research-harness-template

Unattended, scheduled monitoring of external sources (arXiv, OpenAlex, Crossref, Semantic Scholar, PubMed, bioRxiv/medRxiv, GDELT, Hacker News) for operator-selected current-events DOMAINS (monitoringDomains[], weighted, momentum-ranked, digest output) or for a research topic (the topic-bound special case), gated end to end through a mandatory human Editorial Gate before anything publishes. An OPTIONAL methodology pack (enable the `continuous-monitor` pack AND configure a domain or a topic's continuousMonitoring block). Use this skill when the user wants to enable, inspect, or manually run continuous monitoring for a domain or topic, review a run's digest or candidate recommendations, or check the Continuity Log for a skipped/failed source. Triggers on 'continuous monitoring', 'monitor this domain', 'monitor this topic', 'monitoring digest', 'schedule research monitoring', 'check for new sources', 'run the monitoring pipeline', 'review monitoring recommendations'.

Render surviving research findings into a publishable blog post through the typed findings-to-artifact contract. A first-class harness output (not a pack). Use this skill when the user wants to turn research into a blog post, publish findings, draft an article from the corpus, or write up what the research found. Triggers on 'publish blog', 'blog post from research', 'write up these findings', 'draft an article', 'turn this into a post'.

Render surviving research findings into a book chapter through the typed findings-to-artifact contract. An OPTIONAL channel pack (enable the `book` pack to use it), supporting the full genre range. Use this skill when the user wants to author a book, write a chapter from the research, turn the corpus into long-form, or build a manuscript. Triggers on 'write a book', 'author a chapter', 'book from research', 'turn this into a chapter', 'long-form from the corpus', 'manuscript'.

Create, validate, inspect, and convert MIF ontology definition files. Use when the user asks to: create a new ontology, scaffold an ontology, validate an ontology YAML, inspect ontology contents (entities, traits, relationships, namespaces, discovery patterns), convert between YAML and JSON formats, convert to JSON-LD, or get guidance on MIF ontology structure. Triggers on: "ontology", "create ontology", "validate ontology", "inspect ontology", "convert ontology", "ontology schema", "entity types", "traits", "discovery patterns", "namespace hierarchy".

Render a research topic's entire surviving findings corpus into a COMPLETE Diátaxis documentation set — a reference page per finding, a per-dimension explanation, how-to, and guided tutorial, plus landing/index pages — every page carrying MIF Level-1 frontmatter and a diataxis_type marker. An OPTIONAL channel pack (enable the `diataxis` pack). Use this skill when the user wants documentation for a research topic, a full docs set from the findings, or to publish the corpus as Diátaxis docs. Triggers on 'diataxis docs', 'documentation channel', 'render docs from research', 'document the findings', 'reference and explanation docs', 'generate the docs site'.

Render a topic's surviving findings into an AI-ready, agent-executable architecture spec. A genre-shaping of the artifact -> Markdown pipeline (not a new mechanism) — finding_refs become grounded evidence sections, the goal's completion checks become EARS acceptance criteria, and the artifact sections become the document structure. Use to turn research into a buildable spec a downstream coding agent executes.

Search the MIF research index — find findings by free-text query or by structured filters over the index fields (dimension, tags, namespace, verdict). Use this skill when the user wants to find research findings, look up what is known about a subject, query findings, or asks 'what do I have on X', 'find findings about Y', 'search research for Z', 'what research', 'what findings'. Reads the projected index, not raw tag files.

Build, refresh, and query the MIF-native knowledge graph — the typed substrate of concepts and entities (all urn:mif: ids) linked by MIF relationships and mentions. Use this skill when the user asks about relationships between findings, what connects two concepts or entities, entity overlap, or wants a visual map. Triggers on 'what connects', 'how are these related', 'show graph', 'knowledge graph', 'visualize research', 'rebuild the graph', 'entity overlap', 'shared between'. With --concordance, the same verbs operate on the corpus-wide ontological spine spanning ALL topics (build/validate/query reports/concordance.json) — triggers on 'concordance', 'across topics', 'whole corpus', 'cross-topic', 'ontological spine', 'world view of knowledge'. The graph is derived from MIF entities and relations, never from tags.

Discover gaps, clusters, and stale findings across the research corpus — dimensions with thin coverage, concepts and entities that co-occur through graph edges, and findings that are aging or carry a weakened/quarantined verdict. Use this skill when the user asks 'where are my gaps', 'what's stale', 'what clusters exist', 'research coverage', 'what am I missing', 'audit the corpus', 'what needs updating', 'coverage matrix', 'stale findings'.

Open an interactive exploration session over the research corpus — traverse MIF relationships from a seed concept or entity, develop and stress-test hypotheses against the finding set, and reason across topics. Use this skill whenever the user wants to discuss, explore, or riff on their research rather than just retrieve it: 'explore topic X', 'what does my research suggest about Y', 'develop a hypothesis about Z', 'interrogate the findings on W', 'think through this with my research', 'brainstorm from the corpus'.

Remediate non-compliant Markdown — fix the markdownlint classes that `markdownlint-cli2 --fix` cannot auto-resolve (MD025 frontmatter-title, MD060 tables, MD033 placeholders, MD036 bold-as-heading, MD029 list-embedded blocks), repair broken intra-repo links, and triage cSpell/typos, escalating only genuine judgment calls. Use this skill when the user wants to fix Markdown lint errors, clean up markdown, make a doc compliant, fix broken doc links, resolve markdownlint/cSpell/typos diagnostics, or remediate documentation without suppressing warnings. Triggers on 'fix markdown', 'fix lint', 'markdownlint errors', 'fix the docs', 'clean up this markdown', 'fix broken links', 'md-fix', 'remediate markdown'. Never suppresses a diagnostic.

Render surviving research findings into the canonical MIF Level-3 markdown report — the harness's source of truth (reports/<topic>/<slug>.md), held to the same bar as a finding and graded by the falsification gate. A first-class harness output (not a pack), never exempt. Use this skill when the user wants the authoritative MIF report, the L3 report of record, or to emit the canonical report from the corpus. Triggers on 'publish report', 'generate the MIF report', 'render the report of record', 'emit the canonical report', 'L3 report from findings'.

List the registered research topics from the harness manifest, each with its MIF namespace and a finding/verdict rollup from the index. Use this skill when the user asks 'what topics do I have', 'list topics', 'show the registry', 'what am I researching', 'topic status', or wants an overview of the corpus by topic. Reads harness.config.json topics[] over the MIF substrate.

Package a clinical-submission artifact into the FDA eCTD (electronic Common Technical Document) module structure directly FROM THE SOURCES — a research topic's surviving findings corpus and the primary materials its citations point to. Lays out the eCTD module tree (M1 regional administrative, M2 summaries, M3 quality, M4 nonclinical study reports, M5 clinical study reports) and writes the eCTD XML backbone that indexes the leaf files. NEVER built from a rendered report (that would be a copy-of-a-copy). Optional channel; pure mkdir + XML, no external toolchain. Verify-live eCTD v4.0. Use when the user says 'package as eCTD', 'build the eCTD submission', or 'render to eCTD'.

Publish a research topic to GitHub Discussions — an announcement of completed research, a Q&A thread about methodology, or a non-canonical observation. Grounded in the SOURCES (the findings corpus and their primary-source citations) — NEVER a rendered report. Optional channel; depends on the `gh` CLI. Use when the user says 'post to discussions', 'announce the research', 'create a discussion', or 'share the findings'.

Convert a research topic's surviving findings into actionable GitHub issues — atomize each finding's implications into sprint-sized, well-structured issues with acceptance criteria and evidence-justified priority. Grounded in the SOURCES (the findings corpus and their primary-source citations) — NEVER a rendered report. Optional channel; depends on the `gh` CLI. Use when the user says 'create issues from the research', 'turn findings into issues', or 'file action items'.

Render a research topic's surviving findings into well-formed JATS (NISO Z39.96) XML directly FROM THE SOURCES — the findings corpus and the primary materials its citations point to. Composes a scholarly article (article metadata in <front>/<article-meta>, a <body> of one section per finding, and a <back>/<ref-list> reference list) and serializes it as JATS XML. NEVER built from a rendered report (that would be a copy-of-a-copy). Optional channel; verify the current NISO Z39.96 version live before tagging. Use when the user says 'render to JATS', 'export as JATS XML', or 'make a JATS article'.

Produce NotebookLM assets (audio deep-dive, slide deck, infographics, video, mind map) grounded in THE SOURCES — a research topic's surviving findings and the primary materials its citations point to. Adds the actual cited primary sources to the notebook (so NotebookLM synthesizes from source documents, not a pre-digested report) plus a findings digest. NEVER built from a rendered report (a simulacrum of the sources). Optional channel; depends on the external `nlm` CLI. Use when the user says 'notebooklm assets', 'audio overview', 'podcast from the research', or 'nlm assets'.

Produce a styled, print-ready PDF directly FROM THE SOURCES — a research topic's surviving findings corpus and the primary materials its citations point to. Composes a comprehensive document grounded in the findings (one section per finding, with each finding's primary-source citations) and a complete References list, then renders it with pandoc + a title page + rasterized Mermaid diagrams. NEVER built from a rendered report (that would be a copy-of-a-copy). Optional channel; depends on pandoc + a PDF engine, and mermaid-cli for diagrams. Use when the user says 'render to PDF', 'export as PDF', or 'make a PDF'.

Render surviving findings (typically a regulatory-disclosure artifact) into an inline XBRL (iXBRL) document directly from the SOURCES. Tags facts with xbrli:context / xbrli:unit and taxonomy concepts inside a human-readable XHTML shell, anchored to the current SEC inline-XBRL taxonomy. Never built from a rendered report.

Use when the user asks to "analyze competitors", "map competitive landscape", "Porter's 5 Forces analysis", "competitor comparison", "competitive positioning", "identify competitors", "competitive intelligence", or needs competitor research methodology, market positioning analysis, or competitive strategy frameworks.

Use when the user asks to "understand customers", "customer research", "user personas", "customer needs analysis", "buyer journey mapping", "voice of customer", "customer segmentation", "user research", or needs guidance on customer-discovery methodologies, persona development, or buyer behavior.

Use when the user asks to "analyze financials", "revenue model", "unit economics", "pricing analysis", "cost structure", "profitability analysis", "financial projections", "business model economics", or needs financial-metric, revenue-analysis, or economic-viability guidance.

Use when the user asks to "calculate market size", "TAM SAM SOM analysis", "estimate market opportunity", "market sizing", "total addressable market", "serviceable market", "market potential", or needs market-size estimation methodologies, opportunity calculations, or growth projections.

Use when the user asks to "analyze regulations", "regulatory landscape", "compliance requirements", "legal considerations", "regulatory risk", "industry regulations", "compliance analysis", "regulatory trends", or needs guidance on regulatory environments, compliance requirements, or legal market factors.

Use when the user asks to "model trends with limited data", "three-valued logic analysis", "scenario generation", "transitional graphs", "qualitative trend analysis", "uncertain data analysis", "minimal-information modeling", or needs trend-based modeling with INC/DEC/CONST logic, scenario planning with little quantitative data, or transitional scenario graphs.

Create, update, and validate a topic's navigation README at reports/<topic>/README.md — the per-topic index (title, verdict-aware metadata header, purpose, dimensions, synthesis-grade key findings, reports table, findings-by-dimension table, optional artifacts, tags) modeled on the per-directory READMEs of a research corpus. Use this skill whenever a report topic is created or mutated and its README must be reconciled: after a research session synthesizes, after findings or reports change, or when the user asks to 'update the README', 'reindex the topic readme', 'reconcile READMEs', or 'regenerate the topic index'. Reads harness.config.json topics[], reports/<topic>/findings/*.json, and goal.json — counts are computed, key findings are synthesized.