[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-anthropic-fraud-detection":3,"mdc-s6mdxc-key":36,"related-repo-anthropic-fraud-detection":1966,"related-org-anthropic-fraud-detection":2067},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":12,"stars":26,"repoUrl":27,"updatedAt":28,"license":29,"forks":30,"topics":31,"repo":32,"sourceUrl":34,"mdContent":35},"fraud-detection","detect fraud in healthcare claims","Screen a Medicare\u002FMedicaid claims corpus for fraud, waste, and abuse and produce ranked, fully-cited investigation referrals for an SIU \u002F program-integrity team. Use when asked to run a fraud sweep, screen claims for FWA, find billing anomalies, or generate investigation referrals over a claims dataset.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},"anthropic","Anthropic","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fanthropic.png","anthropics",[13,17,20,23],{"name":14,"slug":15,"type":16},"Compliance","compliance","tag",{"name":18,"slug":19,"type":16},"Healthcare","healthcare",{"name":21,"slug":22,"type":16},"Audit","audit",{"name":24,"slug":25,"type":16},"Insurance","insurance",356,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fhealthcare","2026-06-26T07:50:49.129616",null,91,[],{"repoUrl":27,"stars":26,"forks":30,"topics":33,"description":29},[],"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fhealthcare\u002Ftree\u002FHEAD\u002Fplugins\u002Fhealthcare\u002Fskills\u002Ffraud-detection","---\nname: fraud-detection\ndescription: Screen a Medicare\u002FMedicaid claims corpus for fraud, waste, and abuse and produce ranked, fully-cited investigation referrals for an SIU \u002F program-integrity team. Use when asked to run a fraud sweep, screen claims for FWA, find billing anomalies, or generate investigation referrals over a claims dataset.\n---\n\n# Fraud Detection — claims screening → cited investigation\n\nScreens a Medicare\u002FMedicaid claims corpus against the public rulebook (NCCI MUE, OIG LEIE,\nCMS enrollment, PFS) and produces **ranked, fully-cited investigation referrals** for an SIU.\nThe skill **orchestrates a three-tier investigation**: a deterministic floor does the detection,\nthe model judges and narrates on top, and every dollar\u002Frule allegation traces back to the floor.\n\n## Output framing\n- **\"Indicators consistent with [scheme],\" not \"fraud.\"** A pattern match doesn't establish intent\n  — that's a downstream investigative\u002Flegal determination. This is standard SIU language and the\n  framing the renderers use.\n- **Render for review.** The skill writes packets to `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fout\u002F`; the payer's SIU workflow\n  decides what to do with them. The model does not send\u002Fpublish on its own.\n\n## Inputs\n- **The payer's claims** in `corpus.duckdb` (canonical 6-table schema: `claims-schema.sql`). Getting\n  this is **step 1** below — without it nothing else matters.\n- **Quarter** (the NCCI\u002FPFS rule set to cite against, e.g. `2026q3`).\n- **Line of business** (`medicare` \u002F `medicaid`).\n\n## Data root\nAll fetched\u002Fgenerated state lives **outside** the plugin install path (which is wiped on upgrade)\nat `~\u002F.claude\u002Fdata\u002Fhealthcare\u002Ffraud-detection\u002F` — override the parent dir with\n`$CLAUDE_HEALTHCARE_DATA` (each skill appends its own name). Below, `data-cache\u002F` and `out\u002F` are\nsubdirectories of `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection`. Resolve it once at the start of a run:\n```bash\nexport CLAUDE_HEALTHCARE_DATA=\"${CLAUDE_HEALTHCARE_DATA:-$HOME\u002F.claude\u002Fdata\u002Fhealthcare}\"\n```\n\n## Steps\n1. **Get the payer's claims into `corpus.duckdb`.** Open with: *\"Where do your adjudicated claims\n   live?\"* and follow `${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002FLOAD-CLAIMS.md` — it walks you\n   and the user from \"I don't know\" to a populated `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Fcorpus.duckdb`. If they\n   already have a `.duckdb` with the canonical tables (schema:\n   `${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fclaims-schema.sql`), use it directly.\n\n   Draft the brief (corpusDb path, quarter, line of business) and confirm scope.\n2. **Seed the public reference layer (first run \u002F new quarter only).** Detectors cite against\n   `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Freference\u002F\u003Cquarter>\u002Freference.duckdb`. If that file is missing for the\n   requested quarter, fetch it now — this prints per-source `✓ name (size)` progress as ~34 sources land:\n   ```bash\n   node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Ffetch-reference.js\" 2026q3\n   node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Ffetch-enrichment.js\"\n   ```\n   Requires `unzip` and `pdftotext` (poppler) on PATH; both ship with most distros \u002F `brew install\n   poppler`. Needs real network egress — if you see \"Could not resolve host\" for cms.gov \u002F oig.hhs.gov,\n   the command sandbox is blocking it; re-run with sandbox disabled. Policy PDFs (NCCI manual, MLN articles) land under `reference\u002F\u003Cq>\u002Fpolicy\u002F*.txt`\n   for grep; everything keyed lands in `reference.duckdb`. Skip if already present. If a fetch fails\n   or a table is missing, see `REFERENCE-DATA.md` for source URLs and recovery.\n3. **Create the run directory.** Each invocation lands in its own minute-stamped directory so prior\n   runs are preserved side-by-side. Every script honors `FRAUD_OUT_DIR`:\n   ```bash\n   export FRAUD_OUT_DIR=\"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fout\u002Frun-$(date +%Y%m%d-%H%M)\"\n   mkdir -p \"$FRAUD_OUT_DIR\"\n   echo \"$FRAUD_OUT_DIR\"\n   ```\n   Use the printed absolute path verbatim as `outDir` in the next step.\n4. **Run the investigation** by calling the **Workflow** tool with:\n   - `scriptPath`: `${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fworkflows\u002Finvestigate.js`\n   - `args`: `{ \"corpusDb\": \"\u003Cabs path to corpus.duckdb>\", \"quarter\": \"2026q3\", \"lob\": \"medicaid\", \"pluginRoot\": \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\", \"dataRoot\": \"\u003Cabs $CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection>\", \"outDir\": \"\u003Cabs FRAUD_OUT_DIR from step 3>\" }`\n\n   The workflow runs three stages (see \"How it works\"):\n   - **Detect** — runs the deterministic sweep (`scripts\u002Fscreen.js`, zero model) → `$FRAUD_OUT_DIR\u002Freferrals.detect.json`.\n   - **Adjudicate** — one agent per judgment-required finding (D2\u002FD4\u002FD7\u002FD13) sets `status` + `adjudication.reason`; mechanical detectors auto-confirm. Adjudicate may dismiss or downgrade, never add.\n   - **Synthesize** — one agent per provider writes the investigator narrative, plus novel-lead discovery with adversarial verification.\n\n   Tell the user they can watch the fan-out live with `\u002Fworkflows`.\n5. **Materialize the stage snapshots + render** (required — this is the reviewable deliverable). The\n   workflow sandbox has no filesystem, so write its return to disk and let `apply-stages.js` produce\n   the auditable spine. **`FRAUD_OUT_DIR` does not persist across separate Bash calls** — re-export\n   it (to the same absolute path you printed in step 3) at the top of every shell block that needs it:\n   Use the **Write** tool to save the workflow's return JSON verbatim to\n   `$FRAUD_OUT_DIR\u002Fworkflow-result.json` (it can be 50KB+ — don't heredoc it through Bash). Then:\n   ```bash\n   export FRAUD_OUT_DIR=\"\u003Cabs path from step 3>\"\n   node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Fapply-stages.js\"\n   ```\n   This writes `$FRAUD_OUT_DIR\u002Freferrals.adjudicated.json`, `referrals.final.json`, `referrals.json`\n   (canonical, = final), and the renderer sidecars (`source-excerpts.json`, `providers.json`).\n\n   Then render the packets FIRST, then the dashboard (the dashboard only links a provider row to its\n   packet if that packet file already exists), then the xlsx:\n   ```bash\n   export FRAUD_OUT_DIR=\"\u003Cabs path from step 3>\"\n   node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-packet.js\" --all\n   node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-dashboard.js\"\n   node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-xlsx.js\"\n   ```\n   → `$FRAUD_OUT_DIR\u002Fprovider-packet-\u003Cnpi>.html` ×N, `index.html`, `referrals.xlsx`.\n6. **Show the dashboard** so the user can validate it visually.\n   - **Claude Code Desktop with the preview tool available**: serve the run directory and open it\n     in the side-pane preview — `npx serve \"$FRAUD_OUT_DIR\"` (the dashboard is `index.html`,\n     so the root URL is the dashboard; packet links resolve as siblings).\n   - **Otherwise** (terminal CLI, no preview tool): use the OS opener on\n     `$FRAUD_OUT_DIR\u002Findex.html` — try in order, stop at the first that works:\n     ```bash\n     f=\"$FRAUD_OUT_DIR\u002Findex.html\"\n     open \"$f\" 2>\u002Fdev\u002Fnull \\            # macOS\n       || xdg-open \"$f\" 2>\u002Fdev\u002Fnull \\   # Linux\n       || wslview \"$f\" 2>\u002Fdev\u002Fnull \\    # WSL\n       || powershell.exe start \"$(wslpath -w \"$f\")\" 2>\u002Fdev\u002Fnull \\  # WSL→Windows fallback\n       || cmd.exe \u002Fc start \"\" \"$f\" 2>\u002Fdev\u002Fnull \\                   # Windows\n       || echo \"Could not auto-open; open manually: $f\"\n     ```\n   Skip this entirely in a headless\u002Fnon-interactive run (eval, CI, Cowork) — just report the path.\n   Don't fail the run if opening\u002Fserving fails.\n7. **Relay**: the ranked referrals (NPI, schemes, exposure $, confidence) and total exposure from\n   the workflow result, verbatim where it cites numbers. Surface `meta.disclaimer` if it is set. Do\n   not add any dollar or rule the deterministic floor did not produce. **End the response with the\n   run-directory path on its own line** so downstream graders\u002Ftools can locate the artifacts:\n   ```\n   Run directory: \u003Cabsolute $FRAUD_OUT_DIR>\n   ```\n8. **Close the loop** (optional) — see `${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002FPROPOSE-DETECTORS.md`\n   to mine this run for new detector candidates and payer-specific adjudicate-time checks.\n\n## The inviolable line\nThe model adjudicates, explores, and narrates freely, but **any dollar or rule allegation must trace\nto a detect-stage deterministic recompute** (the gate in `scripts\u002Fgate.js`). Adjudicate may **dismiss\nor downgrade** a finding (with an auditable reason) — it never adds one or changes its dollars.\nSynthesize narratives are separate, clearly-marked model output and never introduce a number the\nfloor did not compute.\n\n## Enrichment — local cached data (canonical), MCPs for interactive only\nThe deterministic pipeline reads enrichment from **local cached files** (`scripts\u002Ffetch-enrichment.js`\n→ `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Fenrichment\u002F`, loaded via `scripts\u002Fenrichment.js`) — no runtime auth, no drift, fully\nreproducible. The healthcare plugin's bundled MCP servers (CMS Coverage \u002F ICD-10 \u002F NPI Registry) are\nfor **interactive adjudicate\u002Fsynthesize exploration** only; the pipeline does not depend on them.\n- **ICD-10-CM** — code validity \u002F description (NLM Clinical Tables)\n- **CMS Coverage (LCD\u002FNCD)** — medical-necessity policy index; cached, feeds D4 adjudication\n- **NPI Registry** — provider taxonomy\u002Fstatus\n\n## How it works (plugin layout)\n- **Entry skill** — this file; orchestrates the workflow, never does the math.\n- **Workflow** — `workflows\u002Finvestigate.js` (Claude Code dynamic Workflow): Detect → Adjudicate → Synthesize.\n- **Deterministic sweep** — `scripts\u002Fscreen.js \u003Ccorpus.duckdb> \u003Cquarter> \u003Clob>` runs all detectors and\n  writes `$FRAUD_OUT_DIR\u002Freferrals.json`. Zero model calls.\n- **Detectors** — `scripts\u002FdNN-*.js` (one deterministic module each, sharing the Finding shape).\n- **Pipeline** — `scripts\u002Fpipeline.js` (run → gate → roll up → rank → `referrals.json`).\n- **Citation gate** — `scripts\u002Fgate.js` (independently recomputes every cited number; uncited or\n  non-reproducing findings are dropped — \"citation-or-zero\").\n- **Reference data** — `scripts\u002Freference-data.js` loads `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Freference\u002F` (NCCI\u002FMUE, LEIE, PFS,\n  enrollment), fetched by `scripts\u002Ffetch-reference.js`, versioned by date-of-service quarter.\n- **Enrichment** — `scripts\u002Fenrichment.js` loads `$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Fenrichment\u002F`, fetched by `fetch-enrichment.js`.\n- **Stage merge** — `scripts\u002Fapply-stages.js` (workflow return → `referrals.adjudicated.json` \u002F `.final.json`).\n- **Renderers** — `scripts\u002Frender-dashboard.js` (→ `index.html`), `render-packet.js`, `render-xlsx.js` → `$FRAUD_OUT_DIR\u002F`.\n\nEvery allegation cites a public rule with a value the gate independently recomputes, or it is dropped.\n",{"data":37,"body":38},{"name":4,"description":6},{"type":39,"children":40},"root",[41,50,71,78,120,126,205,211,263,355,361,1585,1591,1618,1624,1666,1699,1705,1955,1960],{"type":42,"tag":43,"props":44,"children":46},"element","h1",{"id":45},"fraud-detection-claims-screening-cited-investigation",[47],{"type":48,"value":49},"text","Fraud Detection — claims screening → cited investigation",{"type":42,"tag":51,"props":52,"children":53},"p",{},[54,56,62,64,69],{"type":48,"value":55},"Screens a Medicare\u002FMedicaid claims corpus against the public rulebook (NCCI MUE, OIG LEIE,\nCMS enrollment, PFS) and produces ",{"type":42,"tag":57,"props":58,"children":59},"strong",{},[60],{"type":48,"value":61},"ranked, fully-cited investigation referrals",{"type":48,"value":63}," for an SIU.\nThe skill ",{"type":42,"tag":57,"props":65,"children":66},{},[67],{"type":48,"value":68},"orchestrates a three-tier investigation",{"type":48,"value":70},": a deterministic floor does the detection,\nthe model judges and narrates on top, and every dollar\u002Frule allegation traces back to the floor.",{"type":42,"tag":72,"props":73,"children":75},"h2",{"id":74},"output-framing",[76],{"type":48,"value":77},"Output framing",{"type":42,"tag":79,"props":80,"children":81},"ul",{},[82,101],{"type":42,"tag":83,"props":84,"children":85},"li",{},[86,99],{"type":42,"tag":57,"props":87,"children":88},{},[89,91,97],{"type":48,"value":90},"\"Indicators consistent with ",{"type":42,"tag":92,"props":93,"children":94},"span",{},[95],{"type":48,"value":96},"scheme",{"type":48,"value":98},",\" not \"fraud.\"",{"type":48,"value":100}," A pattern match doesn't establish intent\n— that's a downstream investigative\u002Flegal determination. This is standard SIU language and the\nframing the renderers use.",{"type":42,"tag":83,"props":102,"children":103},{},[104,109,111,118],{"type":42,"tag":57,"props":105,"children":106},{},[107],{"type":48,"value":108},"Render for review.",{"type":48,"value":110}," The skill writes packets to ",{"type":42,"tag":112,"props":113,"children":115},"code",{"className":114},[],[116],{"type":48,"value":117},"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fout\u002F",{"type":48,"value":119},"; the payer's SIU workflow\ndecides what to do with them. The model does not send\u002Fpublish on its own.",{"type":42,"tag":72,"props":121,"children":123},{"id":122},"inputs",[124],{"type":48,"value":125},"Inputs",{"type":42,"tag":79,"props":127,"children":128},{},[129,162,180],{"type":42,"tag":83,"props":130,"children":131},{},[132,137,139,145,147,153,155,160],{"type":42,"tag":57,"props":133,"children":134},{},[135],{"type":48,"value":136},"The payer's claims",{"type":48,"value":138}," in ",{"type":42,"tag":112,"props":140,"children":142},{"className":141},[],[143],{"type":48,"value":144},"corpus.duckdb",{"type":48,"value":146}," (canonical 6-table schema: ",{"type":42,"tag":112,"props":148,"children":150},{"className":149},[],[151],{"type":48,"value":152},"claims-schema.sql",{"type":48,"value":154},"). Getting\nthis is ",{"type":42,"tag":57,"props":156,"children":157},{},[158],{"type":48,"value":159},"step 1",{"type":48,"value":161}," below — without it nothing else matters.",{"type":42,"tag":83,"props":163,"children":164},{},[165,170,172,178],{"type":42,"tag":57,"props":166,"children":167},{},[168],{"type":48,"value":169},"Quarter",{"type":48,"value":171}," (the NCCI\u002FPFS rule set to cite against, e.g. ",{"type":42,"tag":112,"props":173,"children":175},{"className":174},[],[176],{"type":48,"value":177},"2026q3",{"type":48,"value":179},").",{"type":42,"tag":83,"props":181,"children":182},{},[183,188,190,196,198,204],{"type":42,"tag":57,"props":184,"children":185},{},[186],{"type":48,"value":187},"Line of business",{"type":48,"value":189}," (",{"type":42,"tag":112,"props":191,"children":193},{"className":192},[],[194],{"type":48,"value":195},"medicare",{"type":48,"value":197}," \u002F ",{"type":42,"tag":112,"props":199,"children":201},{"className":200},[],[202],{"type":48,"value":203},"medicaid",{"type":48,"value":179},{"type":42,"tag":72,"props":206,"children":208},{"id":207},"data-root",[209],{"type":48,"value":210},"Data root",{"type":42,"tag":51,"props":212,"children":213},{},[214,216,221,223,229,231,237,239,245,247,253,255,261],{"type":48,"value":215},"All fetched\u002Fgenerated state lives ",{"type":42,"tag":57,"props":217,"children":218},{},[219],{"type":48,"value":220},"outside",{"type":48,"value":222}," the plugin install path (which is wiped on upgrade)\nat ",{"type":42,"tag":112,"props":224,"children":226},{"className":225},[],[227],{"type":48,"value":228},"~\u002F.claude\u002Fdata\u002Fhealthcare\u002Ffraud-detection\u002F",{"type":48,"value":230}," — override the parent dir with\n",{"type":42,"tag":112,"props":232,"children":234},{"className":233},[],[235],{"type":48,"value":236},"$CLAUDE_HEALTHCARE_DATA",{"type":48,"value":238}," (each skill appends its own name). Below, ",{"type":42,"tag":112,"props":240,"children":242},{"className":241},[],[243],{"type":48,"value":244},"data-cache\u002F",{"type":48,"value":246}," and ",{"type":42,"tag":112,"props":248,"children":250},{"className":249},[],[251],{"type":48,"value":252},"out\u002F",{"type":48,"value":254}," are\nsubdirectories of ",{"type":42,"tag":112,"props":256,"children":258},{"className":257},[],[259],{"type":48,"value":260},"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection",{"type":48,"value":262},". Resolve it once at the start of a run:",{"type":42,"tag":264,"props":265,"children":270},"pre",{"className":266,"code":267,"language":268,"meta":269,"style":269},"language-bash shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","export CLAUDE_HEALTHCARE_DATA=\"${CLAUDE_HEALTHCARE_DATA:-$HOME\u002F.claude\u002Fdata\u002Fhealthcare}\"\n","bash","",[271],{"type":42,"tag":112,"props":272,"children":273},{"__ignoreMap":269},[274],{"type":42,"tag":92,"props":275,"children":278},{"class":276,"line":277},"line",1,[279,285,291,297,302,307,312,317,322,328,333,337,342,346,350],{"type":42,"tag":92,"props":280,"children":282},{"style":281},"--shiki-light:#9C3EDA;--shiki-default:#C792EA;--shiki-dark:#C792EA",[283],{"type":48,"value":284},"export",{"type":42,"tag":92,"props":286,"children":288},{"style":287},"--shiki-light:#90A4AE;--shiki-default:#EEFFFF;--shiki-dark:#BABED8",[289],{"type":48,"value":290}," CLAUDE_HEALTHCARE_DATA",{"type":42,"tag":92,"props":292,"children":294},{"style":293},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[295],{"type":48,"value":296},"=",{"type":42,"tag":92,"props":298,"children":299},{"style":293},[300],{"type":48,"value":301},"\"${",{"type":42,"tag":92,"props":303,"children":304},{"style":287},[305],{"type":48,"value":306},"CLAUDE_HEALTHCARE_DATA",{"type":42,"tag":92,"props":308,"children":309},{"style":293},[310],{"type":48,"value":311},":-",{"type":42,"tag":92,"props":313,"children":314},{"style":287},[315],{"type":48,"value":316},"$HOME",{"type":42,"tag":92,"props":318,"children":319},{"style":293},[320],{"type":48,"value":321},"\u002F",{"type":42,"tag":92,"props":323,"children":325},{"style":324},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[326],{"type":48,"value":327},".",{"type":42,"tag":92,"props":329,"children":330},{"style":287},[331],{"type":48,"value":332},"claude",{"type":42,"tag":92,"props":334,"children":335},{"style":293},[336],{"type":48,"value":321},{"type":42,"tag":92,"props":338,"children":339},{"style":287},[340],{"type":48,"value":341},"data",{"type":42,"tag":92,"props":343,"children":344},{"style":293},[345],{"type":48,"value":321},{"type":42,"tag":92,"props":347,"children":348},{"style":287},[349],{"type":48,"value":19},{"type":42,"tag":92,"props":351,"children":352},{"style":293},[353],{"type":48,"value":354},"}\"\n",{"type":42,"tag":72,"props":356,"children":358},{"id":357},"steps",[359],{"type":48,"value":360},"Steps",{"type":42,"tag":362,"props":363,"children":364},"ol",{},[365,427,581,717,851,1157,1532,1567],{"type":42,"tag":83,"props":366,"children":367},{},[368,379,381,387,389,395,397,403,405,411,413,419,421,425],{"type":42,"tag":57,"props":369,"children":370},{},[371,373,378],{"type":48,"value":372},"Get the payer's claims into ",{"type":42,"tag":112,"props":374,"children":376},{"className":375},[],[377],{"type":48,"value":144},{"type":48,"value":327},{"type":48,"value":380}," Open with: ",{"type":42,"tag":382,"props":383,"children":384},"em",{},[385],{"type":48,"value":386},"\"Where do your adjudicated claims\nlive?\"",{"type":48,"value":388}," and follow ",{"type":42,"tag":112,"props":390,"children":392},{"className":391},[],[393],{"type":48,"value":394},"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002FLOAD-CLAIMS.md",{"type":48,"value":396}," — it walks you\nand the user from \"I don't know\" to a populated ",{"type":42,"tag":112,"props":398,"children":400},{"className":399},[],[401],{"type":48,"value":402},"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Fcorpus.duckdb",{"type":48,"value":404},". If they\nalready have a ",{"type":42,"tag":112,"props":406,"children":408},{"className":407},[],[409],{"type":48,"value":410},".duckdb",{"type":48,"value":412}," with the canonical tables (schema:\n",{"type":42,"tag":112,"props":414,"children":416},{"className":415},[],[417],{"type":48,"value":418},"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fclaims-schema.sql",{"type":48,"value":420},"), use it directly.",{"type":42,"tag":422,"props":423,"children":424},"br",{},[],{"type":48,"value":426},"Draft the brief (corpusDb path, quarter, line of business) and confirm scope.",{"type":42,"tag":83,"props":428,"children":429},{},[430,435,437,443,445,451,453,529,532,534,540,541,547,549,555,557,563,565,571,573,579],{"type":42,"tag":57,"props":431,"children":432},{},[433],{"type":48,"value":434},"Seed the public reference layer (first run \u002F new quarter only).",{"type":48,"value":436}," Detectors cite against\n",{"type":42,"tag":112,"props":438,"children":440},{"className":439},[],[441],{"type":48,"value":442},"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Freference\u002F\u003Cquarter>\u002Freference.duckdb",{"type":48,"value":444},". If that file is missing for the\nrequested quarter, fetch it now — this prints per-source ",{"type":42,"tag":112,"props":446,"children":448},{"className":447},[],[449],{"type":48,"value":450},"✓ name (size)",{"type":48,"value":452}," progress as ~34 sources land:",{"type":42,"tag":264,"props":454,"children":456},{"className":266,"code":455,"language":268,"meta":269,"style":269},"node \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Ffetch-reference.js\" 2026q3\nnode \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Ffetch-enrichment.js\"\n",[457],{"type":42,"tag":112,"props":458,"children":459},{"__ignoreMap":269},[460,499],{"type":42,"tag":92,"props":461,"children":462},{"class":276,"line":277},[463,469,474,479,484,489,494],{"type":42,"tag":92,"props":464,"children":466},{"style":465},"--shiki-light:#E2931D;--shiki-default:#FFCB6B;--shiki-dark:#FFCB6B",[467],{"type":48,"value":468},"node",{"type":42,"tag":92,"props":470,"children":471},{"style":293},[472],{"type":48,"value":473}," \"${",{"type":42,"tag":92,"props":475,"children":476},{"style":287},[477],{"type":48,"value":478},"CLAUDE_PLUGIN_ROOT",{"type":42,"tag":92,"props":480,"children":481},{"style":293},[482],{"type":48,"value":483},"}",{"type":42,"tag":92,"props":485,"children":486},{"style":324},[487],{"type":48,"value":488},"\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Ffetch-reference.js",{"type":42,"tag":92,"props":490,"children":491},{"style":293},[492],{"type":48,"value":493},"\"",{"type":42,"tag":92,"props":495,"children":496},{"style":324},[497],{"type":48,"value":498}," 2026q3\n",{"type":42,"tag":92,"props":500,"children":502},{"class":276,"line":501},2,[503,507,511,515,519,524],{"type":42,"tag":92,"props":504,"children":505},{"style":465},[506],{"type":48,"value":468},{"type":42,"tag":92,"props":508,"children":509},{"style":293},[510],{"type":48,"value":473},{"type":42,"tag":92,"props":512,"children":513},{"style":287},[514],{"type":48,"value":478},{"type":42,"tag":92,"props":516,"children":517},{"style":293},[518],{"type":48,"value":483},{"type":42,"tag":92,"props":520,"children":521},{"style":324},[522],{"type":48,"value":523},"\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Ffetch-enrichment.js",{"type":42,"tag":92,"props":525,"children":526},{"style":293},[527],{"type":48,"value":528},"\"\n",{"type":42,"tag":422,"props":530,"children":531},{},[],{"type":48,"value":533},"Requires ",{"type":42,"tag":112,"props":535,"children":537},{"className":536},[],[538],{"type":48,"value":539},"unzip",{"type":48,"value":246},{"type":42,"tag":112,"props":542,"children":544},{"className":543},[],[545],{"type":48,"value":546},"pdftotext",{"type":48,"value":548}," (poppler) on PATH; both ship with most distros \u002F ",{"type":42,"tag":112,"props":550,"children":552},{"className":551},[],[553],{"type":48,"value":554},"brew install poppler",{"type":48,"value":556},". Needs real network egress — if you see \"Could not resolve host\" for cms.gov \u002F oig.hhs.gov,\nthe command sandbox is blocking it; re-run with sandbox disabled. Policy PDFs (NCCI manual, MLN articles) land under ",{"type":42,"tag":112,"props":558,"children":560},{"className":559},[],[561],{"type":48,"value":562},"reference\u002F\u003Cq>\u002Fpolicy\u002F*.txt",{"type":48,"value":564},"\nfor grep; everything keyed lands in ",{"type":42,"tag":112,"props":566,"children":568},{"className":567},[],[569],{"type":48,"value":570},"reference.duckdb",{"type":48,"value":572},". Skip if already present. If a fetch fails\nor a table is missing, see ",{"type":42,"tag":112,"props":574,"children":576},{"className":575},[],[577],{"type":48,"value":578},"REFERENCE-DATA.md",{"type":48,"value":580}," for source URLs and recovery.",{"type":42,"tag":83,"props":582,"children":583},{},[584,589,591,597,599,704,707,709,715],{"type":42,"tag":57,"props":585,"children":586},{},[587],{"type":48,"value":588},"Create the run directory.",{"type":48,"value":590}," Each invocation lands in its own minute-stamped directory so prior\nruns are preserved side-by-side. Every script honors ",{"type":42,"tag":112,"props":592,"children":594},{"className":593},[],[595],{"type":48,"value":596},"FRAUD_OUT_DIR",{"type":48,"value":598},":",{"type":42,"tag":264,"props":600,"children":602},{"className":266,"code":601,"language":268,"meta":269,"style":269},"export FRAUD_OUT_DIR=\"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fout\u002Frun-$(date +%Y%m%d-%H%M)\"\nmkdir -p \"$FRAUD_OUT_DIR\"\necho \"$FRAUD_OUT_DIR\"\n",[603],{"type":42,"tag":112,"props":604,"children":605},{"__ignoreMap":269},[606,655,682],{"type":42,"tag":92,"props":607,"children":608},{"class":276,"line":277},[609,613,618,622,626,630,635,640,645,650],{"type":42,"tag":92,"props":610,"children":611},{"style":281},[612],{"type":48,"value":284},{"type":42,"tag":92,"props":614,"children":615},{"style":287},[616],{"type":48,"value":617}," FRAUD_OUT_DIR",{"type":42,"tag":92,"props":619,"children":620},{"style":293},[621],{"type":48,"value":296},{"type":42,"tag":92,"props":623,"children":624},{"style":293},[625],{"type":48,"value":493},{"type":42,"tag":92,"props":627,"children":628},{"style":287},[629],{"type":48,"value":236},{"type":42,"tag":92,"props":631,"children":632},{"style":324},[633],{"type":48,"value":634},"\u002Ffraud-detection\u002Fout\u002Frun-",{"type":42,"tag":92,"props":636,"children":637},{"style":293},[638],{"type":48,"value":639},"$(",{"type":42,"tag":92,"props":641,"children":642},{"style":465},[643],{"type":48,"value":644},"date",{"type":42,"tag":92,"props":646,"children":647},{"style":324},[648],{"type":48,"value":649}," +%Y%m%d-%H%M",{"type":42,"tag":92,"props":651,"children":652},{"style":293},[653],{"type":48,"value":654},")\"\n",{"type":42,"tag":92,"props":656,"children":657},{"class":276,"line":501},[658,663,668,673,678],{"type":42,"tag":92,"props":659,"children":660},{"style":465},[661],{"type":48,"value":662},"mkdir",{"type":42,"tag":92,"props":664,"children":665},{"style":324},[666],{"type":48,"value":667}," -p",{"type":42,"tag":92,"props":669,"children":670},{"style":293},[671],{"type":48,"value":672}," \"",{"type":42,"tag":92,"props":674,"children":675},{"style":287},[676],{"type":48,"value":677},"$FRAUD_OUT_DIR",{"type":42,"tag":92,"props":679,"children":680},{"style":293},[681],{"type":48,"value":528},{"type":42,"tag":92,"props":683,"children":685},{"class":276,"line":684},3,[686,692,696,700],{"type":42,"tag":92,"props":687,"children":689},{"style":688},"--shiki-light:#6182B8;--shiki-default:#82AAFF;--shiki-dark:#82AAFF",[690],{"type":48,"value":691},"echo",{"type":42,"tag":92,"props":693,"children":694},{"style":293},[695],{"type":48,"value":672},{"type":42,"tag":92,"props":697,"children":698},{"style":287},[699],{"type":48,"value":677},{"type":42,"tag":92,"props":701,"children":702},{"style":293},[703],{"type":48,"value":528},{"type":42,"tag":422,"props":705,"children":706},{},[],{"type":48,"value":708},"Use the printed absolute path verbatim as ",{"type":42,"tag":112,"props":710,"children":712},{"className":711},[],[713],{"type":48,"value":714},"outDir",{"type":48,"value":716}," in the next step.",{"type":42,"tag":83,"props":718,"children":719},{},[720,725,727,732,734,770,773,775,839,842,844,850],{"type":42,"tag":57,"props":721,"children":722},{},[723],{"type":48,"value":724},"Run the investigation",{"type":48,"value":726}," by calling the ",{"type":42,"tag":57,"props":728,"children":729},{},[730],{"type":48,"value":731},"Workflow",{"type":48,"value":733}," tool with:",{"type":42,"tag":79,"props":735,"children":736},{},[737,754],{"type":42,"tag":83,"props":738,"children":739},{},[740,746,748],{"type":42,"tag":112,"props":741,"children":743},{"className":742},[],[744],{"type":48,"value":745},"scriptPath",{"type":48,"value":747},": ",{"type":42,"tag":112,"props":749,"children":751},{"className":750},[],[752],{"type":48,"value":753},"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fworkflows\u002Finvestigate.js",{"type":42,"tag":83,"props":755,"children":756},{},[757,763,764],{"type":42,"tag":112,"props":758,"children":760},{"className":759},[],[761],{"type":48,"value":762},"args",{"type":48,"value":747},{"type":42,"tag":112,"props":765,"children":767},{"className":766},[],[768],{"type":48,"value":769},"{ \"corpusDb\": \"\u003Cabs path to corpus.duckdb>\", \"quarter\": \"2026q3\", \"lob\": \"medicaid\", \"pluginRoot\": \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\", \"dataRoot\": \"\u003Cabs $CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection>\", \"outDir\": \"\u003Cabs FRAUD_OUT_DIR from step 3>\" }",{"type":42,"tag":422,"props":771,"children":772},{},[],{"type":48,"value":774},"The workflow runs three stages (see \"How it works\"):",{"type":42,"tag":79,"props":776,"children":777},{},[778,803,829],{"type":42,"tag":83,"props":779,"children":780},{},[781,786,788,794,796,802],{"type":42,"tag":57,"props":782,"children":783},{},[784],{"type":48,"value":785},"Detect",{"type":48,"value":787}," — runs the deterministic sweep (",{"type":42,"tag":112,"props":789,"children":791},{"className":790},[],[792],{"type":48,"value":793},"scripts\u002Fscreen.js",{"type":48,"value":795},", zero model) → ",{"type":42,"tag":112,"props":797,"children":799},{"className":798},[],[800],{"type":48,"value":801},"$FRAUD_OUT_DIR\u002Freferrals.detect.json",{"type":48,"value":327},{"type":42,"tag":83,"props":804,"children":805},{},[806,811,813,819,821,827],{"type":42,"tag":57,"props":807,"children":808},{},[809],{"type":48,"value":810},"Adjudicate",{"type":48,"value":812}," — one agent per judgment-required finding (D2\u002FD4\u002FD7\u002FD13) sets ",{"type":42,"tag":112,"props":814,"children":816},{"className":815},[],[817],{"type":48,"value":818},"status",{"type":48,"value":820}," + ",{"type":42,"tag":112,"props":822,"children":824},{"className":823},[],[825],{"type":48,"value":826},"adjudication.reason",{"type":48,"value":828},"; mechanical detectors auto-confirm. Adjudicate may dismiss or downgrade, never add.",{"type":42,"tag":83,"props":830,"children":831},{},[832,837],{"type":42,"tag":57,"props":833,"children":834},{},[835],{"type":48,"value":836},"Synthesize",{"type":48,"value":838}," — one agent per provider writes the investigator narrative, plus novel-lead discovery with adversarial verification.",{"type":42,"tag":422,"props":840,"children":841},{},[],{"type":48,"value":843},"Tell the user they can watch the fan-out live with ",{"type":42,"tag":112,"props":845,"children":847},{"className":846},[],[848],{"type":48,"value":849},"\u002Fworkflows",{"type":48,"value":327},{"type":42,"tag":83,"props":852,"children":853},{},[854,859,861,867,869,879,881,886,888,894,896,959,962,964,970,972,978,979,985,987,993,994,1000,1001,1004,1006,1130,1133,1135,1141,1143,1149,1150,1156],{"type":42,"tag":57,"props":855,"children":856},{},[857],{"type":48,"value":858},"Materialize the stage snapshots + render",{"type":48,"value":860}," (required — this is the reviewable deliverable). The\nworkflow sandbox has no filesystem, so write its return to disk and let ",{"type":42,"tag":112,"props":862,"children":864},{"className":863},[],[865],{"type":48,"value":866},"apply-stages.js",{"type":48,"value":868}," produce\nthe auditable spine. ",{"type":42,"tag":57,"props":870,"children":871},{},[872,877],{"type":42,"tag":112,"props":873,"children":875},{"className":874},[],[876],{"type":48,"value":596},{"type":48,"value":878}," does not persist across separate Bash calls",{"type":48,"value":880}," — re-export\nit (to the same absolute path you printed in step 3) at the top of every shell block that needs it:\nUse the ",{"type":42,"tag":57,"props":882,"children":883},{},[884],{"type":48,"value":885},"Write",{"type":48,"value":887}," tool to save the workflow's return JSON verbatim to\n",{"type":42,"tag":112,"props":889,"children":891},{"className":890},[],[892],{"type":48,"value":893},"$FRAUD_OUT_DIR\u002Fworkflow-result.json",{"type":48,"value":895}," (it can be 50KB+ — don't heredoc it through Bash). Then:",{"type":42,"tag":264,"props":897,"children":899},{"className":266,"code":898,"language":268,"meta":269,"style":269},"export FRAUD_OUT_DIR=\"\u003Cabs path from step 3>\"\nnode \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Fapply-stages.js\"\n",[900],{"type":42,"tag":112,"props":901,"children":902},{"__ignoreMap":269},[903,931],{"type":42,"tag":92,"props":904,"children":905},{"class":276,"line":277},[906,910,914,918,922,927],{"type":42,"tag":92,"props":907,"children":908},{"style":281},[909],{"type":48,"value":284},{"type":42,"tag":92,"props":911,"children":912},{"style":287},[913],{"type":48,"value":617},{"type":42,"tag":92,"props":915,"children":916},{"style":293},[917],{"type":48,"value":296},{"type":42,"tag":92,"props":919,"children":920},{"style":293},[921],{"type":48,"value":493},{"type":42,"tag":92,"props":923,"children":924},{"style":324},[925],{"type":48,"value":926},"\u003Cabs path from step 3>",{"type":42,"tag":92,"props":928,"children":929},{"style":293},[930],{"type":48,"value":528},{"type":42,"tag":92,"props":932,"children":933},{"class":276,"line":501},[934,938,942,946,950,955],{"type":42,"tag":92,"props":935,"children":936},{"style":465},[937],{"type":48,"value":468},{"type":42,"tag":92,"props":939,"children":940},{"style":293},[941],{"type":48,"value":473},{"type":42,"tag":92,"props":943,"children":944},{"style":287},[945],{"type":48,"value":478},{"type":42,"tag":92,"props":947,"children":948},{"style":293},[949],{"type":48,"value":483},{"type":42,"tag":92,"props":951,"children":952},{"style":324},[953],{"type":48,"value":954},"\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Fapply-stages.js",{"type":42,"tag":92,"props":956,"children":957},{"style":293},[958],{"type":48,"value":528},{"type":42,"tag":422,"props":960,"children":961},{},[],{"type":48,"value":963},"This writes ",{"type":42,"tag":112,"props":965,"children":967},{"className":966},[],[968],{"type":48,"value":969},"$FRAUD_OUT_DIR\u002Freferrals.adjudicated.json",{"type":48,"value":971},", ",{"type":42,"tag":112,"props":973,"children":975},{"className":974},[],[976],{"type":48,"value":977},"referrals.final.json",{"type":48,"value":971},{"type":42,"tag":112,"props":980,"children":982},{"className":981},[],[983],{"type":48,"value":984},"referrals.json",{"type":48,"value":986},"\n(canonical, = final), and the renderer sidecars (",{"type":42,"tag":112,"props":988,"children":990},{"className":989},[],[991],{"type":48,"value":992},"source-excerpts.json",{"type":48,"value":971},{"type":42,"tag":112,"props":995,"children":997},{"className":996},[],[998],{"type":48,"value":999},"providers.json",{"type":48,"value":179},{"type":42,"tag":422,"props":1002,"children":1003},{},[],{"type":48,"value":1005},"Then render the packets FIRST, then the dashboard (the dashboard only links a provider row to its\npacket if that packet file already exists), then the xlsx:",{"type":42,"tag":264,"props":1007,"children":1009},{"className":266,"code":1008,"language":268,"meta":269,"style":269},"export FRAUD_OUT_DIR=\"\u003Cabs path from step 3>\"\nnode \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-packet.js\" --all\nnode \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-dashboard.js\"\nnode \"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-xlsx.js\"\n",[1010],{"type":42,"tag":112,"props":1011,"children":1012},{"__ignoreMap":269},[1013,1040,1073,1101],{"type":42,"tag":92,"props":1014,"children":1015},{"class":276,"line":277},[1016,1020,1024,1028,1032,1036],{"type":42,"tag":92,"props":1017,"children":1018},{"style":281},[1019],{"type":48,"value":284},{"type":42,"tag":92,"props":1021,"children":1022},{"style":287},[1023],{"type":48,"value":617},{"type":42,"tag":92,"props":1025,"children":1026},{"style":293},[1027],{"type":48,"value":296},{"type":42,"tag":92,"props":1029,"children":1030},{"style":293},[1031],{"type":48,"value":493},{"type":42,"tag":92,"props":1033,"children":1034},{"style":324},[1035],{"type":48,"value":926},{"type":42,"tag":92,"props":1037,"children":1038},{"style":293},[1039],{"type":48,"value":528},{"type":42,"tag":92,"props":1041,"children":1042},{"class":276,"line":501},[1043,1047,1051,1055,1059,1064,1068],{"type":42,"tag":92,"props":1044,"children":1045},{"style":465},[1046],{"type":48,"value":468},{"type":42,"tag":92,"props":1048,"children":1049},{"style":293},[1050],{"type":48,"value":473},{"type":42,"tag":92,"props":1052,"children":1053},{"style":287},[1054],{"type":48,"value":478},{"type":42,"tag":92,"props":1056,"children":1057},{"style":293},[1058],{"type":48,"value":483},{"type":42,"tag":92,"props":1060,"children":1061},{"style":324},[1062],{"type":48,"value":1063},"\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-packet.js",{"type":42,"tag":92,"props":1065,"children":1066},{"style":293},[1067],{"type":48,"value":493},{"type":42,"tag":92,"props":1069,"children":1070},{"style":324},[1071],{"type":48,"value":1072}," --all\n",{"type":42,"tag":92,"props":1074,"children":1075},{"class":276,"line":684},[1076,1080,1084,1088,1092,1097],{"type":42,"tag":92,"props":1077,"children":1078},{"style":465},[1079],{"type":48,"value":468},{"type":42,"tag":92,"props":1081,"children":1082},{"style":293},[1083],{"type":48,"value":473},{"type":42,"tag":92,"props":1085,"children":1086},{"style":287},[1087],{"type":48,"value":478},{"type":42,"tag":92,"props":1089,"children":1090},{"style":293},[1091],{"type":48,"value":483},{"type":42,"tag":92,"props":1093,"children":1094},{"style":324},[1095],{"type":48,"value":1096},"\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-dashboard.js",{"type":42,"tag":92,"props":1098,"children":1099},{"style":293},[1100],{"type":48,"value":528},{"type":42,"tag":92,"props":1102,"children":1104},{"class":276,"line":1103},4,[1105,1109,1113,1117,1121,1126],{"type":42,"tag":92,"props":1106,"children":1107},{"style":465},[1108],{"type":48,"value":468},{"type":42,"tag":92,"props":1110,"children":1111},{"style":293},[1112],{"type":48,"value":473},{"type":42,"tag":92,"props":1114,"children":1115},{"style":287},[1116],{"type":48,"value":478},{"type":42,"tag":92,"props":1118,"children":1119},{"style":293},[1120],{"type":48,"value":483},{"type":42,"tag":92,"props":1122,"children":1123},{"style":324},[1124],{"type":48,"value":1125},"\u002Fskills\u002Ffraud-detection\u002Fscripts\u002Frender-xlsx.js",{"type":42,"tag":92,"props":1127,"children":1128},{"style":293},[1129],{"type":48,"value":528},{"type":42,"tag":422,"props":1131,"children":1132},{},[],{"type":48,"value":1134},"→ ",{"type":42,"tag":112,"props":1136,"children":1138},{"className":1137},[],[1139],{"type":48,"value":1140},"$FRAUD_OUT_DIR\u002Fprovider-packet-\u003Cnpi>.html",{"type":48,"value":1142}," ×N, ",{"type":42,"tag":112,"props":1144,"children":1146},{"className":1145},[],[1147],{"type":48,"value":1148},"index.html",{"type":48,"value":971},{"type":42,"tag":112,"props":1151,"children":1153},{"className":1152},[],[1154],{"type":48,"value":1155},"referrals.xlsx",{"type":48,"value":327},{"type":42,"tag":83,"props":1158,"children":1159},{},[1160,1165,1167,1527,1530],{"type":42,"tag":57,"props":1161,"children":1162},{},[1163],{"type":48,"value":1164},"Show the dashboard",{"type":48,"value":1166}," so the user can validate it visually.",{"type":42,"tag":79,"props":1168,"children":1169},{},[1170,1195],{"type":42,"tag":83,"props":1171,"children":1172},{},[1173,1178,1180,1186,1188,1193],{"type":42,"tag":57,"props":1174,"children":1175},{},[1176],{"type":48,"value":1177},"Claude Code Desktop with the preview tool available",{"type":48,"value":1179},": serve the run directory and open it\nin the side-pane preview — ",{"type":42,"tag":112,"props":1181,"children":1183},{"className":1182},[],[1184],{"type":48,"value":1185},"npx serve \"$FRAUD_OUT_DIR\"",{"type":48,"value":1187}," (the dashboard is ",{"type":42,"tag":112,"props":1189,"children":1191},{"className":1190},[],[1192],{"type":48,"value":1148},{"type":48,"value":1194},",\nso the root URL is the dashboard; packet links resolve as siblings).",{"type":42,"tag":83,"props":1196,"children":1197},{},[1198,1203,1205,1211,1213],{"type":42,"tag":57,"props":1199,"children":1200},{},[1201],{"type":48,"value":1202},"Otherwise",{"type":48,"value":1204}," (terminal CLI, no preview tool): use the OS opener on\n",{"type":42,"tag":112,"props":1206,"children":1208},{"className":1207},[],[1209],{"type":48,"value":1210},"$FRAUD_OUT_DIR\u002Findex.html",{"type":48,"value":1212}," — try in order, stop at the first that works:\n",{"type":42,"tag":264,"props":1214,"children":1216},{"className":266,"code":1215,"language":268,"meta":269,"style":269},"f=\"$FRAUD_OUT_DIR\u002Findex.html\"\nopen \"$f\" 2>\u002Fdev\u002Fnull \\            # macOS\n  || xdg-open \"$f\" 2>\u002Fdev\u002Fnull \\   # Linux\n  || wslview \"$f\" 2>\u002Fdev\u002Fnull \\    # WSL\n  || powershell.exe start \"$(wslpath -w \"$f\")\" 2>\u002Fdev\u002Fnull \\  # WSL→Windows fallback\n  || cmd.exe \u002Fc start \"\" \"$f\" 2>\u002Fdev\u002Fnull \\                   # Windows\n  || echo \"Could not auto-open; open manually: $f\"\n",[1217],{"type":42,"tag":112,"props":1218,"children":1219},{"__ignoreMap":269},[1220,1249,1291,1334,1376,1440,1497],{"type":42,"tag":92,"props":1221,"children":1222},{"class":276,"line":277},[1223,1228,1232,1236,1240,1245],{"type":42,"tag":92,"props":1224,"children":1225},{"style":287},[1226],{"type":48,"value":1227},"f",{"type":42,"tag":92,"props":1229,"children":1230},{"style":293},[1231],{"type":48,"value":296},{"type":42,"tag":92,"props":1233,"children":1234},{"style":293},[1235],{"type":48,"value":493},{"type":42,"tag":92,"props":1237,"children":1238},{"style":287},[1239],{"type":48,"value":677},{"type":42,"tag":92,"props":1241,"children":1242},{"style":324},[1243],{"type":48,"value":1244},"\u002Findex.html",{"type":42,"tag":92,"props":1246,"children":1247},{"style":293},[1248],{"type":48,"value":528},{"type":42,"tag":92,"props":1250,"children":1251},{"class":276,"line":501},[1252,1257,1261,1266,1270,1275,1280,1285],{"type":42,"tag":92,"props":1253,"children":1254},{"style":465},[1255],{"type":48,"value":1256},"open",{"type":42,"tag":92,"props":1258,"children":1259},{"style":293},[1260],{"type":48,"value":672},{"type":42,"tag":92,"props":1262,"children":1263},{"style":287},[1264],{"type":48,"value":1265},"$f",{"type":42,"tag":92,"props":1267,"children":1268},{"style":293},[1269],{"type":48,"value":493},{"type":42,"tag":92,"props":1271,"children":1272},{"style":293},[1273],{"type":48,"value":1274}," 2>",{"type":42,"tag":92,"props":1276,"children":1277},{"style":324},[1278],{"type":48,"value":1279},"\u002Fdev\u002Fnull",{"type":42,"tag":92,"props":1281,"children":1282},{"style":287},[1283],{"type":48,"value":1284}," \\            ",{"type":42,"tag":92,"props":1286,"children":1288},{"style":1287},"--shiki-light:#90A4AE;--shiki-light-font-style:italic;--shiki-default:#546E7A;--shiki-default-font-style:italic;--shiki-dark:#676E95;--shiki-dark-font-style:italic",[1289],{"type":48,"value":1290},"# macOS\n",{"type":42,"tag":92,"props":1292,"children":1293},{"class":276,"line":684},[1294,1299,1304,1308,1312,1316,1320,1324,1329],{"type":42,"tag":92,"props":1295,"children":1296},{"style":293},[1297],{"type":48,"value":1298},"  ||",{"type":42,"tag":92,"props":1300,"children":1301},{"style":465},[1302],{"type":48,"value":1303}," xdg-open",{"type":42,"tag":92,"props":1305,"children":1306},{"style":293},[1307],{"type":48,"value":672},{"type":42,"tag":92,"props":1309,"children":1310},{"style":287},[1311],{"type":48,"value":1265},{"type":42,"tag":92,"props":1313,"children":1314},{"style":293},[1315],{"type":48,"value":493},{"type":42,"tag":92,"props":1317,"children":1318},{"style":293},[1319],{"type":48,"value":1274},{"type":42,"tag":92,"props":1321,"children":1322},{"style":324},[1323],{"type":48,"value":1279},{"type":42,"tag":92,"props":1325,"children":1326},{"style":287},[1327],{"type":48,"value":1328}," \\   ",{"type":42,"tag":92,"props":1330,"children":1331},{"style":1287},[1332],{"type":48,"value":1333},"# Linux\n",{"type":42,"tag":92,"props":1335,"children":1336},{"class":276,"line":1103},[1337,1341,1346,1350,1354,1358,1362,1366,1371],{"type":42,"tag":92,"props":1338,"children":1339},{"style":293},[1340],{"type":48,"value":1298},{"type":42,"tag":92,"props":1342,"children":1343},{"style":465},[1344],{"type":48,"value":1345}," wslview",{"type":42,"tag":92,"props":1347,"children":1348},{"style":293},[1349],{"type":48,"value":672},{"type":42,"tag":92,"props":1351,"children":1352},{"style":287},[1353],{"type":48,"value":1265},{"type":42,"tag":92,"props":1355,"children":1356},{"style":293},[1357],{"type":48,"value":493},{"type":42,"tag":92,"props":1359,"children":1360},{"style":293},[1361],{"type":48,"value":1274},{"type":42,"tag":92,"props":1363,"children":1364},{"style":324},[1365],{"type":48,"value":1279},{"type":42,"tag":92,"props":1367,"children":1368},{"style":287},[1369],{"type":48,"value":1370}," \\    ",{"type":42,"tag":92,"props":1372,"children":1373},{"style":1287},[1374],{"type":48,"value":1375},"# WSL\n",{"type":42,"tag":92,"props":1377,"children":1379},{"class":276,"line":1378},5,[1380,1384,1389,1394,1399,1404,1409,1413,1417,1422,1426,1430,1435],{"type":42,"tag":92,"props":1381,"children":1382},{"style":293},[1383],{"type":48,"value":1298},{"type":42,"tag":92,"props":1385,"children":1386},{"style":465},[1387],{"type":48,"value":1388}," powershell.exe",{"type":42,"tag":92,"props":1390,"children":1391},{"style":324},[1392],{"type":48,"value":1393}," start",{"type":42,"tag":92,"props":1395,"children":1396},{"style":293},[1397],{"type":48,"value":1398}," \"$(",{"type":42,"tag":92,"props":1400,"children":1401},{"style":465},[1402],{"type":48,"value":1403},"wslpath",{"type":42,"tag":92,"props":1405,"children":1406},{"style":324},[1407],{"type":48,"value":1408}," -w ",{"type":42,"tag":92,"props":1410,"children":1411},{"style":293},[1412],{"type":48,"value":493},{"type":42,"tag":92,"props":1414,"children":1415},{"style":287},[1416],{"type":48,"value":1265},{"type":42,"tag":92,"props":1418,"children":1419},{"style":293},[1420],{"type":48,"value":1421},"\")\"",{"type":42,"tag":92,"props":1423,"children":1424},{"style":293},[1425],{"type":48,"value":1274},{"type":42,"tag":92,"props":1427,"children":1428},{"style":324},[1429],{"type":48,"value":1279},{"type":42,"tag":92,"props":1431,"children":1432},{"style":287},[1433],{"type":48,"value":1434}," \\  ",{"type":42,"tag":92,"props":1436,"children":1437},{"style":1287},[1438],{"type":48,"value":1439},"# WSL→Windows fallback\n",{"type":42,"tag":92,"props":1441,"children":1443},{"class":276,"line":1442},6,[1444,1448,1453,1458,1462,1467,1471,1475,1479,1483,1487,1492],{"type":42,"tag":92,"props":1445,"children":1446},{"style":293},[1447],{"type":48,"value":1298},{"type":42,"tag":92,"props":1449,"children":1450},{"style":465},[1451],{"type":48,"value":1452}," cmd.exe",{"type":42,"tag":92,"props":1454,"children":1455},{"style":324},[1456],{"type":48,"value":1457}," \u002Fc",{"type":42,"tag":92,"props":1459,"children":1460},{"style":324},[1461],{"type":48,"value":1393},{"type":42,"tag":92,"props":1463,"children":1464},{"style":293},[1465],{"type":48,"value":1466}," \"\"",{"type":42,"tag":92,"props":1468,"children":1469},{"style":293},[1470],{"type":48,"value":672},{"type":42,"tag":92,"props":1472,"children":1473},{"style":287},[1474],{"type":48,"value":1265},{"type":42,"tag":92,"props":1476,"children":1477},{"style":293},[1478],{"type":48,"value":493},{"type":42,"tag":92,"props":1480,"children":1481},{"style":293},[1482],{"type":48,"value":1274},{"type":42,"tag":92,"props":1484,"children":1485},{"style":324},[1486],{"type":48,"value":1279},{"type":42,"tag":92,"props":1488,"children":1489},{"style":287},[1490],{"type":48,"value":1491}," \\                   ",{"type":42,"tag":92,"props":1493,"children":1494},{"style":1287},[1495],{"type":48,"value":1496},"# Windows\n",{"type":42,"tag":92,"props":1498,"children":1500},{"class":276,"line":1499},7,[1501,1505,1510,1514,1519,1523],{"type":42,"tag":92,"props":1502,"children":1503},{"style":293},[1504],{"type":48,"value":1298},{"type":42,"tag":92,"props":1506,"children":1507},{"style":688},[1508],{"type":48,"value":1509}," echo",{"type":42,"tag":92,"props":1511,"children":1512},{"style":293},[1513],{"type":48,"value":672},{"type":42,"tag":92,"props":1515,"children":1516},{"style":324},[1517],{"type":48,"value":1518},"Could not auto-open; open manually: ",{"type":42,"tag":92,"props":1520,"children":1521},{"style":287},[1522],{"type":48,"value":1265},{"type":42,"tag":92,"props":1524,"children":1525},{"style":293},[1526],{"type":48,"value":528},{"type":42,"tag":422,"props":1528,"children":1529},{},[],{"type":48,"value":1531},"Skip this entirely in a headless\u002Fnon-interactive run (eval, CI, Cowork) — just report the path.\nDon't fail the run if opening\u002Fserving fails.",{"type":42,"tag":83,"props":1533,"children":1534},{},[1535,1540,1542,1548,1550,1555,1557],{"type":42,"tag":57,"props":1536,"children":1537},{},[1538],{"type":48,"value":1539},"Relay",{"type":48,"value":1541},": the ranked referrals (NPI, schemes, exposure $, confidence) and total exposure from\nthe workflow result, verbatim where it cites numbers. Surface ",{"type":42,"tag":112,"props":1543,"children":1545},{"className":1544},[],[1546],{"type":48,"value":1547},"meta.disclaimer",{"type":48,"value":1549}," if it is set. Do\nnot add any dollar or rule the deterministic floor did not produce. ",{"type":42,"tag":57,"props":1551,"children":1552},{},[1553],{"type":48,"value":1554},"End the response with the\nrun-directory path on its own line",{"type":48,"value":1556}," so downstream graders\u002Ftools can locate the artifacts:",{"type":42,"tag":264,"props":1558,"children":1562},{"className":1559,"code":1561,"language":48},[1560],"language-text","Run directory: \u003Cabsolute $FRAUD_OUT_DIR>\n",[1563],{"type":42,"tag":112,"props":1564,"children":1565},{"__ignoreMap":269},[1566],{"type":48,"value":1561},{"type":42,"tag":83,"props":1568,"children":1569},{},[1570,1575,1577,1583],{"type":42,"tag":57,"props":1571,"children":1572},{},[1573],{"type":48,"value":1574},"Close the loop",{"type":48,"value":1576}," (optional) — see ",{"type":42,"tag":112,"props":1578,"children":1580},{"className":1579},[],[1581],{"type":48,"value":1582},"${CLAUDE_PLUGIN_ROOT}\u002Fskills\u002Ffraud-detection\u002FPROPOSE-DETECTORS.md",{"type":48,"value":1584},"\nto mine this run for new detector candidates and payer-specific adjudicate-time checks.",{"type":42,"tag":72,"props":1586,"children":1588},{"id":1587},"the-inviolable-line",[1589],{"type":48,"value":1590},"The inviolable line",{"type":42,"tag":51,"props":1592,"children":1593},{},[1594,1596,1601,1603,1609,1611,1616],{"type":48,"value":1595},"The model adjudicates, explores, and narrates freely, but ",{"type":42,"tag":57,"props":1597,"children":1598},{},[1599],{"type":48,"value":1600},"any dollar or rule allegation must trace\nto a detect-stage deterministic recompute",{"type":48,"value":1602}," (the gate in ",{"type":42,"tag":112,"props":1604,"children":1606},{"className":1605},[],[1607],{"type":48,"value":1608},"scripts\u002Fgate.js",{"type":48,"value":1610},"). Adjudicate may ",{"type":42,"tag":57,"props":1612,"children":1613},{},[1614],{"type":48,"value":1615},"dismiss\nor downgrade",{"type":48,"value":1617}," a finding (with an auditable reason) — it never adds one or changes its dollars.\nSynthesize narratives are separate, clearly-marked model output and never introduce a number the\nfloor did not compute.",{"type":42,"tag":72,"props":1619,"children":1621},{"id":1620},"enrichment-local-cached-data-canonical-mcps-for-interactive-only",[1622],{"type":48,"value":1623},"Enrichment — local cached data (canonical), MCPs for interactive only",{"type":42,"tag":51,"props":1625,"children":1626},{},[1627,1629,1634,1635,1641,1643,1649,1651,1657,1659,1664],{"type":48,"value":1628},"The deterministic pipeline reads enrichment from ",{"type":42,"tag":57,"props":1630,"children":1631},{},[1632],{"type":48,"value":1633},"local cached files",{"type":48,"value":189},{"type":42,"tag":112,"props":1636,"children":1638},{"className":1637},[],[1639],{"type":48,"value":1640},"scripts\u002Ffetch-enrichment.js",{"type":48,"value":1642},"\n→ ",{"type":42,"tag":112,"props":1644,"children":1646},{"className":1645},[],[1647],{"type":48,"value":1648},"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Fenrichment\u002F",{"type":48,"value":1650},", loaded via ",{"type":42,"tag":112,"props":1652,"children":1654},{"className":1653},[],[1655],{"type":48,"value":1656},"scripts\u002Fenrichment.js",{"type":48,"value":1658},") — no runtime auth, no drift, fully\nreproducible. The healthcare plugin's bundled MCP servers (CMS Coverage \u002F ICD-10 \u002F NPI Registry) are\nfor ",{"type":42,"tag":57,"props":1660,"children":1661},{},[1662],{"type":48,"value":1663},"interactive adjudicate\u002Fsynthesize exploration",{"type":48,"value":1665}," only; the pipeline does not depend on them.",{"type":42,"tag":79,"props":1667,"children":1668},{},[1669,1679,1689],{"type":42,"tag":83,"props":1670,"children":1671},{},[1672,1677],{"type":42,"tag":57,"props":1673,"children":1674},{},[1675],{"type":48,"value":1676},"ICD-10-CM",{"type":48,"value":1678}," — code validity \u002F description (NLM Clinical Tables)",{"type":42,"tag":83,"props":1680,"children":1681},{},[1682,1687],{"type":42,"tag":57,"props":1683,"children":1684},{},[1685],{"type":48,"value":1686},"CMS Coverage (LCD\u002FNCD)",{"type":48,"value":1688}," — medical-necessity policy index; cached, feeds D4 adjudication",{"type":42,"tag":83,"props":1690,"children":1691},{},[1692,1697],{"type":42,"tag":57,"props":1693,"children":1694},{},[1695],{"type":48,"value":1696},"NPI Registry",{"type":48,"value":1698}," — provider taxonomy\u002Fstatus",{"type":42,"tag":72,"props":1700,"children":1702},{"id":1701},"how-it-works-plugin-layout",[1703],{"type":48,"value":1704},"How it works (plugin layout)",{"type":42,"tag":79,"props":1706,"children":1707},{},[1708,1718,1735,1760,1777,1800,1816,1849,1878,1909],{"type":42,"tag":83,"props":1709,"children":1710},{},[1711,1716],{"type":42,"tag":57,"props":1712,"children":1713},{},[1714],{"type":48,"value":1715},"Entry skill",{"type":48,"value":1717}," — this file; orchestrates the workflow, never does the math.",{"type":42,"tag":83,"props":1719,"children":1720},{},[1721,1725,1727,1733],{"type":42,"tag":57,"props":1722,"children":1723},{},[1724],{"type":48,"value":731},{"type":48,"value":1726}," — ",{"type":42,"tag":112,"props":1728,"children":1730},{"className":1729},[],[1731],{"type":48,"value":1732},"workflows\u002Finvestigate.js",{"type":48,"value":1734}," (Claude Code dynamic Workflow): Detect → Adjudicate → Synthesize.",{"type":42,"tag":83,"props":1736,"children":1737},{},[1738,1743,1744,1750,1752,1758],{"type":42,"tag":57,"props":1739,"children":1740},{},[1741],{"type":48,"value":1742},"Deterministic sweep",{"type":48,"value":1726},{"type":42,"tag":112,"props":1745,"children":1747},{"className":1746},[],[1748],{"type":48,"value":1749},"scripts\u002Fscreen.js \u003Ccorpus.duckdb> \u003Cquarter> \u003Clob>",{"type":48,"value":1751}," runs all detectors and\nwrites ",{"type":42,"tag":112,"props":1753,"children":1755},{"className":1754},[],[1756],{"type":48,"value":1757},"$FRAUD_OUT_DIR\u002Freferrals.json",{"type":48,"value":1759},". Zero model calls.",{"type":42,"tag":83,"props":1761,"children":1762},{},[1763,1768,1769,1775],{"type":42,"tag":57,"props":1764,"children":1765},{},[1766],{"type":48,"value":1767},"Detectors",{"type":48,"value":1726},{"type":42,"tag":112,"props":1770,"children":1772},{"className":1771},[],[1773],{"type":48,"value":1774},"scripts\u002FdNN-*.js",{"type":48,"value":1776}," (one deterministic module each, sharing the Finding shape).",{"type":42,"tag":83,"props":1778,"children":1779},{},[1780,1785,1786,1792,1794,1799],{"type":42,"tag":57,"props":1781,"children":1782},{},[1783],{"type":48,"value":1784},"Pipeline",{"type":48,"value":1726},{"type":42,"tag":112,"props":1787,"children":1789},{"className":1788},[],[1790],{"type":48,"value":1791},"scripts\u002Fpipeline.js",{"type":48,"value":1793}," (run → gate → roll up → rank → ",{"type":42,"tag":112,"props":1795,"children":1797},{"className":1796},[],[1798],{"type":48,"value":984},{"type":48,"value":179},{"type":42,"tag":83,"props":1801,"children":1802},{},[1803,1808,1809,1814],{"type":42,"tag":57,"props":1804,"children":1805},{},[1806],{"type":48,"value":1807},"Citation gate",{"type":48,"value":1726},{"type":42,"tag":112,"props":1810,"children":1812},{"className":1811},[],[1813],{"type":48,"value":1608},{"type":48,"value":1815}," (independently recomputes every cited number; uncited or\nnon-reproducing findings are dropped — \"citation-or-zero\").",{"type":42,"tag":83,"props":1817,"children":1818},{},[1819,1824,1825,1831,1833,1839,1841,1847],{"type":42,"tag":57,"props":1820,"children":1821},{},[1822],{"type":48,"value":1823},"Reference data",{"type":48,"value":1726},{"type":42,"tag":112,"props":1826,"children":1828},{"className":1827},[],[1829],{"type":48,"value":1830},"scripts\u002Freference-data.js",{"type":48,"value":1832}," loads ",{"type":42,"tag":112,"props":1834,"children":1836},{"className":1835},[],[1837],{"type":48,"value":1838},"$CLAUDE_HEALTHCARE_DATA\u002Ffraud-detection\u002Fdata-cache\u002Freference\u002F",{"type":48,"value":1840}," (NCCI\u002FMUE, LEIE, PFS,\nenrollment), fetched by ",{"type":42,"tag":112,"props":1842,"children":1844},{"className":1843},[],[1845],{"type":48,"value":1846},"scripts\u002Ffetch-reference.js",{"type":48,"value":1848},", versioned by date-of-service quarter.",{"type":42,"tag":83,"props":1850,"children":1851},{},[1852,1857,1858,1863,1864,1869,1871,1877],{"type":42,"tag":57,"props":1853,"children":1854},{},[1855],{"type":48,"value":1856},"Enrichment",{"type":48,"value":1726},{"type":42,"tag":112,"props":1859,"children":1861},{"className":1860},[],[1862],{"type":48,"value":1656},{"type":48,"value":1832},{"type":42,"tag":112,"props":1865,"children":1867},{"className":1866},[],[1868],{"type":48,"value":1648},{"type":48,"value":1870},", fetched by ",{"type":42,"tag":112,"props":1872,"children":1874},{"className":1873},[],[1875],{"type":48,"value":1876},"fetch-enrichment.js",{"type":48,"value":327},{"type":42,"tag":83,"props":1879,"children":1880},{},[1881,1886,1887,1893,1895,1901,1902,1908],{"type":42,"tag":57,"props":1882,"children":1883},{},[1884],{"type":48,"value":1885},"Stage merge",{"type":48,"value":1726},{"type":42,"tag":112,"props":1888,"children":1890},{"className":1889},[],[1891],{"type":48,"value":1892},"scripts\u002Fapply-stages.js",{"type":48,"value":1894}," (workflow return → ",{"type":42,"tag":112,"props":1896,"children":1898},{"className":1897},[],[1899],{"type":48,"value":1900},"referrals.adjudicated.json",{"type":48,"value":197},{"type":42,"tag":112,"props":1903,"children":1905},{"className":1904},[],[1906],{"type":48,"value":1907},".final.json",{"type":48,"value":179},{"type":42,"tag":83,"props":1910,"children":1911},{},[1912,1917,1918,1924,1926,1931,1933,1939,1940,1946,1948,1954],{"type":42,"tag":57,"props":1913,"children":1914},{},[1915],{"type":48,"value":1916},"Renderers",{"type":48,"value":1726},{"type":42,"tag":112,"props":1919,"children":1921},{"className":1920},[],[1922],{"type":48,"value":1923},"scripts\u002Frender-dashboard.js",{"type":48,"value":1925}," (→ ",{"type":42,"tag":112,"props":1927,"children":1929},{"className":1928},[],[1930],{"type":48,"value":1148},{"type":48,"value":1932},"), ",{"type":42,"tag":112,"props":1934,"children":1936},{"className":1935},[],[1937],{"type":48,"value":1938},"render-packet.js",{"type":48,"value":971},{"type":42,"tag":112,"props":1941,"children":1943},{"className":1942},[],[1944],{"type":48,"value":1945},"render-xlsx.js",{"type":48,"value":1947}," → ",{"type":42,"tag":112,"props":1949,"children":1951},{"className":1950},[],[1952],{"type":48,"value":1953},"$FRAUD_OUT_DIR\u002F",{"type":48,"value":327},{"type":42,"tag":51,"props":1956,"children":1957},{},[1958],{"type":48,"value":1959},"Every allegation cites a public rule with a value the gate independently recomputes, or it is dropped.",{"type":42,"tag":1961,"props":1962,"children":1963},"style",{},[1964],{"type":48,"value":1965},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":1967,"total":2066},[1968,1979,1997,2017,2033,2044,2051],{"slug":1969,"name":1969,"fn":1970,"description":1971,"org":1972,"tags":1973,"stars":26,"repoUrl":27,"updatedAt":1978},"clinical-note-extract-skill","extract structured data from clinical notes","Extract structured data from clinical notes with span-level provenance and null-safety. Use when users say \"extract [variables] from this note\", \"abstract this chart\", \"pull structured data from these notes\", \"what does this note say about [field]\", or when building a chart-abstraction, registry, or cohort dataset from unstructured clinical text.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1974,1977],{"name":1975,"slug":1976,"type":16},"Data Analysis","data-analysis",{"name":18,"slug":19,"type":16},"2026-06-28T07:57:18.332447",{"slug":1980,"name":1980,"fn":1981,"description":1982,"org":1983,"tags":1984,"stars":26,"repoUrl":27,"updatedAt":1996},"contracts","analyze contract documents with citations","Answer a question across a corpus of contract documents with verified citations. Use when the user asks what a contract says, which contracts have a clause, what changed between amendments, or any question that needs reading and citing across a set of contract files. The corpus must be on the local filesystem (see README).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[1985,1987,1990,1993],{"name":1986,"slug":1980,"type":16},"Contracts",{"name":1988,"slug":1989,"type":16},"Documents","documents",{"name":1991,"slug":1992,"type":16},"Legal","legal",{"name":1994,"slug":1995,"type":16},"Research","research","2026-07-18T05:15:17.073689",{"slug":1998,"name":1998,"fn":1999,"description":2000,"org":2001,"tags":2002,"stars":26,"repoUrl":27,"updatedAt":2016},"doc-extract","extract text from documents","Extract plain text from a document file - PDF, DOCX, XLSX, PPTX, RTF, or plain text\u002Fmarkdown\u002FHTML. Use when a binary document needs to be turned into text, for example a contract PDF or an EHR DocumentReference attachment. Other skills (fhir) invoke scripts\u002Fextract.ts directly; the contracts MCP server bundles its own copy (servers\u002Fdocuments\u002Fsrc\u002Fextract.mjs) so its bundle stays self-contained — port fixes to both.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2003,2006,2007,2010,2013],{"name":2004,"slug":2005,"type":16},"Data Cleaning","data-cleaning",{"name":1988,"slug":1989,"type":16},{"name":2008,"slug":2009,"type":16},"DOCX","docx",{"name":2011,"slug":2012,"type":16},"PDF","pdf",{"name":2014,"slug":2015,"type":16},"Spreadsheets","spreadsheets","2026-07-18T05:15:15.766116",{"slug":2018,"name":2018,"fn":2019,"description":2020,"org":2021,"tags":2022,"stars":26,"repoUrl":27,"updatedAt":2032},"fhir","extract clinical data from FHIR servers","Connect to a hospital's FHIR R4 server (Epic, Oracle Health\u002FCerner, MEDITECH, athenahealth, or any SMART-on-FHIR endpoint), pull a patient's clinical data and notes, and extract structured findings. Use when users say \"connect to the EHR\", \"connect to Epic\u002FCerner\", \"pull notes for patient X\", \"what do the last 6 months of notes say about Y\", or any task that starts from a live EHR rather than pasted text.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2023,2026,2028,2029],{"name":2024,"slug":2025,"type":16},"API Development","api-development",{"name":2027,"slug":2018,"type":16},"FHIR",{"name":18,"slug":19,"type":16},{"name":2030,"slug":2031,"type":16},"Interoperability","interoperability","2026-07-03T16:28:01.476883",{"slug":2034,"name":2034,"fn":2035,"description":2036,"org":2037,"tags":2038,"stars":26,"repoUrl":27,"updatedAt":2043},"fhir-developer-skill","build FHIR REST healthcare endpoints","FHIR API development guide for building healthcare endpoints. Use when: (1) Creating FHIR REST endpoints (Patient, Observation, Encounter, Condition, MedicationRequest), (2) Validating FHIR resources and returning proper HTTP status codes and error responses, (3) Implementing SMART on FHIR authorization and OAuth scopes, (4) Working with Bundles, transactions, batch operations, or search pagination. Covers FHIR R4 resource structures, required fields, value sets (status codes, gender, intent), coding systems (LOINC, SNOMED, RxNorm, ICD-10), and OperationOutcome error handling.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2039,2040,2041,2042],{"name":2024,"slug":2025,"type":16},{"name":2027,"slug":2018,"type":16},{"name":18,"slug":19,"type":16},{"name":2030,"slug":2031,"type":16},"2026-06-16T09:39:06.471928",{"slug":4,"name":4,"fn":5,"description":6,"org":2045,"tags":2046,"stars":26,"repoUrl":27,"updatedAt":28},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2047,2048,2049,2050],{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":18,"slug":19,"type":16},{"name":24,"slug":25,"type":16},{"slug":2052,"name":2052,"fn":2053,"description":2054,"org":2055,"tags":2056,"stars":26,"repoUrl":27,"updatedAt":2065},"icd10-cm-skill","extract ICD-10-CM diagnosis codes","Extract billable ICD-10-CM diagnosis codes from a clinical note the way a professional coder builds the claim. Use when users say \"code this encounter\", \"assign ICD-10 codes\", \"what diagnosis codes apply\", \"code this chart\", or when turning clinical documentation into claim-ready diagnosis codes.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2057,2060,2061,2062],{"name":2058,"slug":2059,"type":16},"Coding","coding",{"name":18,"slug":19,"type":16},{"name":24,"slug":25,"type":16},{"name":2063,"slug":2064,"type":16},"Medical Necessity","medical-necessity","2026-06-19T09:34:31.578353",9,{"items":2068,"total":2247},[2069,2090,2104,2114,2133,2146,2163,2183,2197,2210,2218,2231],{"slug":2070,"name":2070,"fn":2071,"description":2072,"org":2073,"tags":2074,"stars":2087,"repoUrl":2088,"updatedAt":2089},"algorithmic-art","create algorithmic art with p5.js","Creating algorithmic art using p5.js with seeded randomness and interactive parameter exploration. Use this when users request creating art using code, generative art, algorithmic art, flow fields, or particle systems. Create original algorithmic art rather than copying existing artists' work to avoid copyright violations.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2075,2078,2081,2084],{"name":2076,"slug":2077,"type":16},"Creative","creative",{"name":2079,"slug":2080,"type":16},"Design","design",{"name":2082,"slug":2083,"type":16},"Generative Art","generative-art",{"name":2085,"slug":2086,"type":16},"JavaScript","javascript",161831,"https:\u002F\u002Fgithub.com\u002Fanthropics\u002Fskills","2026-04-06T17:56:15.455818",{"slug":2091,"name":2091,"fn":2092,"description":2093,"org":2094,"tags":2095,"stars":2087,"repoUrl":2088,"updatedAt":2103},"brand-guidelines","apply Anthropic brand colors and typography","Applies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2096,2099,2100],{"name":2097,"slug":2098,"type":16},"Branding","branding",{"name":2079,"slug":2080,"type":16},{"name":2101,"slug":2102,"type":16},"Typography","typography","2026-04-06T17:56:05.042852",{"slug":2105,"name":2105,"fn":2106,"description":2107,"org":2108,"tags":2109,"stars":2087,"repoUrl":2088,"updatedAt":2113},"canvas-design","create posters and visual art as PNG or PDF","Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2110,2111,2112],{"name":2076,"slug":2077,"type":16},{"name":2079,"slug":2080,"type":16},{"name":2011,"slug":2012,"type":16},"2026-04-06T17:56:03.794732",{"slug":2115,"name":2115,"fn":2116,"description":2117,"org":2118,"tags":2119,"stars":2087,"repoUrl":2088,"updatedAt":2132},"claude-api","build apps with the Claude API","Reference for the Claude API \u002F Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration.\nTRIGGER — read BEFORE opening the target file; don't skip because it \"looks like a one-liner\" — whenever: the prompt names Claude\u002FAnthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing\u002Fmodel choice\u002Flimits\u002Fcaching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent\u002FMCP\u002Ftool-definition\u002Fmulti-agent\u002FRAG\u002FLLM-judge\u002Fcomputer-use; generate\u002Fsummarize\u002Fextract\u002Fclassify\u002Frewrite\u002Fconverse over NL; debugging refusals\u002Fcutoffs\u002Fstreaming\u002Ftool-calls\u002Ftokens).\nSKIP only when another provider is being worked on (overrides all triggers): OpenAI\u002FGPT\u002FGemini\u002FLlama\u002FMistral\u002FCohere\u002FOllama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2120,2123,2124,2127,2129],{"name":2121,"slug":2122,"type":16},"Agents","agents",{"name":9,"slug":8,"type":16},{"name":2125,"slug":2126,"type":16},"Anthropic SDK","anthropic-sdk",{"name":2128,"slug":2115,"type":16},"Claude API",{"name":2130,"slug":2131,"type":16},"LLM","llm","2026-07-28T05:36:08.213335",{"slug":2134,"name":2134,"fn":2135,"description":2136,"org":2137,"tags":2138,"stars":2087,"repoUrl":2088,"updatedAt":2145},"doc-coauthoring","co-author documentation and technical specs","Guide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2139,2142],{"name":2140,"slug":2141,"type":16},"Documentation","documentation",{"name":2143,"slug":2144,"type":16},"Technical Writing","technical-writing","2026-04-06T17:56:14.18897",{"slug":2009,"name":2009,"fn":2147,"description":2148,"org":2149,"tags":2150,"stars":2087,"repoUrl":2088,"updatedAt":2162},"create and edit Word documents","Use this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2151,2152,2153,2156,2159],{"name":1988,"slug":1989,"type":16},{"name":2008,"slug":2009,"type":16},{"name":2154,"slug":2155,"type":16},"Office","office",{"name":2157,"slug":2158,"type":16},"Templates","templates",{"name":2160,"slug":2161,"type":16},"Word","word","2026-07-18T05:16:23.136271",{"slug":2164,"name":2164,"fn":2165,"description":2166,"org":2167,"tags":2168,"stars":2087,"repoUrl":2088,"updatedAt":2182},"frontend-design","design production-grade frontend interfaces","Guidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2169,2170,2173,2176,2179],{"name":2079,"slug":2080,"type":16},{"name":2171,"slug":2172,"type":16},"Frontend","frontend",{"name":2174,"slug":2175,"type":16},"React","react",{"name":2177,"slug":2178,"type":16},"Tailwind CSS","tailwind-css",{"name":2180,"slug":2181,"type":16},"UI Components","ui-components","2026-04-06T17:56:16.723469",{"slug":2184,"name":2184,"fn":2185,"description":2186,"org":2187,"tags":2188,"stars":2087,"repoUrl":2088,"updatedAt":2196},"internal-comms","write internal company communications","A set of resources to help me write all kinds of internal communications, using the formats that my company likes to use. Claude should use this skill whenever asked to write some sort of internal communications (status reports, leadership updates, 3P updates, company newsletters, FAQs, incident reports, project updates, etc.).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2189,2192,2193],{"name":2190,"slug":2191,"type":16},"Communications","communications",{"name":2157,"slug":2158,"type":16},{"name":2194,"slug":2195,"type":16},"Writing","writing","2026-04-06T17:56:20.695522",{"slug":2198,"name":2198,"fn":2199,"description":2200,"org":2201,"tags":2202,"stars":2087,"repoUrl":2088,"updatedAt":2209},"mcp-builder","build MCP servers","Guide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node\u002FTypeScript (MCP SDK).",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2203,2204,2205,2206],{"name":2121,"slug":2122,"type":16},{"name":2024,"slug":2025,"type":16},{"name":2130,"slug":2131,"type":16},{"name":2207,"slug":2208,"type":16},"MCP","mcp","2026-04-06T17:56:10.357665",{"slug":2012,"name":2012,"fn":2211,"description":2212,"org":2213,"tags":2214,"stars":2087,"repoUrl":2088,"updatedAt":2217},"read edit and manipulate PDF files","Use this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text\u002Ftables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting\u002Fdecrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2215,2216],{"name":1988,"slug":1989,"type":16},{"name":2011,"slug":2012,"type":16},"2026-04-06T17:56:02.483316",{"slug":2219,"name":2219,"fn":2220,"description":2221,"org":2222,"tags":2223,"stars":2087,"repoUrl":2088,"updatedAt":2230},"pptx","create and edit PowerPoint presentations","Use this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions \"deck,\" \"slides,\" \"presentation,\" or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2224,2227],{"name":2225,"slug":2226,"type":16},"PowerPoint","powerpoint",{"name":2228,"slug":2229,"type":16},"Presentations","presentations","2026-07-18T05:16:24.1471",{"slug":2232,"name":2232,"fn":2233,"description":2234,"org":2235,"tags":2236,"stars":2087,"repoUrl":2088,"updatedAt":2246},"skill-creator","create and optimize agent skills","Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[2237,2238,2239,2242,2245],{"name":2121,"slug":2122,"type":16},{"name":2140,"slug":2141,"type":16},{"name":2240,"slug":2241,"type":16},"Evals","evals",{"name":2243,"slug":2244,"type":16},"Performance","performance",{"name":2143,"slug":2144,"type":16},"2026-04-19T06:45:40.804",490]