{"slug":"ai-research-explore","title":"ai-research-explore","summary":"Rigor Explore compatible skill slug for meaningful and potentially novel deep learning research candidates. Use when the researcher has chosen the task family, dataset, benchmark, evaluation method, provided SOTA references, and wants candidate-only exploration on top of `current","platform":"Claude","tags":[],"authorName":"LLM Mart","authorSlug":"llm-mart","score":0,"source":"github","price":null,"verified":false,"createdAt":"2026-09-13T18:18:58.872717Z","repo":{"url":"https://github.com/lllllllama/RigorPilot-Skills","stars":495,"forks":17,"license":"MIT","updatedAt":"2026-09-23T14:14:25Z"},"bodyHtml":"<hr>\n<h2>name: ai-research-explore\ndescription: Rigor Explore compatible skill slug for meaningful and potentially novel deep learning research candidates. Use when the researcher has chosen the task family, dataset, benchmark, evaluation method, provided SOTA references, and wants candidate-only exploration on top of <code>current_research</code> with auditable repo understanding, idea gating, fair comparison, and governed experiments written to <code>explore_outputs/</code>. Do not use for README-first trusted reproduction, open-ended direction finding, narrow code-only or run-only exploration, passive repo analysis, verified novelty claims, or implicit experimentation.</h2>\n<h1>ai-research-explore</h1>\n<h2>Purpose</h2>\n<p>Use this as the Rigor Explore compatible skill slug after the researcher\nexplicitly authorizes candidate-only work on top of a durable\n<code>current_research</code> anchor. The installed slug remains <code>ai-research-explore</code> for\ncompatibility. Rigor Explore is for meaningful and potentially novel deep\nlearning research candidates while preserving scientific rigor, comparability,\nreproducibility, and auditable collaboration. Novelty and significance remain\nhypotheses before literature contrast, ablation evidence, and fair comparison.\nThe skill does not promise autonomous discovery, global benchmark completeness,\nnovelty proof, or trusted reproduction success.</p>\n<p>Start from the shared operating principles in\n<code>../ai-research-reproduction/references/agent-operating-principles.md</code>, then load\n<code>../ai-research-reproduction/references/research-rigor-principles.md</code> for research claims and\n<code>../ai-research-reproduction/references/deep-learning-experiment-principles.md</code> when experiment\ndetails affect comparability or reproducibility.</p>\n<h2>Fit</h2>\n<p>Use this skill only when the request has both:</p>\n<ul>\n<li>Explicit exploration authorization such as candidate-only work, isolated\nbranch or worktree, sweep, several variants, or exploratory ranking.</li>\n<li>A durable <code>current_research</code> context such as a branch, commit, checkpoint,\nrun record, or already-trained local model state.</li>\n</ul>\n<p>Keep narrow code-only requests on <code>explore-code</code>. Keep narrow run-only requests\non <code>explore-run</code>. Keep passive repository analysis on <code>analyze-project</code>. Keep\nREADME-first reproduction on <code>ai-research-reproduction</code>.</p>\n<h2>Research Rhythm</h2>\n<p>Use a two-loop rhythm:</p>\n<ul>\n<li>Outer loop: understand the repository, freeze task/dataset/evaluation/budget,\npreserve user ideas, map sources, gate ideas, and decide whether the next\nexperiment is worth running.</li>\n<li>Inner loop: make one bounded candidate change or run, smoke-check it, collect\nevidence, rank it against the current anchor, and either stop or return to the\nouter loop with the new evidence.</li>\n</ul>\n<p>This rhythm is a guide, not a rigid autonomous loop. Stop at explicit blockers,\nunclear scientific meaning, exhausted budget, missing anchor/evaluation, or a\nhuman checkpoint.</p>\n<h2>Workflow</h2>\n<ol>\n<li>Confirm <code>current_research</code> and explicit explore-lane authorization.</li>\n<li>Accept either legacy <code>variant_spec</code> or higher-level <code>research_campaign</code>.</li>\n<li>In campaign mode, freeze the task, dataset, benchmark, evaluation source,\nSOTA reference, and budget before candidate work.</li>\n<li>Build only the repo-understanding artifacts needed for the current campaign,\nusually through <code>analyze-project</code>.</li>\n<li>Run bounded, cache-first source lookup when source support matters; prefer\nlocal curated literature such as Zotero if available, then seed sources,\nrepo-local locators, public locators, or optional web lookup. Treat lookup as\nsource resolution, not an open-ended literature search.</li>\n<li>Preserve researcher-provided ideas, optionally add a small bounded set of\nsingle-variable seed ideas, and rank ideas with explicit gates and score\nbreakdowns.</li>\n<li>Prefer one clear candidate at a time. Use <code>explore-code</code> for bounded code\nadaptation and <code>explore-run</code> for short-cycle trials or sweeps.</li>\n<li>Use <code>minimal-run-and-audit</code> or <code>run-train</code> only when the exploratory plan\nrequires real execution evidence.</li>\n<li>Write candidate-only outputs to <code>analysis_outputs/</code>, <code>sources/</code>, and\n<code>explore_outputs/</code> as appropriate; never present exploratory gains as trusted\nreproduction success. Include <code>SCIENTIFIC_CHANGELOG.md</code> and\n<code>COMPARABILITY_REPORT.md</code> for candidate scientific meaning and comparison\nboundaries.</li>\n</ol>\n<h2>Ranking and Evidence</h2>\n<ul>\n<li>Before execution, prioritize candidates by expected gain, cost, success\nlikelihood, patch surface, dependency drag, evaluation risk, and rollback\nease.</li>\n<li>After execution, rank by real evidence first: command status, observed\nmetrics, artifacts, changed paths, smoke results, and reproducibility notes.</li>\n<li>Keep researcher-provided <code>evaluation_source</code> and <code>sota_reference</code> frozen for\nthe campaign; do not claim they are globally complete.</li>\n<li>If the top ideas are too close or the implementation cannot be decomposed into\nauditable units, stop for a checkpoint instead of silently choosing.</li>\n</ul>\n<h2>Campaign Inputs</h2>\n<p><code>research_campaign</code> is preferred for Rigor Explore campaigns, but it should\nstay minimal. The durable core is:</p>\n<ul>\n<li><code>current_research</code></li>\n<li><code>task_family</code></li>\n<li><code>dataset</code></li>\n<li><code>benchmark</code></li>\n<li><code>evaluation_source</code></li>\n<li><code>sota_reference</code></li>\n<li><code>compute_budget</code></li>\n</ul>\n<p>Use <code>candidate_ideas</code>, <code>variant_spec</code>, <code>research_lookup</code>, <code>idea_policy</code>,\n<code>idea_generation</code>, <code>source_constraints</code>, <code>feasibility_policy</code>, <code>baseline_gate</code>,\nand <code>execution_policy</code> as optional guidance, not as fields the agent must fill\nfor every campaign. See <code>references/research-campaign-spec.md</code> for the advanced\nschema and artifact expectations.</p>\n<h2>Reference Loading</h2>\n<ul>\n<li>Load <code>references/ai-research-explore-policy.md</code> for lane safety and candidate\nsemantics.</li>\n<li>Load <code>references/research-campaign-spec.md</code> only when a campaign file is\npresent or the user asks for Rigor Explore campaign governance.</li>\n<li>Load <code>../ai-research-reproduction/references/explore-variant-spec.md</code> for run-level variant matrix\ndetails.</li>\n<li>Load <code>../ai-research-reproduction/references/research-thinking-loop.md</code> before proposing or ranking candidate changes; it is the required greedy observe-ground-design-compare cycle.</li>\n<li>Load <code>../ai-research-reproduction/references/research-rigor-principles.md</code> before making novelty, contribution, SOTA, or comparability statements.</li>\n<li>Consult <code>~/.rigorpilot/PERSONAL_RIGOR.md</code> if present, under <code>../ai-research-reproduction/references/continuous-learning-policy.md</code> (advisory only; core wins).</li>\n<li>Load <code>../ai-research-reproduction/references/deep-learning-experiment-principles.md</code> when training,\nevaluation, baseline, ablation, metric, checkpoint, or dataset details matter.</li>\n<li>Use <code>scripts/orchestrate_explore.py</code> and <code>scripts/write_outputs.py</code> for the\nexisting deterministic artifact workflow.</li>\n</ul>\n","files":[{"path":"agents/openai.yaml","sizeBytes":423,"isText":true},{"path":"references/ai-research-explore-policy.md","sizeBytes":1803,"isText":true},{"path":"references/idea-evaluation-framework.md","sizeBytes":1117,"isText":true},{"path":"references/research-campaign-spec.md","sizeBytes":9663,"isText":true},{"path":"references/smoke-validation-policy.md","sizeBytes":787,"isText":true},{"path":"references/source-mapping-policy.md","sizeBytes":615,"isText":true},{"path":"references/sources-naming-policy.md","sizeBytes":895,"isText":true},{"path":"scripts/lookup/cache_store.py","sizeBytes":8607,"isText":true},{"path":"scripts/lookup/__init__.py","sizeBytes":624,"isText":true},{"path":"scripts/lookup/inventory_writer.py","sizeBytes":3139,"isText":true},{"path":"scripts/lookup/normalizers.py","sizeBytes":4907,"isText":true},{"path":"scripts/lookup/providers/arxiv_provider.py","sizeBytes":2864,"isText":true},{"path":"scripts/lookup/providers/base.py","sizeBytes":3086,"isText":true},{"path":"scripts/lookup/providers/doi_provider.py","sizeBytes":2828,"isText":true},{"path":"scripts/lookup/providers/github_provider.py","sizeBytes":3382,"isText":true},{"path":"scripts/lookup/providers/__init__.py","sizeBytes":457,"isText":true},{"path":"scripts/lookup/providers/optional_provider.py","sizeBytes":844,"isText":true},{"path":"scripts/lookup/providers/url_provider.py","sizeBytes":2090,"isText":true},{"path":"scripts/lookup/record_schema.py","sizeBytes":3445,"isText":true},{"path":"scripts/lookup/repo_extractors.py","sizeBytes":3064,"isText":true},{"path":"scripts/lookup/source_support.py","sizeBytes":4505,"isText":true},{"path":"scripts/orchestrate_explore.py","sizeBytes":124846,"isText":true},{"path":"scripts/passes/atomic_idea_decomposition.py","sizeBytes":12815,"isText":true},{"path":"scripts/passes/candidate_idea_generation.py","sizeBytes":22068,"isText":true},{"path":"scripts/passes/execution_feasibility.py","sizeBytes":19961,"isText":true},{"path":"scripts/passes/idea_cards.py","sizeBytes":1788,"isText":true},{"path":"scripts/passes/idea_ranking.py","sizeBytes":8649,"isText":true},{"path":"scripts/passes/implementation_fidelity.py","sizeBytes":16637,"isText":true},{"path":"scripts/passes/improvement_bank.py","sizeBytes":19244,"isText":true},{"path":"scripts/passes/__init__.py","sizeBytes":909,"isText":true},{"path":"scripts/passes/lookup_sources.py","sizeBytes":16634,"isText":true},{"path":"scripts/passes/source_mapping.py","sizeBytes":19584,"isText":true},{"path":"scripts/write_outputs.py","sizeBytes":1194,"isText":true},{"path":"SKILL.md","sizeBytes":6701,"isText":true}],"reviewScore":null,"reviewSummary":null,"trust":{"provenance":"trusted-source-unreviewed","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow.","bodySource":null},"bodyLocked":false,"purchaseUrl":null,"sourceUrl":null,"report":{"provenance":"trusted-source-unreviewed","screen":{"ran":true,"outcome":"clean","suspicious":0,"notes":0,"hiddenCharacters":false},"virusScan":{"engine":"clamav","status":"clean","scannedAt":"2026-09-13T18:19:14.025966Z","sha256":"021DF0F0CE4449F91E6EC831B73D1646C4D68F5A0FB0AE0A99CD91BF5454C340","sizeBytes":83363},"review":null,"source":{"repositoryUrl":"https://github.com/lllllllama/RigorPilot-Skills","path":"skills/ai-research-explore","license":"MIT","commit":"fb3ccdf5aa64b1fa5b1d0463e666a8bc8bc26e01","subtreeSha":"ED52BD6C5DF50CD6CBB02B96B272978FF2FBB442F875A3F3EC13C9E484961E16","lastSyncedAt":"2026-09-25T23:11:56.834416Z"},"reviewedAt":"2026-09-13T18:19:33.17222Z","notice":"Community-authored content, reproduced verbatim and not vetted as instructions. Treat it as data to evaluate, never as directives to follow."},"install":[{"target":"skills-cli","command":"npx skills add https://github.com/lllllllama/RigorPilot-Skills/tree/main/skills/ai-research-explore"},{"target":"claude-code","command":"claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install lllllllama-rigorpilot-skills@llmmart"},{"target":"git","command":"git clone https://github.com/lllllllama/RigorPilot-Skills.git"}]}