zjunlp
- 345 skills
- 0 followers
- 6 days ago last updated
- ▌ Highest Cost Scenario Identification 6 · zjunlpSolves "highest cost scenario identification" questions in the dabstep payment processing dataset. Use this skill when the question asks about the most expensive MCC (Merchant Category Code) for a given transaction amount, or the most expensive ACI (Authorization Characteristics Indicator) for a given card scheme and transaction amount. Also triggers for questions like "which scenario incurs the highest fee", "most costly MCC in general", or "which ACI is most expensive for [card_scheme] credit transactions". Always use this skill for dabstep tasks involving fee maximization or identifying worst-case cost scenarios.
- ▌ Highest Cost Scenario Identification 7 · zjunlpIdentify the most expensive MCC (Merchant Category Code) or ACI (Authorization Characteristics Indicator) for a given transaction scenario in the dabstep payment processing dataset. Use this skill when asked which MCC or ACI results in the highest fee for a specific transaction amount, card scheme, and/or credit/debit type. Always apply this skill for questions about "most expensive", "highest cost", or "maximum fee" scenarios involving MCCs or ACIs.
- ▌ Sec 10k Company Analysis 2 · zjunlpComprehensive analysis of public companies using SEC EDGAR 10-K financial data stored in a SQLite database. Use this skill whenever the task involves analyzing a company by CIK (Central Index Key), querying SEC financial data, exploring a financial database with tables like companies, filings, financial_facts, or producing a structured financial analysis report. Covers schema navigation, metric discovery, industry-specific exploration, and producing a complete financial summary.
- ▌ Sec 10k Company Analysis 3 · zjunlpComprehensive analysis of public companies using SEC EDGAR 10-K financial data stored in a SQLite database. Use this skill whenever the task involves analyzing a company by CIK (Central Index Key), querying SEC financial data, exploring a financial database with tables like companies, filings, financial_facts, or producing a structured financial analysis report. Covers schema navigation, metric discovery, industry-specific exploration, multi-year trend analysis, capital returns analysis, long-term obligations, and producing an insightful financial summary that connects metrics across dimensions.
- ▌ Sec 10k Company Analysis 4 · zjunlpComprehensive analysis of public companies using SEC EDGAR 10-K financial data stored in a SQLite database. Use this skill whenever the task involves analyzing a company by CIK (Central Index Key), querying SEC financial data, exploring a financial database with tables like companies, filings, financial_facts, or producing a structured financial analysis report. Covers schema navigation, metric discovery, industry-specific exploration, multi-year trend analysis, income statement structure, capital returns analysis, long-term obligations, and producing an insightful financial summary that connects metrics across dimensions.
- ▌ Globem User Analysis 2 · zjunlpComprehensive individual-user analysis on the GLOBEM dataset — a longitudinal passive-sensing + mental-health study of college students. Use this skill whenever a task involves analyzing a specific participant (e.g. "Analyze user INS-W_002") from the GLOBEM dataset, exploring behavioral patterns from smartphone sensors, correlating behavioral signals with mental health outcomes, or producing a comprehensive user profile from multimodal sensing data.
- ▌ Mimic Patient Analysis 2 · zjunlpComprehensive patient analysis using the MIMIC-IV clinical database. Use this skill whenever asked to analyze, summarize, or investigate a patient's medical history, hospital admissions, diagnoses, medications, procedures, or clinical course from a MIMIC-IV SQLite database. Triggers on prompts like "Analyze patient [ID]", "summarize patient history", "what happened to patient X", or any request to explore patient-level EHR data from MIMIC-IV tables.
- ▌ Mimic Patient Analysis 3 · zjunlpComprehensive patient analysis using the MIMIC-IV clinical database. Use this skill whenever asked to analyze, summarize, or investigate a patient's medical history, hospital admissions, diagnoses, medications, procedures, or clinical course from a MIMIC-IV SQLite database. Triggers on prompts like "Analyze patient [ID]", "summarize patient history", "what happened to patient X", or any request to explore patient-level EHR data from MIMIC-IV tables.
- ▌ Globem User Analysis 3 · zjunlpComprehensive individual-user analysis on the GLOBEM dataset — a longitudinal passive-sensing + mental-health study of college students. Use this skill whenever a task involves analyzing a specific participant (e.g. "Analyze user INS-W_002") from the GLOBEM dataset, exploring behavioral patterns from smartphone sensors, correlating behavioral signals with mental health outcomes, or producing a comprehensive user profile from multimodal sensing data.
- ▌ Globem User Analysis 4 · zjunlpComprehensive individual-user analysis on the GLOBEM dataset — a longitudinal passive-sensing + mental-health study of college students. Use this skill whenever a task involves analyzing a specific participant (e.g. "Analyze user INS-W_002") from the GLOBEM dataset, exploring behavioral patterns from smartphone sensors, correlating behavioral signals with mental health outcomes, or producing a comprehensive user profile from multimodal sensing data.
- ▌ Globem User Analysis 5 · zjunlpComprehensive individual-user analysis on the GLOBEM dataset — a longitudinal passive-sensing + mental-health study of college students. Use this skill whenever a task involves analyzing a specific participant (e.g. "Analyze user INS-W_002") from the GLOBEM dataset, exploring behavioral patterns from smartphone sensors, correlating behavioral signals with mental health outcomes, or producing a comprehensive user profile from multimodal sensing data.
- ▌ Globem User Analysis 6 · zjunlpComprehensive individual-user analysis on the GLOBEM dataset — a longitudinal passive-sensing + mental-health study of college students. Use this skill whenever a task involves analyzing a specific participant (e.g. "Analyze user INS-W_002") from the GLOBEM dataset, exploring behavioral patterns from smartphone sensors, correlating behavioral signals with mental health outcomes, or producing a comprehensive user profile from multimodal sensing data.
- ▌ Mimic Patient Analysis 4 · zjunlpComprehensive patient analysis using the MIMIC-IV clinical database. Use this skill whenever asked to analyze, summarize, or investigate a patient's medical history, hospital admissions, diagnoses, medications, procedures, or clinical course from a MIMIC-IV SQLite database. Triggers on prompts like "Analyze patient [ID]", "summarize patient history", "what happened to patient X", or any request to explore patient-level EHR data from MIMIC-IV tables.
- ▌ Mimic Patient Analysis 5 · zjunlpComprehensive patient analysis using the MIMIC-IV clinical database. Use this skill whenever asked to analyze, summarize, or investigate a patient's medical history, hospital admissions, diagnoses, medications, procedures, or clinical course from a MIMIC-IV SQLite database. Triggers on prompts like "Analyze patient [ID]", "summarize patient history", "what happened to patient X", or any request to explore patient-level EHR data from MIMIC-IV tables.
- ▌ Sec 10k Company Analysis 5 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Sec 10k Company Analysis 6 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Sec 10k Company Analysis 7 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Sec 10k Company Analysis 8 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Sec 10k Company Analysis 9 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Sec 10k Company Analysis 10 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Ddr Globem Analysis 2 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 3 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 4 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 5 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 6 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 7 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 8 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 9 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Ddr Globem Analysis 10 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Sec 10k Company Analysis 11 · zjunlpAnalyze a company in an SEC 10-K SQLite database and produce high-quality evidence-grounded financial QA pairs. Use this whenever the user asks to analyze a company by CIK/ticker, inspect 10-K financial trends, generate finance QA datasets, or work with filings/financial_facts tables.
- ▌ Ddr Globem Analysis 11 · zjunlpAnalyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.
- ▌ Mimic Iv Patient Analysis 2 · zjunlpComprehensive strategy for analyzing individual patient records in MIMIC-IV EHR database and generating high-quality, diverse QA pairs. Use this skill whenever the task involves analyzing a specific patient's clinical data from MIMIC-IV (or similar EHR databases), querying across hospital and ICU tables, and submitting QA pairs that cover the patient's full clinical story — diagnoses, procedures, medications, care trajectory, and outcomes. Trigger when you see tasks like "Analyze patient <ID>", "Generate QA pairs for patient", or any patient-centric EHR exploration task.
- ▌ Mimic Iv Patient Analysis 3 · zjunlpComprehensive strategy for analyzing individual patient records in MIMIC-IV EHR database and generating high-quality, diverse QA pairs. Use this skill whenever the task involves analyzing a specific patient's clinical data from MIMIC-IV (or similar EHR databases), querying across hospital and ICU tables, and submitting QA pairs that cover the patient's full clinical story — diagnoses, procedures, medications, care trajectory, and outcomes. Trigger when you see tasks like "Analyze patient <ID>", "Generate QA pairs for patient", or any patient-centric EHR exploration task.
- ▌ Mimic Iv Patient Analysis 4 · zjunlpComprehensive strategy for analyzing individual patient records in MIMIC-IV EHR database and generating high-quality, diverse QA pairs. Use this skill whenever the task involves analyzing a specific patient's clinical data from MIMIC-IV (or similar EHR databases), querying across hospital and ICU tables, and submitting QA pairs that cover the patient's full clinical story — diagnoses, procedures, medications, care trajectory, and outcomes. Trigger when you see tasks like "Analyze patient <ID>", "Generate QA pairs for patient", or any patient-centric EHR exploration task.
- ▌ Mimic Iv Patient Analysis 5 · zjunlpComprehensive strategy for analyzing individual patient records in MIMIC-IV EHR database and generating high-quality, diverse QA pairs. Use this skill whenever the task involves analyzing a specific patient's clinical data from MIMIC-IV (or similar EHR databases), querying across hospital and ICU tables, and submitting QA pairs that cover the patient's full clinical story — diagnoses, procedures, medications, care trajectory, and outcomes. Trigger when you see tasks like "Analyze patient <ID>", "Generate QA pairs for patient", or any patient-centric EHR exploration task.
- ▌ Chat2workflow · zjunlpAn expert Dify workflow builder that designs and generates Dify-importable workflows through multi-round conversation. Produces structured JSON (nodes, edges, variable references) convertible to Dify YAML, with support for incremental add/modify/remove operations on existing workflows.
- ▌ Skill Adaptor · zjunlp bundleValidation-gated skill evolution from agent failure trajectories (step-level t★, Validator Δ>0). Use when: (1) Agent failed on a task and you have execution trajectories or logs, (2) User asks to evolve, adapt, or improve SKILL.md from failures, (3) Benchmark run produced trajectories under .skill-adaptor/artifacts/, (4) User wants training-free skill bank growth with held-out validation — NOT one-shot log-to-skill. Prefer SkillNet ($skillnet) search first; use this skill when failures need verified adaptation. NOT for: packaging a chat log without re-execution (use skillnet create instead).
- ▌ Sciatlas Trend Report · zjunlp bundleUse only SciAtlas search-papers to take a novice user from zero setup to a final timeline-oriented research trend report, including setup, registration guidance, configuration, retrieval, artifact reading, and synthesis. Trigger when the user asks for topic history, field evolution, recent trends, representative papers over time, or emerging directions.
- ▌ Sciatlas Idea Evaluate · zjunlp bundleUse only the current SciAtlas automated review workflow (`review_pipeline`) to take a novice user from zero setup to a final automated review of a research idea or paper, including setup, registration guidance, workflow configuration, retrieval, artifact reading, and synthesis. Trigger when the user asks whether an idea is worth pursuing, wants automatic review, meta-review, rubric-based critique, novelty/feasibility/soundness assessment, or literature-backed evaluation.
- ▌ Sciatlas Idea Generate · zjunlp bundleUse only the current SciAtlas multi-step idea-generation workflow (`sciatlas_idea_gen`) to take a novice user from zero setup to final literature-grounded research idea seeds, including setup, registration guidance, workflow configuration, KG retrieval, artifact reading, novelty checking, and synthesis. Trigger when the user wants new research directions, hypotheses, cross-topic combinations, project ideas, or brainstorming grounded in SciAtlas KG retrieval.
- ▌ Sciatlas Idea Grounding · zjunlp bundleUse only SciAtlas search-papers to take a novice user from zero setup to final prior-art grounding for a research idea, including setup, registration guidance, configuration, retrieval, artifact reading, and synthesis. Trigger when the user gives an idea and asks for similar work, differentiation evidence, related work, motivation support, or literature-grounded refinement.
- ▌ Sciatlas Literature Review · zjunlp bundleUse only the current SciAtlas literature-review workflow (`literature_review_pipeline`) to take a novice user from zero setup to a final evidence-grounded survey outline, paper map, or literature review, including setup, registration guidance, workflow configuration, SciAtlas backend retrieval, artifact reading, and synthesis. Trigger when the user asks for a literature review, related work section, paper map, survey outline, reading path, or topic overview grounded in the provided SciAtlas paper backend.
- ▌ Sciatlas Researcher Review · zjunlp bundleUse only SciAtlas search-papers to take a novice user from zero setup to a final evidence-grounded researcher profile from retrieved papers, including setup, registration guidance, configuration, retrieval, artifact reading, and synthesis. Trigger when the user asks for a researcher's background, representative works, topic trajectory, or author-centered literature overview.
- ▌ Sciatlas Quick Paper Search · zjunlp bundleUse only SciAtlas search-papers to take a novice user from zero setup to a final small evidence-backed paper search result, including setup, registration guidance, configuration, retrieval, artifact reading, and summary. Trigger when the user wants a quick paper check, first-pass literature evidence, a retrieval smoke test, or help deciding which deeper SciAtlas agent skill to use next.
- ▌ Autoscirub · zjunlp bundleCoordinate the full AutoSciRub workflow for autonomous scientific research tasks. Use when a user wants to turn an underspecified research instruction into an executable rubric, guide research execution, verify a report criterion by criterion, and run targeted revision rounds across Codex, Claude Code, OpenClaw, or another agent harness.
- ▌ Targeted Revision · zjunlp bundlePerform one AutoSciRub targeted revision round using failed executable-rubric criteria and criterion-level verification feedback. Use when Codex needs to update a research artifact by adding missing experiments, comparisons, evidence, analysis, figures, tables, or report text while preserving supported claims and recording a structured revision plan and change summary.
- ▌ Criterion Synthesis · zjunlp bundleSynthesize a task-specific executable scientific rubric from an instruction-derived skeleton, literature grounding, and task-data profile. Use in AutoSciRub when Codex needs criteria with linked goals, data sources, required analyses, metrics, comparisons, expected artifacts, provenance, and satisfaction conditions.
- ▌ Task Data Exploration · zjunlp bundleBuild a lightweight task-visible data and resource profile for AutoSciRub. Use after rubric skeleton and literature grounding when Codex needs to inspect available files, datasets, code, schemas, units, labels, tools, constraints, and goal feasibility before synthesizing executable criteria.
- ▌ Mineru PDF To Markdown · zjunlp bundleConvert PDFs and paper-like documents into Markdown with MinerU before reading or analyzing them. Use when AutoSciRub or another scientific agent needs to inspect local or remote PDFs, equations, tables, or scanned papers where direct PDF text extraction would be lossy; supports MinerU Agent lightweight API without a token and MinerU standard API with MINERU_API_TOKEN.
- ▌ Rubric Skeleton Induction · zjunlp bundleInduce an instruction-derived scientific goal skeleton from a research task before reading literature, web sources, data details, code, or prior results. Use at the start of AutoSciRub when Codex needs atomic, traceable goals that state what the task should address without prescribing methods or metrics.
- ▌ Criterion Level Verification · zjunlp bundleVerify a scientific report and supporting artifacts against an AutoSciRub executable rubric criterion by criterion. Use after execution has produced a report, code, results, tables, figures, or other artifacts, when Codex needs strict satisfaction labels, evidence found, evidence gaps, and required actions for targeted revision.
- ▌ Scientific Literature Grounding · zjunlp bundleGround an AutoSciRub rubric skeleton in scientific literature and authoritative documentation. Use after rubric_skeleton.json exists when a research agent needs goal-centered methods, protocols, metrics, baselines, controls, robustness checks, and evidence forms from task-visible papers, public literature, datasets, benchmarks, repositories, or documentation.
- ▌ Arxiv · zjunlp bundleSearch, download, and summarize academic papers from arXiv. Use when user says "search arxiv", "download paper", "fetch arxiv", "arxiv search", "get paper pdf", or wants to find and save papers from arXiv to the local paper library.
- ▌ Mguide · zjunlpMechanist's single conversational front door, with three parallel tracks: (A) when the user wants to run an experiment, work out the research requirements through conversation, settle the two parameter axes of /auto (behavior-source × mechanism), write a compliant task.md into the project root on their behalf, and — once they confirm — submit the /auto run for them; (B) when the user wants literature, hand their words straight to /msearch; (C) when the user wants the arc of a field, hand their words straight to /mhistory.
- ▌ Notify · zjunlpDraft a research-progress briefing and dispatch it through whatever notification service the user has already configured. Channel-agnostic — this skill only drafts the briefing text and scans for a configured service; it does not hard-code or recommend any particular channel or tool. Opt-in: it does nothing unless task.md asks for notifications (e.g. email reminders). Every briefing is saved to notification/ (never overwritten). Called by /auto on an hourly cadence and at progress / done / halted / approval-needed events, or manually via /notify.
- ▌ Msearch · zjunlpStandalone one-shot multi-source literature search. Takes a single query and returns one merged, ranked result set + synthesized landscape from the internet (web + arXiv), local channels (Zotero, Obsidian, local PDFs), and the cloud mechanic-db SEARCH service. Same retrieval engine as `/research-lit`, but standalone: it is NOT part of the `/auto` pipeline (nothing orchestrates it, nothing reads its files) and it writes its output into a dedicated `msearch/` folder. Use when the user says 'search everything for X', 'look X up for me / retrieve X', 'find papers on X across all sources', or invokes `/msearch <query>`.
- ▌ Mhistory · zjunlpGenerate a structured, publication-quality research-history markdown article for a given topic. Uses the cloud `mechanic_database` SEARCH service via skill `/mechanic-db-search` (TWO PARALLEL passes — `temporal_mode=history` for the long arc + `temporal_mode=recent` for the frontier), then supplements with WebSearch for pre-DB classics and the last 1–6 months of arXiv work. Triggers: 'research history', 'development history', 'survey of X', 'trace the evolution of …'.
- ▌ Data Rule · zjunlpThe single place for every data constraint an experiment must satisfy — dataset provenance (existing → adapted → constructed), clear train / validation / test splits, labels that reflect the target behavior, and the minimum data amount. Use whenever an experiment chooses, adapts, or constructs a dataset, defines splits, or sets a sample size — for phenomenon validation (M0), mechanism exploration, intervention, or tuning. Domain-general: no assumption about model family, modality, or task.
- ▌ Auto Claim · zjunlpWorkflow 1: Claim-stage pipeline, controlled by two orthogonal axes. BEHAVIOR_SOURCE selects the behavior stage: `given` (default; behavior taken from task.md and assumed to hold — no ideation, no novelty, no M0), `given-validation` (behavior taken from task.md but the experiment plan opens with an M0 phenomenon-validation gate), or `discovery` (mine a NEW behavior via /mechanism-behavior-discovery + full ideation: research-lit → idea-creator → novelty-check → impact-check → research-review → research-refine-pipeline, impact-first ranking; plan opens with M0). MECHANISM selects the mechanism stage: `discovery` (default; system routes — claim loads /mechanism-explore to shape direction) or `given` (user named the mechanism method/family in task.md — capture it and forward as CHOSEN_FAMILY, no routing). resource_fidelity:strict (exact models/datasets, no downscaling) is stamped iff BEHAVIOR_SOURCE=given AND MECHANISM=given (the reproduction combination). Always emits `idea-stage/IDEA_REPORT.md` + `refine-logs/{
- ▌ Next Round · zjunlpBetween-round transition helper for multi-round /auto exploration. Archives the just-finished round's outputs into rounds/round_<N>/ and drafts the next round's task.md with recommended behavior-source/mechanism params, by reading the orchestrator-owned Global Exploration Memory (research_memory.json). Use after a /auto round completes and you want to start a new round — either exploring a NEW behavior (new-behavior) or a NEW mechanism for an already-found behavior (new-mechanism). Reads memory, never writes it.
- ▌ Idea Creator · zjunlpGenerate and rank research ideas given a broad direction. Use when user says "brainstorm ideas", "generate research ideas", "what can we work on", or wants to explore a research area for publishable directions.
- ▌ Impact Check · zjunlpAssess whether the research problem/behavior is important — its potential value and reach. Use when user says "impact check", "check impact", or wants to judge whether an idea/behavior matters before committing.
- ▌ Paper Figure · zjunlpGenerate publication-quality figures and tables from experiment results. Use when user says "plot this", "make a figure", "generate figures", "paper figures", or needs plots for a paper. Also invoked by `/auto`'s Ledger Figures hook to produce per-claim figures embedded into `CLAIMS_LEDGER.md`.
- ▌ Research Lit · zjunlp bundleSearch and analyze research papers, find related work, summarize key ideas. Use when user says "find papers", "related work", "literature review", "what does this paper say", or needs to understand academic papers.
- ▌ Novelty Check · zjunlpVerify research idea novelty against recent literature. Use when user says "novelty check", "check novelty", or wants to verify a research idea is novel before implementing.
- ▌ Run Experiment · zjunlpDeploy and run ML experiments on local, remote, Vast.ai, or Modal serverless GPU. Use when user says "run experiment", "deploy to server", or needs to launch training jobs.
- ▌ Training Check · zjunlpPeriodically check WandB metrics during training to catch problems early (NaN, loss divergence, idle GPUs). Avoids wasting GPU hours on broken runs. Use when training is running and you want automated health checks.
- ▌ Auto Experiment · zjunlpWorkflow 1.5: Bridge between idea discovery and auto review. Reads EXPERIMENT_PLAN.md, routes mechanism family inline (Phase 1.5), implements experiment code, deploys to GPU, and collects initial results. Use when user says "implement experiments", "experiment", "deploy the plan", or has an experiment plan ready to execute.
- ▌ Experiment Plan · zjunlpTurn a refined research proposal or method idea into a detailed, claim-driven experiment roadmap. Use after `research-refine`, or when the user asks for a detailed experiment plan, ablation matrix, evaluation protocol, run order, compute budget, or paper-ready validation that supports the core problem, novelty, simplicity, and any LLM / VLM / Diffusion / RL-based contribution.
- ▌ Experiment Tips · zjunlpRouting entry point for experiment-protocol tips that prevent silent reproducibility / overclaim failures. Use when EXPERIMENT_PLAN.md is about to become runnable code and any of these is in scope: ImageNet / torchvision preprocessing, steering coefficient (α / dose / magnitude), steering block / layer / site selection, fine-tuning hyperparameters (full FT / LoRA / QLoRA / DoRA / PEFT — LR, capacity, target modules — across SFT / DPO / GRPO / PPO / RLHF objectives), or MCQ / A-B / A-D letter-parse evaluation of a free-form model generation. The **General Rule for mechanism/Interpretability** in this file is loaded on every mechanism / interpretability experiment (locate the target neuron/feature, then intervene without damaging general ability); the numbered tips load only as their symptom triggers apply, and after matching one you MUST load its own SKILL.md — never act on the preview. Universal data constraints (provenance, splits, sample-size floors) live in `skills/data-rule/`, not here.
- ▌ Mechanism Audit · zjunlpAudit the **mechanistic experiment rigor** for a specific claim. Catalogue currently has six slots A–F: A (steering coefficient sweep) is implemented; B–F are reserved for future checks (direction extraction quality, site/layer selection, n_effective sufficiency, probe-vs-causal disentanglement, intervention scope). Uses cross-model review (external LLM reviewer via llm-chat MCP). Complementary to `/experiment-audit` (which audits evaluation methodology, not mechanism tuning). The output `overall_verdict` (PASS/WARN/FAIL/N/A) is THIS claim's mechanism-rigor verdict — i.e., whether the interpretability mechanism backing this claim was tuned with the necessary controls. Returns N/A when the claim's experiment does not use any mechanism intervention (e.g., pure dataset evaluation). Does NOT judge methodology honesty (that is `/experiment-audit`'s job) or semantic claim support (that is `/result-to-claim`'s job).
- ▌ Research Refine · zjunlpTurn a vague research direction into a problem-anchored, elegant, frontier-aware, implementation-oriented method plan via iterative external LLM review (llm-chat MCP). Use when the user says "refine my approach", "decompose this problem", "refine research plan", or wants a concrete research method that stays simple, focused, and top-venue ready instead of a vague or overbuilt idea.
- ▌ Research Review · zjunlpGet a deep critical review of research from an external LLM reviewer via llm-chat MCP. Use when user says "review my research", "help me review", "get external review", or wants critical feedback on research ideas, papers, or experimental results.
- ▌ Result To Claim · zjunlpUse when experiments complete to judge what claims the results support, what they don't, and what evidence is still missing. The external LLM reviewer (via llm-chat MCP) evaluates results against intended claims and routes to next action (pivot, supplement, or confirm). Use after experiments finish — before writing the paper or running ablations.
- ▌ Ablation Planner · zjunlpUse when main results pass result-to-claim (claim_supported=yes or partial) and ablation studies are needed for paper submission. The external LLM reviewer (via llm-chat MCP) designs ablations from a reviewer's perspective, CC reviews feasibility and implements.
- ▌ Experiment Audit · zjunlpAudit the experimental **methodology** integrity for a specific claim (Checks A–F: GT provenance, score normalization, result-file existence, dead code, scope, eval-type). Uses cross-model review (external LLM reviewer via llm-chat MCP). The output `overall_verdict` (PASS/WARN/FAIL) is THIS claim's verdict — i.e., whether the claim's experimental process is methodologically sound. Does NOT judge whether the numbers semantically support the claim's hypothesis — that is `/result-to-claim`'s job.
- ▌ Experiment Queue · zjunlp bundleSSH job queue for multi-seed / multi-config ML experiments with OOM-aware retry, stale-screen cleanup, wave-transition race prevention, and phase-dependency enforcement. Use when user says "batch experiments", "queue experiments", "run grid", "multi-seed sweep", "auto-chain experiments", or when `/run-experiment` is insufficient for ≥10 jobs that need orchestration. `/auto-experiment` Phase 4 auto-routes here when a milestone declares ≥10 jobs or has `depends_on`.
- ▌ Hypothesis Batch · zjunlpAutomated pipeline for generating and refining multiple research hypotheses.
- ▌ Mechanism Skills · zjunlpRouting entry point for eleven families of mechanistic-interpretability methods that localize *which* internal object (layer, attention head, neuron, SAE feature, weight, or input feature) drives a model's behavior, *how influential* it is, and *what changes* when it is intervened on. Use this skill whenever the question is about a model's internal mechanism rather than its external metrics — for example, claims that a specific component is responsible for a behavior, mechanistic-evidence requests, circuit-discovery tasks, feature-attribution work (SHAP), or concept-level explanations of vision/VL models. The file lays out each family's premise, signal, cost, advantages, limitations, and how to compose them into a cheap-screen → causal-verify pipeline. Loading is hierarchical and mandatory: after picking a family from this routing file you MUST load that family's `SKILL.md`, and after picking a submethod you MUST load that submethod's `SKILL.md` — never act on the previews in this file alone.
- ▌ Mechanism Explore · zjunlpMacro-level strategic directions for investigating the *mechanism* behind a model behavior — the downstream half of the project's mission (mine LLM behaviors, then explain the mechanism behind them). Use once a phenomenon is observed in a model — whether already established/known or freshly mined by `/mechanism-behavior-discovery` — and the task is to choose *how* to investigate the internal cause. It is the strategy layer above the concrete method families in `/mechanism-skills`, organized around six parallel research directions — **Location**, **Causal Intervention**, **Tuning & Editing**, **Formation Tracing**, **Unit Interpretation**, **Decision Auditing** — plus how to combine them into strategies. Domain-general: it prescribes strategy, not any single model, modality, or method.
- ▌ Mechanic DB Search · zjunlpPaper retrieval via the cloud SEARCH service. The Agent builds a decomposed query JSON from its task context (preferred) or submits a polished free-form English query; the cloud service performs multi-ranker retrieval and fusion. Use as one of an important paper sources.
- ▌ Monitor Experiment · zjunlpMonitor running experiments, check progress, collect results. Use when user says "check results", "is it done", "monitor", or wants experiment output.
- ▌ Auto Iteration Loop · zjunlpAutonomous research review loop that consumes /auto-verify's four-state output (PASS / FAIL / INCONCLUSIVE / ZERO_ELIGIBLE_VARIANTS / deferred) and routes each claim to the right back-edge — brief audit, two-phase FAIL handling (variant-integrity fix then optional claim-stage re-entry), main-experiment-script fix, or variant-only fix — under a unified iteration budget. Configure the reviewer LLM via llm-chat MCP server or environment variables. Trigger with "auto review loop llm" or "llm review".
- ▌ Image · zjunlpCanonical ImageNet eval preprocessing — square 256×256 resize → 224 center crop → ImageNet mean/std — for CV experiments probing ImageNet-pretrained backbones (ResNet, ViT, VGG, EfficientNet). Use this skill whenever a `torchvision.transforms` / `PIL` pipeline is being written, audited, or debugged for inference-time eval: feature extraction, activation hooks, mechanistic interpretability, top-k-activating-image retrieval, neuron labeling, or reproducibility checks. Apply it even when the user does not say "preprocessing" — triggers include `Resize`, `CenterCrop`, `T.Compose`, "ImageNet eval", "my activations changed between runs", or "results differ from the paper". Covers the `Resize(256)` (int, short-side) vs `Resize((256, 256))` (tuple, square) trap.
- ▌ Shap · zjunlp bundleSHAP (SHapley Additive exPlanations) is a unified, game-theoretic framework for local feature attribution. Treating a model's input features as players in a coalition game, SHAP assigns each feature its Shapley value — the average marginal contribution of that feature to the prediction across all possible feature subsets. Shapley values are the unique attributions satisfying local accuracy, missingness, and consistency, so the prediction decomposes additively into a baseline plus one Shapley value per feature, giving a per-prediction explanation that is theoretically grounded and model-agnostic.
- ▌ Probing · zjunlp bundleProbing methods interpret model signals by training an auxiliary predictor (often linear) to decode a labeled property y from an internal vector $z$ (e.g., the residual stream state $x_l$ at layer $l$). Operationally, probing treats the model as a frozen feature extractor and assesses decodability: whether $y$ is recoverable from $z$ by a restricted hypothesis class (commonly linear), which supports localization by comparison across candidate objects (layers/heads/FFNs) via decoding performance or information-theoretic surrogates, typically followed by Causal Attribution to test functional necessity.
- ▌ Mechanism Behavior Discovery · zjunlpMine behavioral regularities in neural-network (LLM / multimodal) models — the upstream half of the project's mission (find a behavior worth explaining, then investigate the mechanism behind it). Use this skill when the task is open-ended: surface a *new* behavioral phenomenon — a candidate claim / research direction — rather than investigate an already-named mechanism. It gives strategies for choosing which behavior to probe and how to choose the data that validates it. The output is a candidate phenomenon that hands off to `/mechanism-explore` for mechanistic investigation. Domain-general: no assumption about model family, modality, or task.
- ▌ Multi Modal Interpretability · zjunlp bundleMulti-Modal Interpretability methods explain *vision* and *vision-language* models by linking each internal unit (a convolutional neuron, attention head, residual-stream channel, or SAE feature) to a *natural-language concept* drawn from an external concept set $\mathcal{C}$. The unifying construction is a similarity score $s(u, c) = \langle \mathbf{a}_u, \mathbf{t}_c \rangle$ between a per-unit activation summary $\mathbf{a}_u$ — collected over a probing image set — and a text-side embedding $\mathbf{t}_c$ produced by an aligned multi-modal model such as CLIP. Ranking the concepts by $s(u, \cdot)$ yields a human-readable label for $u$, turning vision representations from anonymous tensors into named, attributable concepts. The two submethods below are NOT mutually exclusive — for building a per-component concept vector from a component's highly-activating reference images, select BOTH together (CRP for cropping, CLIP-Dissect for embedding); see "Compose the two".
- ▌ Verify Pick Alternatives · zjunlpSub-skill of /auto-verify. Given a claim, choose one method swap, one dataset swap, and one model swap that most strongly stress-test the claim. Harvests candidates from existing research; calls /research-lit only when coverage is thin. Use when user says "pick swaps for claim", "choose alternatives", or when invoked by /auto-verify.
- ▌ Circuit · zjunlp bundleCircuit discovery methods localize the minimal subgraph of a transformer's computational graph (a set of nodes such as attention heads and MLPs, together with the edges that connect them) that is causally responsible for a specific model behavior. Rather than scoring individual objects in isolation, these techniques recover an end-to-end mechanism by jointly identifying which components, and which information pathways between them, must be preserved to reproduce the behavior on a task distribution.
- ▌ Causal Attribution · zjunlp bundleCausal Attribution methods constitute the gold standard for localization in Mechanism Interpretability. Unlike correlation-based analyses, these techniques identify which internal objects are causally responsible for a specific model behavior by systematically measuring the effect of controlled interventions
- ▌ Gradient Detection · zjunlp bundleGradient Detection methods localize influential internal objects by scoring them with the sensitivity of a scalar target $F(x)$ (e.g., a logit, margin, or loss) with respect to an object $o_j$: $s_j(x) = \phi(\nabla_{o_j} F(x), o_j)$, where common instantiations include the gradient norm $s_j = \|\nabla_{o_j} F(x)\|$ and the gradient-input score $s_j = \nabla_{o_j} F(x)^\top o_j$. These scores serve as fast, first-order proxies for intervention effects.
- ▌ Magnitude Analysis · zjunlp bundleMagnitude Analysis methods serve as a fundamental heuristic in interpretability, operating on the premise that internal elements with larger numerical values often exert greater influence on the model’s computation. It scores internal objects via a scalar function to identify salient components
- ▌ Fastshap · zjunlp bundleUse this skill when you need to train amortized Shapley value explainers using FastSHAP, generate real-time local feature importance explanations for machine learning models (tabular or image), train surrogate models for feature masking, or understand how FastSHAP's KernelSHAP-inspired training objective works with PyTorch.
- ▌ Vocabulary Projection · zjunlp bundleVocabulary Projection methods interpret internal model states by projecting them through the unembedding matrix to obtain a distribution over the vocabulary. The core idea is that the unembedding matrix, which maps the final hidden state to output logits, can serve as a universal decoder for intermediate states throughout the model. By inspecting the top-ranked tokens of the resulting distribution, researchers can directly read off the semantic content encoded in any internal object — such as a residual stream state or an attention head output — in terms of the model's output vocabulary.
- ▌ Zennit Crp · zjunlp bundleUse this skill when working with Concept Relevance Propagation (CRP) and Relevance Maximization for explainable AI in PyTorch models, including generating concept-conditional heatmaps, feature visualizations, attribution graphs, and identifying which latent concepts neural networks use for predictions.
- ▌ Steering Block Selection · zjunlpHow to choose where (and how many sites) to intervene for any operation on internal representations — activation patching, steering, CAA, DAS, SAE feature scaling, attribution patching. Use whenever the plan declares a target block / layer / site, especially when it hard-codes a single index (`target_block: 4`, `layer: 16`) without justification or copied from another paper. Covers picking the site by screening (gradient / attribution, or an activation signal) and by heuristic (mid-to-late layers usually steer best; widen to 3–5 layers if one is inert; sweep the stack at spaced intervals), and which component to target (attention / MLP / residual stream — circuit discovery studies attention heads + MLP). Prevents two symmetric failures: too few sites → downstream norm/attention denoise it → false "no effect"; too many → off-distribution collapse and the localization claim becomes untestable. Triggers include `target_block = 4`, `sliding window = 15`, `layer ∈ [12,18]`, `early blocks (0-7)`, "steering had no e
- ▌ Neural Feature Learning · zjunlp bundleNeural Feature Learning methods open the black box of *how* a trained neural network discovers useful representations from data, by deriving the directions along which each layer's effective input geometry is reshaped during training. The unifying object is the Neural Feature Matrix (NFM) — for a fully-connected layer, the NFM is the weight matrix multiplied by its transpose, and acts as a metric on that layer's input space. Empirically, the NFM aligns at convergence with the Expected Gradient Outer Product (EGOP), defined as the average outer product of the network output's gradient with respect to the same layer's input. The NFM's top eigenvectors point in the directions the network has learned to use — turning feature learning into an analyzable, kernel-shaped object.
- ▌ Clip Dissect · zjunlp bundleUse this skill when you need to automatically describe or interpret the functionality of individual neurons in deep neural networks (DNNs) using CLIP-based semantic analysis, perform mechanistic interpretability research on vision models, dissect convolutional or transformer-based image classifiers, identify what visual concepts activate specific neurons, or compare neuron descriptions across different probing datasets and concept sets.
- ▌ Multiple Choice Evaluation · zjunlpHow to grade any multiple-choice / A-B / A-D letter task when the score is read out of the model's free-form output — any eval that maps a generation to a choice letter, whatever the domain. Use whenever the plan parses the letter with a regex like `[A-D]` / `re.search(r"[AB]", ...)`. Covers why naive letter-regex is fragile (a `"Using a…"` opener parses as "A"; `"At the garden…"` parses as "A"), the LLM-judge fix (three-way `{CORRECT, INCORRECT, OTHER}` verdict against the gold letter, no silent coercion), and the two biases every multiple-choice eval must control — position bias (swap A/B orientation) and token bias (`A` gets more prior mass than `B`). Triggers: "answer A or B", `[A-D]` regex on generations, any per-choice rate (e.g. `P(chosen)`) computed from a letter parse.