Research & Analysis Workflows

Turns raw health materials (PDFs, scanned documents, images, spreadsheets, exported files) into two layers: first a faithful, fully traceable extraction of each source; then a set of cross-source compiled views (timeline, lab-trends, and others) built once for pattern-spotting without re-burning tokens. Every item traces back to a verbatim source quote. Reorganises data; never interprets, concludes, or flags. Use when handed raw health materials that need to become structured, traceable, pattern-ready data before any analysis. Called by /investigate-health Step 1, or directly when extraction is the whole task. Triggers: "extract this data", "process these records", "pull the values out of these PDFs", "/extract-health-data".

Autonomous procedure for investigating someone's complex multi-system health picture. Works backward from what the person is actually experiencing to the underlying processes that could be producing it, narrows down by evidence (including direct interview with the person), and offers them possibilities they could consider trialing or testing — never in directive language, always as options to discuss with their clinician. Use when handed a person's health records, lab results, or symptom history with the goal of understanding what might be going on and what they might consider next. Triggers: "investigate this person's health", "help me biohack X", "work out what might be going on with Y", "/investigate-health". Not for single research questions — that is /research, which this procedure calls.

Deep-verified supplement and product recommendations that enforce ingredient-by-ingredient cross-checking against the user's documented constraints in MEMORY.md and medical-history.md. Prevents the common failure mode of recommending products based on brand reputation or search-result titles without actually reading the product page. Every recommendation requires fetching the real product page, extracting the full "Other Ingredients" list, checking each ingredient against the user's hard-no list, reading user reviews, and verifying brand reputation. Maintains persistent state (verified-products, brand-database, search-log) so repeat searches don't redo work. Use this skill WHENEVER the user says /product-search, asks "find me a source for X", "where do I buy X", "what's the best brand of X", "compare [product A] and [product B]", "which [product] should I get", "direct URLs for those products", or explicitly asks for a product recommendation, supplement source, or brand comparison. Do NOT use proactively — only when user explicitly requests product sourcing. Do NOT use for general mechanism or rationale questions.

Practitioner-grounded deep research that extends the standard /research engine with structural enforcement against shallow vendor-flavored synthesis. Use this skill whenever the user says "/research-deep", "deep research", "not just docs", "find practitioners", "what experts who shipped X learned", "deeper research", or asks any research question where prior research output was rejected as superficial. Also trigger automatically when the topic is a fast-moving technical domain (AI agents, LLM infrastructure, novel architectures) where documentation-deep answers reliably mislead and only practitioner experience reports carry signal. Trigger especially when the user is visibly frustrated with a previous shallow research output, when they explicitly ask for named experts/practitioners, when they say "I need to actually understand X", or when the user mentions wanting to build / ship / operate something rather than just compare options. The skill enforces tier-tagged claims, verbatim primary-source quotes, mandatory named-practitioner citations, three-pass adversarial iteration, banned confidence theater, and Norton-safe output. Includes a reference-doc sub-protocol for building structured foundational understanding split across multiple files per question cluster.