mirror of
https://github.com/paperclipai/paperclip.git
synced 2026-11-13 15:55:46 +01:00
f4337bb5fb
## 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>
108 lines
6.6 KiB
Markdown
108 lines
6.6 KiB
Markdown
# Preserved chat catalog source
|
|
|
|
This branch consolidates reviewed eval changes for all **244** frozen scenarios:
|
|
242 lazy catalog entries and the two existing standalone chat flows. It preserves
|
|
source implementations, controls and evidence contracts. It does not establish
|
|
that all scenarios are ready to run or that this composed head has passed a live
|
|
campaign. The machine-readable inventory is `chat-catalog-preservation.json`.
|
|
|
|
The production baseline remains `39e811c9c5226979fce42e052a27cfb28301018a`.
|
|
No production, workflow, lockfile or SDK source change is included. The real
|
|
new-company, formal-hire and routine-history definitions retain prerequisites
|
|
18103, 18105 and 18033 respectively. Those focused fixes are excluded here; their
|
|
candidate outcomes remain separate from frozen-baseline qualification.
|
|
|
|
Composition preserves the per-scenario descriptor union, both independent SDK
|
|
privacy and new-company security sections, the final formal-hire clock module,
|
|
and the stronger shared Notes provider used by status cards and routine history.
|
|
The accepted original-deadline and multipart controls are included, as is the
|
|
new-company successor's original phase deadline and remaining-time wizard clamp.
|
|
Frozen queries, contexts, outcomes, turn/human counts and overall caps are retained.
|
|
|
|
The accepted SDK2 overlay binds least-scope setup to the actual public Agent
|
|
lifecycle/version and a new real plugin resume receipt within the existing
|
|
90-second wait. Managed-routine setup identifies only its saved routine and
|
|
plugin through ordinary conversation context and independent CEO readback;
|
|
structural JSON comparisons ignore object key order while preserving all values,
|
|
arrays, ownership and explanation checks. A pending question still fails the
|
|
required outcome.
|
|
|
|
The task prerequisite overlay binds two genuine ready owners to their completed
|
|
public manual pauses, exact versions and holds before inventory seeding. It also
|
|
adds the owned browser Origin to the existing human attachment upload, preserving
|
|
the real board mutation guard and exact company/task/comment/byte checks. Neither
|
|
overlay changes canonical messages, native/human counts, overall caps or an
|
|
exhausted case's result. The production core fixes remain excluded.
|
|
|
|
The document and decision-audience overlay preserves exact public document
|
|
identity, revision and bytes through the shipped Plans control, and reads a
|
|
pending addressed decision through the actual authenticated recipient's feed.
|
|
It requires both the owner and recipient identities. Its measured final failures
|
|
remain unchanged and exhausted; source preservation gives them no new cycle.
|
|
|
|
The outbound-email observer accepts ordinary standing-chat work with a null
|
|
parent only when the canonical source run, creator, public audit and real native
|
|
assignment admission agree. It also preserves the supported public issue-create
|
|
wake shape. The email task still links explicitly to that work and the exact
|
|
owned thread/endpoint; original draft, recipient, native send, provider receipt,
|
|
duplicate and truthful delivery checks remain. Actual authorized AgentMail
|
|
resources and input delivery are separate prerequisites, with no source-only
|
|
runtime pass or implicit credential-transfer permission.
|
|
|
|
Readiness is separate from importability. Declared unsupported fresh contexts,
|
|
public capability/profile gaps and external resource/authorization requirements
|
|
remain explicit. The preserved plugin configure/disable fixture still has a known
|
|
secret-ref schema mismatch: it sends an object where its public string schema
|
|
requires a secret UUID. Disable's returned final reached that setup request and
|
|
received HTTP 400. Configure's returned final stopped earlier while waiting for
|
|
the initial chat response; its later UI/configuration leg was not exercised, so
|
|
the schema mismatch there is a source finding. Both spent raw failures remain
|
|
unchanged and are not rerun by this consolidation. Other measured failures retain
|
|
their actual interpretation.
|
|
The legacy Projects manual-order surface required by sidebar-order is absent
|
|
from both frozen-production layouts; this source does not revive it or inject a
|
|
classic interface to manufacture the frozen context.
|
|
|
|
Every historical result remains attached to its original source revision, full
|
|
fingerprint, provider/model/build/database evidence, ordinal, usage, cost, privacy
|
|
and cleanup. The integrated source changes broad fingerprint closures; none of
|
|
the older qualified passes is transferred to this head. No attempt budget is reset,
|
|
and no live launch or publication authority comes from this source preservation.
|
|
|
|
|
|
## Integration candidate, 2026-10-31
|
|
|
|
This owned integration branch combines preserved source `e2b35be76f` with
|
|
master `51ea624344`. The original preservation branch and its historical results
|
|
remain unchanged. No historical pass transfers to this composed source.
|
|
The frozen 244-case OpenCode catalog keeps its historical IDs and DeepSeek 1017
|
|
model. The two shipped subscription aliases and the separate TestOps backlog cell
|
|
remain selectable alongside it; they do not change the frozen denominator.
|
|
The standalone aliases retain the current shipped 120-second turn limit.
|
|
Catalog phase and attempt budgets remain separately declared by their fixtures.
|
|
|
|
The integration compatibility checks distinguish the 244 historical cells from
|
|
the three shipped aliases. Fixtures implement current run-observation and
|
|
conversation-link custody contracts. The standalone flows keep shipped timeout
|
|
behavior; catalog-native setup still verifies its separately declared policy.
|
|
Synthetic public-schema tests hash their copied synthetic inputs, while the
|
|
production frozen source manifest and all private-value rejection checks remain
|
|
unchanged. No SDK runtime qualification follows from those synthetic tests.
|
|
|
|
The refreshed base includes current master's committed Agent Chat reply recovery
|
|
and durable retry deadlines. It also contains event-driven issue/receipt recovery
|
|
and the guard against a native run changing its own task status. These are
|
|
upstream product changes, not qualifications
|
|
of the preserved scenarios. Fresh pilots remain bound to their exact source.
|
|
|
|
## TestOps completion pilot
|
|
|
|
The dedicated main-only `chat-completion-testops.yml` workflow may select only
|
|
`projects.project-work`, `projects.project-defaults`, and
|
|
`instance-operations.restart-service` on the native OpenCode 1017 profile.
|
|
It needs the TestOps admission marker, a real job-run identity, attempt one,
|
|
and the same source-bound AWS host verification as the existing campaign.
|
|
The entry point preserves original queries, grading, clocks and zero retries.
|
|
Registration is separate from a real TestOps pass and verified result upload.
|
|
The broader catalog remains on this integration branch.
|