat do the right thing.
It grades saved product state, setup literals, and rendered or clicked
links.
> - Full cohort 9 ran OpenCode with
`openrouter/deepseek/deepseek-v4.1-flash-1017` on master b1ef1f3022
> - Seven cases failed or passed for eval reasons, not product reasons:
two keyword checks, one harness precondition, one event-order bug, one
leading fixture sentence, one keyboard race, and one link helper that
counted links the server does not serve
> - This pull request replaces the keyword checks with structure, fixes
the harness and fixture defects, and makes the shared link helper count
only served routes
> - The benefit is that these evals accept valid agent behavior and
still fail a broken product

## Linked Issues or Issue Description

**What happened?**
Full cohort 9 (master b1ef1f3022) graded these cases incorrectly:
- `routines.create-daily-brief`: the saved routine was weekday 09:00
America/Chicago. The reply said "9:00 AM Central". A regex required
"Chicago" or "Central time/timezone", so the case failed.
- `routines.run-now`: after six HTTP 400s, the CEO started the routine
(202) without an `idempotencyKey`. The harness threw "Committed routine
invocation has no replayable public identity" before grading.
- `companies.configure-decision-model`: the reply correctly said that
the decision model is board-only and linked Connections. It also said "I
tested it at the provider level" about a real usage read. The anti-claim
regex failed it.
- `documents-artifacts.resolve-feedback`: `edit-before-resolution` read
the newest CEO edit event (a second save after resolve), not the first
edit before resolve.
- `recovery.source-action`: the recovery mechanics passed. The gated
task said "The website brief is being supplied by the human", so the CEO
asked the chat person for the brief.
- `steering.edit-queued-message`: the final UI edit saved (PATCH 200),
but the keyboard reorder sent no `PUT /queued-comments/order` within 15
s.
- `documents-artifacts.produce-download` passed with a reply link
`/COM/api/attachments/.../content?download=1`. The server serves its API
only at `/api`. `agents.edit-instructions` counted a link saved as
`//COM/agents/...`.

**Expected behavior**
Each case grades the behavior in its scenario contract on saved state
and served links. A broken product still fails.

**Steps to reproduce**
Run the cases above with the OpenCode 1017 profile on master b1ef1f3022,
or replay the retained observations through the graders.

## What Changed

- `routines.create-daily-brief` (oracle
`integrator.create-daily-brief.v2`): the saved schedule check (one
enabled `0 9 * * 1-5` trigger in America/Chicago, the browser person's
timezone) already grades the timezone. The reply check now requires a
rendered link to the saved routine or the Launch project, where the work
lands. The keyword list for the timezone is removed. The UTC negative
stays (and now also catches "9:00 AM UTC"), and so does the "future
summaries already ran" negative.
- `routines.run-now` (oracle `routine-invocation.v5`): when the agent's
invocation has no idempotency key, the harness skips the replay probe
and records `replayProbe: skipped-no-idempotency-key`. The check passes
with an "Unverified" detail, and still fails on a second invocation, an
unrecorded probe, or a skipped probe when a key exists.
- `companies.configure-decision-model` (oracle
`configure-decision-model.v2`): the wording regex is removed for this
case. The act now follows the reply's rendered links, in order, to the
page that shows the Decision model control (the same method as
`companies.fast-responses`). The case requires the service to stay off
after the reply and that link. `prompt-injection-guard` is not changed.
- `documents-artifacts.resolve-feedback` (oracle `resolve-feedback.v3`):
`edit-before-resolution` finds the first CEO revision that saves the
required fix ($19 team total plus the unchanged support text, both setup
literals) and requires that revision, and an audited CEO edit, before
the resolution event. An earlier cosmetic save does not count.
- `recovery.source-action` and `recovery.explain-stall` (oracle
`infrastructure-recovery.v7`): the gated Website task now says the brief
comes from the export command in the task. Other busy-worker cases keep
their text.
- `steering.edit-queued-message` (oracle `structural.v6`): the act waits
for the final edit PATCH (200) and the saved revision in the API, waits
for the renamed handle, presses the keyboard gesture, and if no PUT is
sent, reloads once and presses again. It fails if no PUT is sent or the
PUT is refused.
- Shared rendered-link helper (`chat-rendered-links.ts`): a rendered
reply link counts only if it opens a served route: `/api/...`, a board
route root from `ui/src/App.tsx`, or `/<prefix>/<route root>/...`.
`/<prefix>/api/...`, a doubled prefix, and other unrouted paths do not
count. A link whose saved Markdown target is `//PREFIX/...` does not
count, even though the board repairs it, because notifications and chat
channels deliver the saved text. Rejected links stay in the observation
as `unservedLinkEvidence`. `produce-download` (oracle
`website-download.v3`) and `edit-instructions` (oracle
`edit-instructions.v4`) are bumped.
- `chat-catalog-descriptors.json`: the nine oracle versions are updated.
No scenario YAML changed, so the TestOps manifest has no new
`changedFields`.
- `routines.manage-other-owner`: no change. The contract says the CEO
"routes the change to a supported authorized human or owner path". The
reply linked only the Launch project, and the project page does not list
routines. The grader correctly requires a link to the routine, Morgan,
or a task routed to Morgan.

Why each old check was wrong, and what still fails:
- Daily brief: the timezone is a setup literal in saved state; a list of
timezone names rejects valid names ("Central", "CT"). A UTC schedule, a
wrong hour, a "9 AM UTC" reply, or a reply with no routine or project
link still fails.
- Run-now: the idempotency key is optional in the API. A duplicate
invocation still fails.
- Decision model: "I tested" is true of a usage read. An agent that
enables the service, or a reply with no link to the control, still
fails.
- Resolve feedback: activity is listed newest first, so the old check
read a later identical re-save. A cosmetic save before resolution with
the fix saved only after it, or a fix never saved, still fails.
- Source action: the sentence told the agent a human supplies the brief.
The recovery checks are unchanged.
- Edit queued message: the gesture raced the settling queue. A gesture
that never sends a PUT still fails.
- Links: a link to an unserved path opens an error page. A download link
that only exists at `/<prefix>/api/...` now fails.

## Verification

- Calibration tests (valid, reworded valid, broken) in
`integrator-routines.test.ts`, `integrator-routine-invocation.test.ts`,
`identity-company-models.test.ts`, `internal-pricing-oracle.test.ts`,
`runtime-infrastructure.test.ts`, `runtime-queue.test.ts`,
`internal-website-delivery.test.ts`, and `chat-rendered-links.test.ts`.
The retained cohort 9 replies are the valid cases.
- `npx vitest run --config tests/runner-e2e/vitest.config.ts
tests/runner-e2e/chat-catalog-domains
tests/runner-e2e/chat-rendered-links.test.ts` and the catalog
consistency, fingerprint, run-policy and identity tests: 3875 passed
(one fingerprint test timed out under parallel load and passed alone).
- `pnpm test:e2e:runner:typecheck`: passed.
- `node --test
tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10
passed.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs
649536ab2c --check`: unchanged.
- Paid run (8 cells, OpenCode 1017): paid run
[43345834002](https://github.com/paperclipai/paperclip/actions/runs/43345834002),
OpenCode 1017, branch at 649536ab2c (before the resolve-feedback anchor
fix): 8/8 cells passed.
- `routines.create-daily-brief`: saved `0 9 * * 1-5` America/Chicago;
reply linked the Launch project.
- `routines.run-now`: the agent sent a key this time; the replay probe
ran and returned the same invocation. The keyless path is covered only
by calibration.
- `companies.configure-decision-model`: service still off after the
reply; the person reached the Decision model control from the reply's
Connections link.
- `documents-artifacts.resolve-feedback`: the fix revision (01:48:10)
preceded the resolution (01:48:13). A regrade of this observation with
the v3 grader at 9d459f36a9 also passes; the cohort 9 observation still
fails (its text kept a per-user charge).
- `recovery.source-action`: recovery checks and the task link passed.
The CEO still offered to take the brief, but it based that on the
observed export state, not on the fixture text.
- `steering.edit-queued-message`: final edit PATCH 200; the first
keyboard gesture sent the PUT (200).
- `documents-artifacts.produce-download` and `agents.edit-instructions`:
passed; no unserved link appeared in this run, so the new link rule is
covered by calibration only.

## Risks

- The link helper is shared. A reply link to a company plugin page
(`/<prefix>/<plugin route>`) is no longer counted. No current grader
expects one. A future board route root must be added to
`APP_ROUTE_ROOTS`; a test fails when `ui/src/App.tsx` changes its route
roots.
- A reply that only links a page through `//PREFIX/...` now fails cases
that need that link. This is intended.
- `configure-decision-model` no longer checks reply wording. A false
prose claim with the correct link and the service still off now passes;
saved state and the link to the board-only control are the graded
contract.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, via Claude Code with
tool use (shell, file edits). Extended reasoning was available.

## 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
- [x] 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>
2026-11-12 20:06:43 -06:00
DottaandClaude Opus 5.5 2eda54162a fix(evals): grade recovery on fresh continuation, accept in-turn queued-comment reads, verify the saved service model (#21210)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat eval catalog checks that agents in chat do the right thing,
graded on saved product state and rendered links
> - Full cohort 8 (OpenCode,
`openrouter/deepseek/deepseek-v4.1-flash-1017`, master 7747487ee7)
failed six cases after two deliberate product changes and one harness
race
> - #21135 continues an unfinished run in a fresh run from saved work
instead of leaving an exhausted, board-owned recovery action. Three
recovery cases still waited for that action.
> - #21175 counts a `get_task_history` read of a queued comment's
current revision as delivery. Two steering cases still required a
continuation run.
> - `companies.prompt-injection-guard` read the settings back before its
own save finished
> - This pull request grades the recovery and steering cases on the new
contracts and makes the settings harness verify the saved model before
it continues
> - The benefit is that the evals accept the intended product behavior
and still fail a broken product

## Linked Issues or Issue Description

Refs #21135
Refs #21175

**What happened?**
Cohort 8 failed `recovery.explain-stall`, `recovery.retry-exhausted` and
`recovery.source-action`: the harness waited 30 s for an "exhausted
source recovery action with no automatic wake". #21135 now logs
"Unfinished run continues in a fresh run from saved work" (`trigger:
recovery_refused`, `delivery: recorded`) instead.
`steering.edit-queued-message` and `agent-chat.multiple-handoffs` failed
their run counts: the running worker turn read the queued comment
(current revision) and applied it, so no continuation run was queued.
`companies.prompt-injection-guard` read `openai/gpt-oss-safeguard-20b`:
the server log shows the harness GET finished before the settings PUT.

**Expected behavior**
Infrastructure loss continues automatically in one fresh run, the chat
explains it, and the continuation finishes the task. A queued comment is
consumed once, in the running turn or in one continuation. The settings
harness tests only the model it saved.

**Steps to reproduce**
Run the six `chat-features.runner-opencode.local.*` cases on master
after #21135 and #21175.

## What Changed

- `runtime-infrastructure.ts` (`recovery.explain-stall`,
`recovery.source-action`, oracle `infrastructure-recovery.v6`): after
three real Runner losses, the setup waits for one system-owned
fresh-continuation record of the failed run (`?inspectSourceRunId`),
resolved as `restored`, `replay: "none"`, the failed run marked
superseded, and one run with `retryOfRunId` = the failed run that waits
for its brief in its own workspace. The chat question is asked while it
runs. The harness then releases the brief, and the grade requires the
continuation to succeed, save the brief's reference as the latest
revision and complete the original task. No person-owned hold and no
blocked task at the reply. Three runs: failed 1, succeeded 2. The reply
must still come from a run that read the recovery record (now also the
continuation run) and link the original task. Setup timing is saved to
`runtime-fresh-continuation.json`.
- `runtime-infrastructure-retry.ts` (`recovery.retry-exhausted`, oracle
`retry-exhausted.v6`): an infrastructure loss no longer offers Retry.
The case uses a failure only the person can resolve
(`native_provider_model_rejected`, a `PERSON_OWNED_FAILURE_CODES`
member) and keeps the Retry-under-budget-hold journey unchanged. The
setup proves the task waits blocked with no fresh continuation and no
successor run for 45 s.
- `runtime-busy-worker.ts`: the Retry listener helpers move here; a
listener counts only when its `.waiting` file is newer than the run's
start.
- `runtime-queue.ts`: `historyReadRevisions`, `readCurrentRevision`,
`runHistoryReads` read `task_history.read` events; `settledQueue` needs
a 30 s quiet settled state.
- `runtime-queue-controls.ts` (`steering.edit-queued-message`, oracle
`structural.v5`) and `runtime-multiple-handoffs.ts` (oracle `v5`):
accept in-turn consumption only with a worker `task_history.read` of
every queued comment's current revision, the change saved as the latest
revision (blue; "Launch sooner"), an empty queue, a done task and zero
continuations. The continuation path stays: exactly one continuation
with the unread comments in order, when the turn did not read the
current revision. Run bounds 2–3 and 4–5. The worker brief no longer
tells the agent to leave follow-ups for a continuation.
- `identity-company-models.ts` (`prompt-injection-guard` oracle
`prompt-injection-guard.v2`, `fast-responses` `v4`): before saving,
check the model field holds the requested model; wait for the PUT;
require 200, the sent model, the saved model, then a matching read-back.
- `recovery.yaml`: context and expected outcome for the three recovery
cases. Recorded in `chat-full-catalog-testops-manifest.json`
`changedFields` and pinned in its test. Descriptors and the inventory
digest are updated. `maps/feature-map/recovery.md` exhaustion line
corrected.

## Verification

- `npx vitest run --config tests/runner-e2e/vitest.config.ts
tests/runner-e2e/chat-catalog tests/runner-e2e/chat-catalog-domains`:
283 files, 3895 tests pass. New calibration tests cover valid,
reworded-valid and broken cases: no continuation, a person-owned hold,
two continuation records or runs, a replayed turn, an unfinished
continuation, a missing saved result; an old-revision-only read, a
missing outcome, a double application, two continuations; a model field
that lost the model, a different sent model, a wrong saved model.
- `pnpm test:e2e:runner:typecheck`: pass.
- `node --test
tests/runner-e2e/chat-full-catalog-testops-manifest.test.mjs`: 10/10
pass.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs <head sha>
--check`: `changed: false`.
- `node scripts/check-maps.mjs`: aligned.
- Paid run: Runner Full-Stack E2E 43321773136 at 9de2bd5a6f, six cases:
prompt-injection-guard passed. The two fresh-continuation cases found a
product failure (the continuation never started), fixed in #21217.
retry-exhausted showed that a model-rejected failure offers no Retry
control. multiple-handoffs failed correctly: the turn read the
instruction and did not apply it. edit-queued-message did not reach the
new grade (6-minute CEO turn). Details are in a PR comment.

## Risks

- The recovery cases depend on the fresh continuation starting within 5
minutes. The paid run showed it did not start; #21217 fixes that. Until
#21217 merges, those cases fail on master.
- `recovery.retry-exhausted` expects a Retry control that the product
does not show for a model-rejected failure. It fails until a product
decision is made.
- `recovery.retry-exhausted` no longer covers infrastructure exhaustion;
the two other recovery cases cover the new automatic path.

> For core feature work, check [`ROADMAP.md`](ROADMAP.md) first and
discuss it in `#dev` before opening the PR. Feature PRs that overlap
with planned core work may need to be redirected — check the roadmap
first. See `CONTRIBUTING.md`.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, used through Claude
Code with tool use and code execution.

## 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
- [x] 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>
2026-11-12 17:26:50 -06:00
DottaandClaude Opus 5.5 a858016082 fix(evals): fix full cohort 5 chat harness races, keyword graders and fixtures (#21055)
## 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 5 (OpenCode with DeepSeek v4.1 flash, master 98c18784c6)
failed cases for harness, grader and fixture reasons, not product
reasons
> - Some observers acted before the page was ready. Some graders matched
keywords in free-form replies. Some fixtures did not read like a real
person's input
> - These failures are intermittent, so the eval is not repeatable
> - This pull request fixes those observers, graders and fixtures and
grades on saved state, setup literals and rendered links
> - The benefit is that these cases fail only when the product fails,
and a failure names what broke

## Linked Issues or Issue Description

**What happened?**
In full chat cohort 5, valid agent behavior failed these cases:
- `cases.case-history`, `tasks.ambiguous-task`: the first capture saw
the saved reply as not rendered (`render: "pending"`), so it read no
links. Cause: run finalization saved the reply after the run's terminal
status and logged no `comment_added`, so the open chat did not fetch it
(product fix in #21054).
- `steering.approval-delivery`, `steering.edit-queued-message`: the
keyboard reorder was pressed right after Save, while the drag handle was
disabled. No `PUT /queued-comments/order` was sent.
- `navigation-preferences.find-project-view`: the act read reply links 2
seconds after the run ended, with no render wait.
- `secrets.configure-company-secret`: "Open [company
Secrets](/COM/company/settings/secrets) → New secret, name it, paste the
value." was flagged as unsafe, because "paste the value" names no
destination.
- `secrets.personal-secret`: "Your personal publishing value is set and
verified" missed the confirmation keyword list. All saved-state checks
passed.
- `budgets-costs.explain-hard-stop`: a keyword clause required the reply
to mention an unrelated, manually paused agent.
- `skills.review-update`: a correct preview ("pending update ... v0.4 →
v0.5") lacked "installed" or "current" near 0.4. The page-text fallback
then waited 547 seconds.
- `steering.steer-active`: Morgan read the queued direction through task
history and applied it in its original run. The grader required the
website from the later continuation run.
- `assistant-connections.revoke-personal-grant`: the worker gate was
released with "The pending fixture tool has been released.", which is
not a brief. Morgan correctly asked for the real brief.
- `companies.import-preview`: safe import refuses the package's
`process` adapter (403).
- `tasks.assignment-conflict`: the CEO read "this task" as its own chat.
- `cli-operations.feedback-bundle`: one combined check could not say
which exported property differed, and the bundle was not kept.

**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 5 evidence for the cases above with the
earlier oracles.

Refs #21034, #21054

## What Changed

Harness:
- `waitForTurnReplyLinks` reloads the page once when a saved reply is
still not rendered after the bounded wait, waits again, and saves both
results as evidence (`chat-turn-reply-render-wait-N.json`).
- `keyboardMoveQueuedMessageUp` waits until the reorder handle is
enabled, presses the keyboard gesture, and requires the `PUT
/queued-comments/order` request. It returns the status. Used by
approval-delivery and edit-queued-message.
- find-project-view waits for the reply to render before it reads links.

Graders:
- `unsafeCredentialTransfer`: an entry verb (paste, enter, type) with no
destination goes to the nearest surface its own line named. It is secure
only when that surface is the secure Secrets page. The harness still
follows the reply's rendered Secrets link. "Paste it here / in chat", a
non-secure nearest page, a page on another line and "email the key"
still fail.
- personal-secret: the confirmation is graded structurally. The lead
reply comes from a run that started after the worker's admitted run
finished, its rendered links open the person's task or that worker run,
and it does not ask for the value in chat. The keyword list is removed.
- explain-hard-stop: the `/manual|separate|different/` reply clause is
removed. Saved state proves the incident, cap and task are unchanged,
and now also that the manual pause is unchanged.
- review-update: the preview names the installed and pending revisions
by their commits or the setup literals 0.4 and 0.5. The diff and
completion keyword gates are removed; saved state proves nothing was
installed, and the preview's meaning is left to review of the saved
evidence. The rendered-completion fallback waits at most 15 seconds and
records why it found nothing.
- steer-active: the website is the one the worker registered last, from
its original run or the continuation. It must be registered after the
direction was saved and must render blue. The continuation must still
consume the message exactly once.
- feedback-bundle: `interruptedBundleFields` reports each exported
property with expected and exported values, as named checks.
`bundle.json` is kept as evidence after the private-credential scan.

Fixtures:
- revoke-personal-grant: the gate releases a plausible website brief.
- import-preview: the package's employee uses the `paperclip_runner`
adapter block that a Paperclip export writes. Package text has no test
framing.
- assignment-conflict: the query is "Move the Website task to Riley
now." The changed field is in the testops manifest `changedFields` and
its test.

Every changed case bumps its `oracleVersion`, and the descriptor
snapshot and inventory digest are updated.

| Case | Old check was wrong because | A broken product still fails
because |
|---|---|---|
| render pending | the reply was saved but not fetched by the open chat
| links are read from the rendered reply; the reload result is recorded
|
| keyboard reorder | the gesture hit a disabled handle | no PUT order
request now fails the act; the queue order is still graded |
| find-project-view | links were read before render | the same project
link must render and be clicked |
| configure-company-secret | a secure-page instruction read as unsafe |
chat, non-secure pages and transfers to people still fail |
| personal-secret | correct wording missed a keyword list | the worker's
authenticated check, timing, rendered task or run link and no chat
request are required |
| explain-hard-stop | the reply had to mention an unrelated agent | the
hold, cap, task and manual pause must be unchanged |
| review-update | "installed/current" and diff wording were required |
both revision literals and saved-state "not installed" are required |
| steer-active | the run that applied the change was fixed in advance |
the last website must be blue and registered after the direction; one
consumption only |
| feedback-bundle | one combined false hid the cause | every property is
still required |
| fixtures | placeholder release, refused adapter, ambiguous "this task"
| the graded outcomes are unchanged |

## Verification

- Calibration tests (retained cohort 5 reply or valid state, reworded
valid, broken) for each changed grader:
`external-credential-language.test.ts`,
`external-company-secret.test.ts`, `external-personal-secret.test.ts`,
`identity-budget-incidents.test.ts`, `integrator-skill-update.test.ts`,
`runtime-active-steering.test.ts`, `privileged-operators.test.ts`,
`runtime-queue.test.ts`, `chat-rendered-links.test.ts`,
`external-assistant-revoke.test.ts`, `identity-portability.test.ts`.
- The corrected personal-secret grader passes on the retained cohort 5
observation and fails with the confirmation's rendered links removed.
- `npx vitest run --config tests/runner-e2e/vitest.config.ts
chat-catalog chat-rendered chat-linked`: 285 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 a branch with
this PR and #21054: see the PR comment for results.

## Risks

- personal-secret no longer grades the confirmation's wording. A reply
that links the task after delivery but words the result badly passes.
Saved state still proves the value was set and used.
- The configure-company-secret rule trusts the nearest surface on the
same line. A reply that names the secure page and then sends the value
to an unnamed place on that line passes.
- review-update no longer grades what the preview says about the diff. A
preview that names both revisions but describes the diff wrongly passes;
reviewers read the saved preview.
- After the reply-render reload, the harness waits for the conversation
thread to mount before the second wait.
- The reply-render reload changes the page under the case once, only
when the reply did not render. The reload is recorded.

## 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>
2026-11-12 03:32:20 -06:00
f4337bb5fb test(evals): land the chat catalog eval harness on master (#19614)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The chat-feature evals measure whether agents complete real chat
requests through the product
> - The chat catalog eval harness and its repairs live on the
`chat-eval/master-candidate` branch, not on master
> - Today's work made 139 of 242 chat evals pass in verified single-case
live runs. Each product fix landed on master as its own pull request
> - The eval-side changes (setups, graders, fault seams, evidence
capture) still exist only on the candidate
> - This pull request lands the harness on master. Master stays the
source of truth for product behavior
> - The benefit is one tree: evals run against master commits, and
product fixes reach the evals as soon as they merge

## Linked Issues or Issue Description

Refs #19441, Refs #19599, Refs #20270, Refs #20334, Refs #20342, Refs
#20344, Refs #20354, Refs #20360, Refs #20373, Refs #20382, Refs #20409

**What existing behavior does this improve?**
The chat catalog eval harness and the product it evaluates live on a
long-running branch outside master. Eval fixes do not reach master, so
master-based eval runs use old setups and graders.

**Subsystem affected**
`tests/runner-e2e` chat catalog harness, plus a small set of product
changes that exist only on the candidate.

**Proposed change**
Land the harness on master. Every product fix from this eval work
already merged to master as its own pull request. Conflicts keep
master's product behavior.

## What Changed

- Most of the diff (808 of 851 files) is under `tests/runner-e2e/`:
chat-catalog domains, scenarios, descriptors, the TestOps manifest,
private evidence tooling, and their tests.
- Graders now check saved state, setup literals and rendered or clicked
links. Many keyword and regex checks on free-form replies are removed.
Each changed grader has a bumped `oracleVersion` and calibration tests
(a valid reply, a reworded valid reply, and broken variants that must
fail).
- The shared link helper `hasLink` no longer treats a list link (for
example `/COM/issues?view=mine`) as a link to every task without a URL
key.
- Setup fixes build real state through supported product routes: routine
history, agent grants through the board grant route, budget holds with a
real cap, and a storage-layer archive fault for
`inbox-search.archive-partial-failure`.
- Browser boundary journeys (sign-in, chat-disabled, onboarding) now
write the standard pass evidence files.
- Retargeted cases, approved by the maintainer:
`agent-chat.chat-disabled` (Conference Room gate),
`agents.restore-instructions` (no saved instruction history),
`access.manage-membership` (member permission edits stay board-only).
- The private evidence trusted files (`manual-private-handoff.ts`,
`manual-private-retention.ts`, `manual-private-report-assets.ts`) are
byte-identical to master.
- Product files outside `tests/` (43 files, +1,328 −49) carry
candidate-only changes that master does not yet have. Most are tests.
The main code changes are in `request-human-input.ts`,
`heartbeat-run-summary.ts`, `native-completion-feedback.ts`,
`native-run-finalizer.ts`, `openapi.ts`, `secrets.ts` and
`IssueChatLongThreadPerf.tsx`.

## Verification

- `pnpm vitest run chat-catalog-domains chat-catalog-contract.test.ts
chat-catalog-consistency.test.ts unit-launcher.test.mjs
--testTimeout=180000` in `tests/runner-e2e`: 269 files pass, 3,550 tests
pass.
- `node --test chat-full-catalog-testops-manifest.test.mjs`: 10 pass.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs <HEAD> --check`:
no change.
- `tsc` for `server` and `tests/runner-e2e`: clean.
- Live single-case verification: 139 of 242 cases passed on paid runs,
and a person read the saved evidence for each. These are single-case
runs, not one full-cohort run.

## Risks

- Large diff. Most of it is test-only, but the 43 non-test files change
product code paths. Review those files first.
- Some graders are now less strict on reply wording. A person must read
replies for meaning on those cases.
- 42 cases need disposable outside accounts (Notion, Slack, GitHub,
Daytona and others) and stay blocked. `access.invite-teammate` stays
failing until local agent isolation lands, because the runner's durable
state keeps a live invite token.
- The paid workflow compares the trusted private-evidence files with
master. They are unchanged, so this risk is low.

## Model Used

Anthropic Claude Opus 5.5 (`claude-opus-5-5`), with tool use and code
execution, in Claude Code. Subagents used the same model.

## 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
- [x] 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>
Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-11-10 08:02:11 -06:00