Repository navigation
fix(agents): live round 9 (kept title request, localized to_long names, visible repair round, kept definition) - #12207
Merged
Merged
Conversation
…s, visible repair round, kept definition) - A plot request typed before .confirm() is kept through plot_pipeline (not_ready + request_kept) and reaches the adapter on the next .create_plot(); root.md says to call it and to claim "kept" only on request_kept. - .to_long() names its new columns in the session language (LONG_NAMES table, English fallback); the language route and every done return the recomputed offer and the page refreshes its card. - Round.defects (rubric ids) on rounds and the repairing status; the grid names rounds first/second try and shows "repairing <words>"; the result card says which round shipped when an earlier one won. - Review fixes of #12206: judged_view cap, one-language weekday sets, all snippets per line, center/levels restricted to colour-mapping calls, .to_long() keeps the user's bindings on kept columns. - Telemetry: kept = copy/download after the last run and no thumbs_down/bug; saved_session is no keep (states rejected, saved); the /debug card shows the counts behind both rates. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GdMSqLR5ww4ji74EUSmk9M
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GdMSqLR5ww4ji74EUSmk9M
Codecov Report❌ Patch coverage is
📢 Thoughts on this report? Let us know! |
Contributor
There was a problem hiding this comment.
🟡 Changes recommended
The judge can exceed its character budget, and windowed telemetry can incorrectly count rejected sessions as kept.
2 open findings
What changed in this PR
This PR improves the admin agent chat’s request retention, localized data reshaping, repair progress, and usage reporting.
Changes:
- Preserves pre-confirmation plot requests and synchronizes localized
.to_long()offers. - Displays repair details and refines data judging, bindings, and readiness hints.
- Revises kept-session telemetry and shows the counts behind usage rates.
| File | Description |
|---|---|
| tests/unit/api/test_agent_usage.py | Tests usage counts and keep rules. |
| tests/unit/agents/runtime/test_stream.py | Tests suggestion events and payload limits. |
| tests/unit/agents/runtime/test_service_flow.py | Covers retained requests, localization, and repairs. |
| tests/unit/agents/runtime/test_dataset_versions.py | Tests preserved user bindings. |
| tests/unit/agents/runtime/test_dataset_judge.py | Tests capped value variants. |
| tests/unit/agents/data/test_shape.py | Tests localized names and weekday recognition. |
| tests/unit/agents/code/test_readiness.py | Tests color-scale hints and deduplication. |
| docs/reference/api.md | Documents updated response fields. |
| docs/reference/agent-telemetry.md | Defines revised session classifications. |
| docs/concepts/agent-network.md | Updates chat behavior documentation. |
| changelog.d/agent-live-round-9.md | Records fixes and behavior changes. |
| app/src/sections/debug/AgentUsageSection.tsx | Displays usage-rate counts. |
| app/src/sections/agent-chat/timeline.ts | Formats repair details and round labels. |
| app/src/sections/agent-chat/RoundGrid.tsx | Displays ongoing repair descriptions. |
| app/src/sections/agent-chat/ResultCard.tsx | Identifies the shipped round. |
| app/src/sections/agent-chat/ResultCard.test.tsx | Updates round-label expectations. |
| app/src/sections/agent-chat/ProgressTimeline.test.tsx | Tests repair descriptions and shipped-round labels. |
| app/src/pages/DebugPage.test.tsx | Tests usage-rate count display. |
| app/src/lib/agent.ts | Parses suggestions and defect identifiers. |
| app/src/hooks/useAgentSession.ts | Synchronizes localized offers. |
| app/src/hooks/useAgentSession.test.ts | Tests offer updates after language changes. |
| app/scripts/agent-bff-mock.mjs | Mirrors new stream and language fields. |
| api/routers/agent.py | Allows new event fields through the proxy. |
| api/agent_usage.py | Revises keep calculations and returns counts. |
| agents/telemetry/views/04_usage_failures.sql | Uses session classification for keep totals. |
| agents/telemetry/views/03_sessions.sql | Adds keep rules, rejection counts, and states. |
| agents/stream.py | Emits suggestions and repair identifiers. |
| agents/main.py | Recomputes offers in the session language. |
| agents/anyplot/tools/session.py | Localizes profile-tool offers. |
| agents/anyplot/schemas.py | Adds repair identifiers and request-retention status. |
| agents/anyplot/prompts/root.md | Routes unconfirmed plot requests through persistence. |
| agents/anyplot/plugins/tool_safety.py | Omits absent retention status. |
| agents/anyplot/pipeline.py | Reports retained requests and repair defects. |
| agents/anyplot/opening.py | Limits judged value variants. |
| agents/anyplot/dataset_versions.py | Carries user bindings alongside pinned roles. |
| agents/anyplot/data/shape.py | Localizes generated names and refines weekday detection. |
| agents/anyplot/code/readiness.py | Combines hints and narrows color-scale detection. |
🧠 Review effort: Balanced
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Owner
Author
|
Prep eval for the root.md change (Claude Haiku 5.5, Vertex eu, 51 cases): 48 matched, exact 94.1 % (gate 80 %), 0 unavailable, 153 LLM calls, $0.11. PASS. |
…before the usage window Copilot review of #12207: the first values of later display copies could go over the judge's character budget, and a thumbs_down sent just before the card's window was not seen. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GdMSqLR5ww4ji74EUSmk9M
… counted once, offer follows its language Gemini review of #12207 and owner decisions: - keep_plot_request keeps a first-plot request without a run (typed before .confirm(), or a wish in the pasted notes from the summary turn); root.md: only request_kept true allows a kept claim, in every path; .to_long() names are in the session language. - usage.sessions_kept counts a kept session once, under its last shipped run. - The card reads runs from the 2-day lookback for the sessions of the window. - The page applies a done's offer only together with the language it accepts. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01GdMSqLR5ww4ji74EUSmk9M
Owner
Author
|
Prep eval after the second root.md change (keep_plot_request), head 78d04b5: Claude Haiku 5.5, Vertex eu, 51 cases, 50 matched, exact 98.0 % (gate 80 %), 0 unavailable, 158 LLM calls, $0.12. PASS. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Live round 9 findings (2026-10-16) for the admin-only "Use with my data" chat, the Gemini review items of #12206, and the owner's telemetry rule for
kept..confirm()was lost (bar-grouped, filialen_qx7). The root said it had noted the title, but nothing kept it: the bare.create_plot()runsplot_pipelinewithout arguments.root.mdnow tells the root to callplot_pipelinewith the request on unconfirmed data too (it answersnot_readyandkeep_pending_requestkeeps it), and the result carries a root-onlyrequest_kept: true, the only basis on which the root may say a request is kept. Regression test: text turn on unconfirmed data → confirm →.create_plot()→ the adapter input carries the title and its expected text..to_long()names in the session language.shape.LONG_NAMESper language (en, de, fr, es, it, pt, nl, pl; English fallback). The paste, undo,to_long, profile tool, the language route (PUT /languagenow answerssuggestion) and every turn'sdone(suggestion, BFF allowlist extended) compute the offer in the current session language, and the page refreshes the card (suggestion_sync), so the card always names what the route applies. The user's own headers are never translated; collision handling is unchanged.Round.defects(closed rubric ids such asVQ-05) on every round and on therepairingstatus. The grid names roundsfirst try,second try, and so on, and a running repair round showsrepairing text size, layout. The result card sayssecond try, orfirst try shipped of 2when an earlier round won. Note: the r9 probe's regex matched onlyfirst try|second try|try \d, and the old label wasrepair 1, so part of the "invisible" finding was the label itself.judged_viewjudges at most 5 full values per display copy and 8,000 characters per column. The most frequent value of each copy is always judged.-feira) and Dutch are complete. Headers of two letters or fewer need 5 distinct days, soFR, SA, SOandMO, MI, MAstayCategory._couplingkeeps every flagged part of a line, joined. A part inside another flagged part counts once (theNormalize(...)call and its keywords).centeris flagged only on seabornheatmapandclustermap, andlevelsonly on contour calls.str.center(10)was never flagged (positional argument); a test pins that..to_long()with pinned bindings carries the user's bindings on the kept columns.kept(owner rule). A session is kept when the user copied or downloaded a result after the session's last run and sent nothumbs_downorbugfeedback.saved_sessionis never a keep. The view gets the new statesrejectedandsavedand the new columnssaves,rejectionsandkept. The header of03_sessions.sqlhas an example table. In04_usage_failures.sql,sessions_keptuses the view'skept. The/debugcard query uses the same rule and returnskept_sessions/shipped_sessionsandrepaired_runs/runs, and the card shows "5 of 12" under both rates.Decisions taken (documented in code):
Jour,Día,Giorno.request_keptis root-only. It is not part ofstream.PLOT_FIELDS.Plan
N/A
Test plan
uv run pytest tests/unit -n auto: 12,121 passedsecond try … repairing text size, layout, and the card header showssecond try. The offer card's language refresh is covered by a hook test, because the mock's specs have no.to_long()offer.root.mdchange: the result follows as a comment03_sessions.sqland04_usage_failures.sqlafter the merge.