[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-79909":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":8,"htmlUrl":8,"language":9,"languages":8,"totalLinesOfCode":8,"stars":10,"forks":11,"watchers":12,"openIssues":13,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":14,"stars7d":15,"stars30d":16,"stars90d":14,"forks30d":14,"starsTrendScore":14,"compositeScore":17,"rankGlobal":8,"rankLanguage":8,"license":18,"archived":19,"fork":19,"defaultBranch":20,"hasWiki":19,"hasPages":19,"topics":21,"createdAt":8,"pushedAt":8,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":14,"starSnapshotCount":14,"syncStatus":16,"lastSyncTime":29,"discoverSource":30},79909,"reviewer","Ingar30\u002Freviewer","Ingar30",null,"Python",92,18,5,3,0,1,2,41.54,"MIT License",false,"main",[22,23,24,25],"codex","economics","paper-review","research-tools","2026-06-12 04:01:25","# Reviewer\n\nA reproducible multi-agent reviewer for academic economics papers. The repository contains the workflow machinery: preprocessing scripts, reviewer prompts, schemas, validation, normalization, editor assembly, tests, and Codex project instructions. It does not include papers or generated review outputs.\n\nThe main entry point is:\n\n```powershell\n.\\.venv\\Scripts\\python.exe scripts\\review_paper.py --pdf \"inputs\\\u003Cpaper_id>.pdf\"\n```\n\nOn macOS\u002FLinux, use `.\u002F.venv\u002Fbin\u002Fpython` instead of `.\\.venv\\Scripts\\python.exe`.\n\n## What This Does\n\nFor a fresh paper, the wrapper:\n\n1. preprocesses the PDF into structured artifacts under `work\u002F\u003Cpaper_id>\u002Fparsed\u002F`\n2. renders run-specific prompts under `work\u002F\u003Cpaper_id>\u002Fprompts\u002F`\n3. runs parser-quality preflight before substantive review\n4. optionally runs an experimental parser repair LLM agent when parser-quality preflight reports high- or medium-severity parser artifacts\n5. dynamically selects optional reviewers while always running mandatory reviewers\n6. validates every reviewer JSON output against schema and semantic checks\n7. normalizes and deduplicates reviewer findings into an editor bundle\n8. builds editor input from the normalized bundle and original reviewer JSON files\n9. runs the editor to write `outputs\u002F\u003Cpaper_id>\u002Freport.md`\n10. smoke-checks final report structure and traceability\n\nOnly the project machinery is meant to be shared on GitHub. Source PDFs, parsed artifacts, reviewer logs, and final reports are local\u002Fprivate by default.\n\n## Quick Start\n\n### 1. Get The Repository\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FIngar30\u002Freviewer.git\ncd reviewer\n```\n\nGit is convenient for cloning and contributing, but it is not required to run the reviewer. You can also download the repository as a ZIP from GitHub and open a shell in the extracted folder.\n\n### 2. Install Prerequisites\n\nYou need:\n\n- Python 3.12 or newer\n- Codex CLI installed and authenticated\n- access to the model\u002Fsearch features needed by your reviewer configuration\n\n### 3. Set Up Python\n\nWindows PowerShell:\n\n```powershell\n.\\setup.ps1\n.\\.venv\\Scripts\\Activate.ps1\n```\n\nmacOS\u002FLinux:\n\n```bash\nbash setup.sh\nsource .venv\u002Fbin\u002Factivate\n```\n\nManual setup is also fine:\n\n```powershell\npython -m venv .venv\n.\\.venv\\Scripts\\python.exe -m pip install --upgrade pip\n.\\.venv\\Scripts\\python.exe -m pip install -r requirements.txt\n```\n\n### 4. Check The Install\n\n```powershell\n.\\.venv\\Scripts\\python.exe -m unittest\n.\\.venv\\Scripts\\python.exe scripts\\check_environment.py\n```\n\n### 5. Add A Paper Locally\n\nPut a source PDF in `inputs\u002F`. Files in `inputs\u002F` are ignored by Git.\n\n```text\ninputs\u002Fmy-paper.pdf\n```\n\n### 6. Run A Review\n\n```powershell\n.\\.venv\\Scripts\\python.exe scripts\\review_paper.py --pdf \"inputs\\my-paper.pdf\"\n```\n\nThe final report will be written to:\n\n```text\noutputs\u002Fmy-paper\u002Freport.md\n```\n\nThe intermediate parsed artifacts, prompts, logs, reviewer outputs, selection output, and editor bundle will be written to:\n\n```text\nwork\u002Fmy-paper\u002F\n```\n\n## Repository Map\n\nTracked project machinery:\n\n- `AGENTS.md`: Codex-facing workflow and safety instructions.\n- `.codex\u002Fconfig.toml`: project-level Codex defaults.\n- `.agents\u002Fskills\u002Fpaper-reviewer\u002FSKILL.md`: reusable workflow playbook.\n- `config\u002Freviewers.json`: enabled reviewer roster and reviewer metadata.\n- `prompts\u002Ftemplates\u002F*.txt`: reusable prompt templates.\n- `schemas\u002F*.json`: structured output contracts.\n- `scripts\u002F*.py`: deterministic preprocessing, validation, orchestration, normalization, and report checks.\n- `scripts\u002Fpipeline_paths.py`: shared runtime path conventions for wrappers and forked workflows.\n- `tests\u002F`: focused unit tests for reviewer config, validation, normalization, editor brief behavior, and report checks.\n- `.github\u002F`: CI, issue templates, and pull request template.\n- `.github\u002Fdependabot.yml`: weekly dependency checks for GitHub Actions and Python requirements.\n- `setup.ps1` and `setup.sh`: local bootstrap helpers.\n- `scripts\u002Fcheck_environment.py`: fast local readiness check for dependencies, project files, and Codex CLI.\n- `scripts\u002Fcheck_tracked_sensitive_names.py`: pre-push scanner for unexpected sensitive variable names in shareable files.\n- `docs\u002Ffirst_review_walkthrough.md`: step-by-step path for a new user running a first private review.\n- `docs\u002Fextension_guide.md`: reviewer and wrapper extension points for forks.\n- `docs\u002Frepository_settings.md`: recommended GitHub settings for public or private repository use.\n\nLocal\u002Fprivate runtime locations:\n\n- `inputs\u002F`: source PDFs.\n- `work\u002F\u003Cpaper_id>\u002Fparsed\u002F`: parsed page text, page images, inventories, tables, figures, citations, crossrefs, and manifest files.\n- `work\u002F\u003Cpaper_id>\u002Fprompts\u002F`: rendered run-specific prompts.\n- `work\u002F\u003Cpaper_id>\u002Frepair\u002F`: optional parser repair plan, reviewer-facing repair notes, repair manifest, and repaired overlay artifacts.\n- `work\u002F\u003Cpaper_id>\u002Fselection\u002F`: reviewer selector output and selected reviewer roster.\n- `work\u002F\u003Cpaper_id>\u002Freviews\u002F`: reviewer JSON outputs.\n- `work\u002F\u003Cpaper_id>\u002Feditor\u002F`: normalized bundle and editor input.\n- `outputs\u002F\u003Cpaper_id>\u002Freport.md`: final human-readable report.\n\nPrivate papers and generated review artifacts are local by default. Do not commit source PDFs, `work\u002F` artifacts, `outputs\u002F` reports, logs, rendered prompts, reviewer JSON, or credentials. See `SECURITY.md` and `docs\u002Fpublic_release_checklist.md` for the full release checklist.\n\n## Open Development\n\nThis project is intended to support reproducible AI-assisted paper-review workflows without publishing the papers being reviewed. Issues, pull requests, examples, and tests should use synthetic fixtures, public-domain examples, or short non-sensitive snippets rather than private manuscripts or generated review outputs.\n\nUseful contributions include:\n\n- better deterministic preprocessing and artifact inventories\n- reviewer prompts, schemas, validators, and normalization rules that improve traceability\n- tests that capture parser, reviewer-selection, editor, or privacy-hygiene failures\n- documentation for running the workflow on new platforms or adapting it to related review settings\n\nForks can usually extend the workflow by adding reviewer entries in `config\u002Freviewers.json`, prompt templates in `prompts\u002Ftemplates\u002F`, and matching validation or normalization tests when the output contract changes. Shared runtime paths live in `scripts\u002Fpipeline_paths.py` so wrappers can reuse the same `inputs\u002F`, `work\u002F`, and `outputs\u002F` layout.\n\nSee `docs\u002Fextension_guide.md` for the main reviewer, schema, prompt, normalization, and wrapper extension points.\n\nSee `CONTRIBUTING.md` for pull request expectations and local checks.\n\n## Reviewer Roster\n\nReviewers are configured in `config\u002Freviewers.json`. Each entry declares:\n\n- reviewer name\n- prompt template\n- output filename\n- finding ID prefix\n- whether search is required\n- normalization role\n- stage: `preflight` or `review`\n- selection policy: `mandatory` or `optional`\n\nMandatory reviewers always run:\n\n- `parser_quality_auditor`: preflight check for parser artifacts that could poison downstream review\n- `crossref_auditor`: internal reference, numbering, and appendix-label checks\n- `reference_auditor`: bibliography and cited-reference verification\n- `grammar_auditor`: copyediting and grammar issues\n\nAfter `parser_quality_auditor`, an optional parser-repair step can be enabled. It adds repair guidance and narrow overlay artifacts that help reviewers avoid unsafe parsed tables, figures, or captions, but it adds runtime and token usage and is off by default. To run the review with the parser repair overlay enabled, use the following command:\n\n```powershell\n.\\.venv\\Scripts\\python.exe scripts\\review_paper.py --pdf \"inputs\\my-paper.pdf\" --parser-repair overlay\n```\n\nOptional reviewers are selected dynamically by default:\n\n- core substantive reviewers: `numerical_auditor`, `claim_evidence_auditor`, `literature_auditor`, `identification_auditor`, `robustness_auditor`, `sample_construction_auditor`, `abstract_conclusion_consistency_auditor`, `limitations_external_validity_auditor`, `model_equation_auditor`, and `data_availability_replication_auditor`\n- narrower pilot reviewers: `institutional_context_auditor`, `power_multiple_testing_auditor`, `design_randomization_auditor`, and `economic_magnitude_auditor`\n\nUse dynamic selection for normal runs. Use static mode only when all enabled review-stage reviewers should run.\n\nSearch-enabled reviewers require Codex search mode. Literature and reference verification should not be guessed; use `cannot_verify` when evidence is missing.\n\nRun all enabled review-stage reviewers without selector filtering:\n\n```powershell\n.\\.venv\\Scripts\\python.exe scripts\\review_paper.py --pdf \"inputs\\my-paper.pdf\" --reviewer-selection static\n```\n\nUse an explicit paper id when needed:\n\n```powershell\n.\\.venv\\Scripts\\python.exe scripts\\review_paper.py --pdf \"inputs\\my-paper.pdf\" --paper-id \"my-custom-id\"\n```\n\n## License\n\nMIT License. See `LICENSE.md`.\n","Reviewer 是一个用于学术经济学论文的可复现多代理审稿工具。该项目使用Python编写，结合了Codex等技术，能够自动处理论文PDF文件，生成结构化数据，并通过一系列预设的审稿流程对论文进行评估，最终生成审查报告。核心功能包括论文解析、动态选择审稿人、验证和标准化审查结果以及编辑报告生成。适合需要快速且一致地评审大量学术论文的研究机构或个人使用。项目遵循MIT许可协议开放源代码，但实际使用的论文及其审查输出默认保持私有。","2026-06-11 03:58:29","CREATED_QUERY"]