{"id":51649411,"url":"https://github.com/wd041216-bit/auto-research","last_synced_at":"2026-07-14T02:03:14.387Z","repository":{"id":366137889,"uuid":"1275196086","full_name":"wd041216-bit/auto-research","owner":"wd041216-bit","description":"Evidence-gated research workflows for AI agents: proposal councils, literature triage, claim audits, review packets, and paper packages.","archived":false,"fork":false,"pushed_at":"2026-06-20T11:34:43.000Z","size":52,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-20T13:24:03.237Z","etag":null,"topics":["ai-agents","ai-research","codex-skill","literature-review","paper-writing","reproducibility","research","scientific-discovery"],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/wd041216-bit.png","metadata":{"files":{"readme":"README.md","changelog":"CHANGELOG.md","contributing":"CONTRIBUTING.md","funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":"CITATION.cff","codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-20T11:29:34.000Z","updated_at":"2026-06-20T11:34:31.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/wd041216-bit/auto-research","commit_stats":null,"previous_names":["wd041216-bit/auto-research"],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/wd041216-bit/auto-research","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wd041216-bit%2Fauto-research","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wd041216-bit%2Fauto-research/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wd041216-bit%2Fauto-research/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wd041216-bit%2Fauto-research/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/wd041216-bit","download_url":"https://codeload.github.com/wd041216-bit/auto-research/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/wd041216-bit%2Fauto-research/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35442828,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-07-14T02:00:06.603Z","response_time":114,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-agents","ai-research","codex-skill","literature-review","paper-writing","reproducibility","research","scientific-discovery"],"created_at":"2026-07-14T02:03:13.705Z","updated_at":"2026-07-14T02:03:14.382Z","avatar_url":"https://github.com/wd041216-bit.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Auto Research\n\n**Evidence-gated research workflows for AI agents.**\n\nAuto Research is a Codex skill and standalone research toolkit for turning rough ideas into paper-ready research packages without losing evidence discipline. It gives agents a structured path through proposal councils, literature triage, experiment planning, claim-evidence mapping, simulated review, revision closure, and submission packaging.\n\n\u003e Research agents need judgment, but they also need brakes. Auto Research gives them gates.\n\n## Why This Exists\n\nAI agents can write polished research prose before they have earned the claims. That is dangerous: fake citations, weak novelty, missing baselines, hidden negative results, and \"submission-ready\" drafts that never survived real critique.\n\nAuto Research makes the agent stop at the right places:\n\n- No final claim without traceable evidence.\n- No related work without a literature matrix.\n- No empirical conclusion without a prior experiment plan and results ledger.\n- No top-venue proposal without frontier grounding and unresolved-veto checks.\n- No submission-ready language without review and revision closure.\n\n## What It Does\n\n- **Proposal / Council Gate**: turns rough ideas into scored, adversarially reviewed proposal dossiers.\n- **Literature Protocol**: records search logs, source metadata, triage categories, and literature quality scores.\n- **Experiment Protocol**: requires hypotheses, baselines, metrics, ablations, negative controls, and result ledgers.\n- **Claim-Evidence Audit**: checks whether claims are empirical, literature-backed, theoretical, conjectural, engineering, or negative-result claims.\n- **Review Packet Builder**: packages the current state for skeptical review.\n- **Process Constraints**: defines stage contracts, evidence ID rules, status states, downgrade rules, and allowed gate decisions for every workflow step.\n- **Workspace Templates**: creates a complete paper-package workspace with all required artifacts.\n\n## Quick Start\n\nClone the repository:\n\n```bash\ngit clone https://github.com/wd041216-bit/auto-research.git\ncd auto-research\n```\n\nCreate a research workspace:\n\n```bash\npython3 scripts/init_research_workspace.py \\\n  --mode hybrid \\\n  --title \"My Research Direction\" \\\n  --output ./my-research\n```\n\nValidate gates:\n\n```bash\npython3 scripts/validate_research_gates.py ./my-research --mode hybrid\n```\n\nAn empty workspace should fail. That is the point. Fill artifacts as evidence becomes available, then rerun validators.\n\n## Use As A Codex Skill\n\nCopy or symlink this repository into your Codex skills directory:\n\n```bash\nmkdir -p ~/.codex/skills\nln -s \"$(pwd)\" ~/.codex/skills/auto-research\n```\n\nThen invoke:\n\n```text\nUse $auto-research to turn this research direction into a publication-quality paper package.\n```\n\n## Lifecycle\n\n```text\n0. Proposal / Council Gate\n1. Intake\n2. Research Question\n3. Literature Recall\n4. Literature Triage\n5. Contribution Plan\n6. Experiment / Analysis\n7. Claim-Evidence Mapping\n8. Paper Package\n9. Peer Review \u0026 Revision\n```\n\n## Repository Layout\n\n```text\nSKILL.md                         Codex skill entrypoint\nreferences/                      Protocols and rubrics\nreferences/process-constraints.md Stage and process constraints\nscripts/                         Deterministic validators and workspace tools\nassets/research-workspace/       Blank research package template\nexamples/                        Minimal passing example workspace\nagents/openai.yaml               Skill UI metadata\n```\n\n## Core Commands\n\n```bash\npython3 scripts/check_literature_matrix.py ./my-research\npython3 scripts/audit_claims_evidence.py ./my-research\npython3 scripts/build_review_packet.py ./my-research\npython3 scripts/validate_research_gates.py ./my-research --mode hybrid\n```\n\n## The Council Gate\n\nFor rough original or hybrid research ideas, Auto Research runs a proposal council before letting the idea harden into a contribution plan.\n\nCouncil roles include:\n\n- Chair\n- Domain scientist\n- Methods inventor\n- Data and benchmark specialist\n- Reviewer skeptic\n- Replication engineer\n- Venue strategist\n- Ethics and safety critic\n\nThe council can end as:\n\n- `converged`\n- `checkpoint`\n- `blocked`\n- `killed`\n\nOnly `converged` permits final proposal packaging.\n\n## Validation Philosophy\n\nAuto Research is intentionally strict. A validator failure is a useful research signal, not a bad user experience.\n\nExamples:\n\n- Empty workspaces fail because no evidence exists yet.\n- Empirical claims fail unless they reference `result:` IDs.\n- Literature-backed claims fail unless they reference `lit:` IDs.\n- Converged proposals fail if the debate log still has unresolved vetoes.\n\n## Example\n\nRun validation against the bundled example:\n\n```bash\npython3 scripts/validate_research_gates.py examples/converged-council-hybrid --mode hybrid\npython3 scripts/build_review_packet.py examples/converged-council-hybrid\n```\n\n## Roadmap\n\n- arXiv / DBLP / Semantic Scholar metadata helpers\n- LaTeX compilation helper\n- Citation graph expansion\n- More example research packages\n- Optional multi-agent council orchestration\n- GitHub Action artifacts for review packets\n\n## License\n\nMIT. Use it, fork it, remix it, and make research agents more honest.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwd041216-bit%2Fauto-research","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fwd041216-bit%2Fauto-research","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fwd041216-bit%2Fauto-research/lists"}