增强 Auto Research 全流程、三模型审稿与 Teaser 生成 - #3
Merged
Merged
Conversation
Use independent maximum-reasoning reviewers on an isolated rendered artifact so feedback reflects the actual submission. Co-authored-by: Cursor <cursoragent@cursor.com>
Review compiled PDFs with a Cursor model panel
Gate every reviewer turn on a complete rendered paper and return deterministic failures to the author instead of spending review on unfinished work. Co-authored-by: Cursor <cursoragent@cursor.com>
Require submission-ready papers before review
Integrate the Research Factory, split paper workspace, figure skills, cost tracking, and adaptive loop controls while preserving the fixed three-model PDF reviewer panel and readiness gate. Co-authored-by: Cursor <cursoragent@cursor.com>
Rename the results and teaser variants consistently so author prompts and documentation expose unambiguous 1/2 choices. Co-authored-by: Cursor <cursoragent@cursor.com>
Expose all four AR figure examples from one Markdown page without duplicating image assets. Co-authored-by: Cursor <cursoragent@cursor.com>
Preserve the reference assets, generated candidates, review notes, and script snapshots in a self-contained gallery for comparison and reuse. Co-authored-by: Cursor <cursoragent@cursor.com>
Honor persisted work/paper manuscript locations so upgrades keep existing PDFs, builds, readiness checks, and reviews accessible. Co-authored-by: Cursor <cursoragent@cursor.com>
Provide a tested controlled restart workflow that retains auth, Turbogate identity, tmux agents, and disk-backed task state without exposing secrets. Co-authored-by: Cursor <cursoragent@cursor.com>
Add teaser-figure-3 with a semantic-blueprint and correction workflow, and make Auto Research invoke it proactively whenever a page-one teaser needs creation or refresh. Co-authored-by: Cursor <cursoragent@cursor.com>
Integrate the staged Research Factory, grounded knowledge graph, paper metadata cache, and exact agent completion hooks while preserving the fixed three-model PDF reviewer panel and AR readiness workflow. Co-authored-by: Cursor <cursoragent@cursor.com>
Persist each model's review separately and render all three reports in Loom and Research Factory while clearly marking the lowest-scoring final verdict. Co-authored-by: Cursor <cursoragent@cursor.com>
JamesSand
changed the title
Remove trailing Markdown whitespace so the pull request passes the repository diff hygiene check. Co-authored-by: Cursor <cursoragent@cursor.com>
There was a problem hiding this comment.
Pull request overview
This PR upgrades Loom’s Auto Research (AR) paper workflow by adding a deterministic “ready-to-submit” gate before reviews, switching paper reviews to a fixed three-model Cursor PDF reviewer panel with per-model persistence, and expanding figure/teaser generation workflows (including a new default teaser skill). It also adds a developer-only hot-restart skill to redeploy Loom while preserving tunnel URL and auth environment, and updates both Loom’s main UI and Research Factory to display per-model reviewer results.
Changes:
- Add a hard Review Readiness Gate that blocks reviewer calls until the compiled PDF is complete (no placeholders, missing figures, unresolved references, etc.).
- Replace single-reviewer flow with a fixed three-model Cursor reviewer panel, persisting individual model reports and updating UI/API payloads accordingly.
- Introduce/rename AR figure skills (including default
teaser-figure-3) plus add a hot-restart developer skill and supporting tests/docs.
Reviewed changes
Copilot reviewed 28 out of 57 changed files in this pull request and generated 4 comments.
Show a summary per file
| File | Description |
|---|---|
| zhizhou-note/figure-redraw-v2/scripts/when-to-weights-teaser-v2.py | Adds a versioned teaser rendering script for the “when to write” figure candidate set. |
| zhizhou-note/figure-redraw-v2/scripts/when-to-weights-streaming-gate-v2.py | Adds a deterministic streaming-gate diagram script with render audits. |
| zhizhou-note/figure-redraw-v2/scripts/self-distillation-teaser-v2.py | Adds a framework teaser script driven by committed experiment artifacts. |
| zhizhou-note/figure-redraw-v2/FIGURE_REDRAW_V2.md | Documents the figure-redraw candidate set, verification notes, and manual review outcomes. |
| zhizhou-note/figure-redraw-v2/CURSOR_IMAGE_GENERATION.md | Documents Cursor GenerateImage usage and constraints for reproducible image generation. |
| tests/test_review_panel_ui.py | Tests per-model review persistence and API payload recovery for combined panel files. |
| tests/test_hot_restart_skill.py | Tests hot-restart behavior, including dry-run safety and secret non-disclosure. |
| tests/test_ar_task.py | Adds extensive coverage for readiness gating, reviewer panel isolation, plateau logic, and legacy layout handling. |
| README.md | Updates AR pipeline documentation to reflect readiness gating and the three-model panel behavior. |
| pyproject.toml | Adds pypdf dependency used by readiness checks to inspect rendered PDFs. |
| loom/web.py | Implements reviewer panel persistence/API payloads, readiness-gate loop behavior, plateau pausing, and updates review endpoint responses. |
| loom/web_static/factory.js | Updates Research Factory UI to show per-model reviewer cards and full per-model reports in the modal. |
| loom/web_static/factory.css | Styles the reviewer card grid and per-model report sections in Research Factory. |
| loom/web_static/app.js | Updates Loom main UI to display readiness status and per-model reviewer cards/reports. |
| loom/web_static/app.css | Adds styling for reviewer cards/reports in Loom main UI. |
| loom/skills/dev/loom-hot-restart/SKILL.md | Adds documentation and safety rules for hot restart (preserve tunnel, secrets, tasks). |
| loom/skills/dev/loom-hot-restart/scripts/hot_restart.py | Adds the Linux-only hot-restart helper implementation. |
| loom/skills/ar/figures/teaser-figure-3/SKILL.md | Adds the default AI-based teaser workflow (Cursor GenerateImage) with semantic-blueprint safeguards. |
| loom/skills/ar/figures/teaser-figure-3/PROMPT_TEMPLATE.md | Provides prompt templates and correction workflow for teaser-figure-3. |
| loom/skills/ar/figures/teaser-figure-2/SKILL.md | Renames/updates the “plain” deterministic teaser skill to teaser-figure-2. |
| loom/skills/ar/figures/teaser-figure-2/scripts/plain_style.py | Adds the drawing kit for the plain teaser style. |
| loom/skills/ar/figures/teaser-figure-2/scripts/overview_style.py | Adds shared deterministic overview drawing utilities for teaser figures. |
| loom/skills/ar/figures/teaser-figure-2/example.py | Adds a runnable example for teaser-figure-2. |
| loom/skills/ar/figures/teaser-figure-1/SKILL.md | Renames/updates the tinted deterministic teaser skill to teaser-figure-1. |
| loom/skills/ar/figures/teaser-figure-1/scripts/overview_style.py | Adds the drawing kit for tinted teaser style. |
| loom/skills/ar/figures/teaser-figure-1/example.py | Adds a runnable example for teaser-figure-1. |
| loom/skills/ar/figures/results-figure-2/SKILL.md | Renames/updates replicate-focused results plotting skill to results-figure-2. |
| loom/skills/ar/figures/results-figure-2/scripts/replicate_style.py | Adds replicate plotting helpers (strip plots, vrefs, stat blocks, legends). |
| loom/skills/ar/figures/results-figure-2/scripts/plot_style.py | Adds shared results plotting rcParams and PDF output checks. |
| loom/skills/ar/figures/results-figure-2/example.py | Adds a runnable example for results-figure-2. |
| loom/skills/ar/figures/results-figure-1/SKILL.md | Renames/updates baseline results plotting skill to results-figure-1. |
| loom/skills/ar/figures/results-figure-1/scripts/plot_style.py | Adds shared results plotting style and PDF output validation. |
| loom/skills/ar/figures/results-figure-1/example.py | Adds a runnable example for results-figure-1. |
| loom/skills/ar/figures/results-figure-1/example_data.json | Adds example data for results-figure-1’s runnable demo. |
| loom/skills/ar/figures/display.md | Adds a lightweight index page that displays figure skill examples by relative path. |
| loom/skills/ar/AR-REVIEWER.md | Updates reviewer instructions to enforce “PDF-only” review and page-based location references. |
| loom/skills/ar/AR-AUTHOR.md | Updates author methodology to enforce readiness completion and proactive teaser generation. |
| FIGURE_SKILL_FLOW.md | Documents end-to-end figure skill discovery → author usage → readiness gate → reviewer checks flow. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Comment on lines
+4520
to
+4524
| attempts.append( | ||
| { | ||
| "attempt": attempt_n, | ||
| "ended_at": _iso_now(), | ||
| "note": str(attempt_note), |
| HERE = Path(__file__).resolve().parent | ||
| CODE = HERE.parent | ||
| RESULTS = CODE / "results" / "results.json" | ||
| OUT = HERE.parents[3] / "figure-redraw-v2" |
|
|
||
|
|
||
| HERE = Path(__file__).resolve().parent | ||
| OUT = HERE.parents[3] / "figure-redraw-v2" |
|
|
||
|
|
||
| ROOT = Path(__file__).resolve().parents[3] | ||
| OUT = ROOT.parent.parent / "figure-redraw-v2" |
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.
概要
该 PR 将
zhizhou-dev中完整验证过的 Auto Research 增强合并到main，覆盖三模型 PDF 审稿、投稿就绪门控、默认 AI Teaser、旧任务兼容、Hot Restart，以及两套前端的 Reviewer 结果展示。主要更新
三模型 PDF Reviewer Panel
lowest · final。投稿就绪门控与循环控制
Figure 与 Teaser 工作流
teaser-figure-3，Auto Research 在需要 Teaser、Figure 1、architecture 或 pipeline 图时主动使用 Cursor GenerateImage / Nano Banana 工作流。兼容性与运维
work/paper/目录，恢复 PDF 下载、构建、Readiness 和 Reviewer 能力。.RUD状态。合并就绪状态
origin/main已完整包含在zhizhou-dev中251 passedgit diff --check通过app.js与factory.js语法检查通过推荐合并方式
使用 Merge commit。当前分支已包含最新
main，无需 rebase 或 force push；该方式能够保留已经验证过的 commit 历史。合并后操作
合并并部署后，对正在运行的 Loom 8766 执行保留域名和认证信息的 Hot Restart，然后复查 PDF API 与两套 Reviewer 前端。