ogmios

Academic work with scientific rigor — covers literature search, source evaluation, and writing. Configurable language. [WHAT] End-to-end support: (1) systematic literature search and source evaluation, (2) thematic synthesis, (3) writing essays, papers, reports, dissertations, literature reviews, method discussions. IMRAD structure, Harvard/APA, peer review. [WHEN] Use when: academic, scientific, essay, report, dissertation, literature review, paper, thesis, doctoral, peer-reviewed, systematic review, find sources, zotero, Harvard reference, APA, IMRAD, source criticism, literature review, research synthesis. NOT for: opponent role (academic-opponent), interview coding (gioia-analysis). [LANGUAGE] Configurable. Matches input. [EXPERTISE] Academic writing, IMRAD, source criticism, systematic literature search, Harvard/APA citations, thematic synthesis, qualitative/quantitative methods.

Weekly dual-memory flow: review project auto-memory (inbox) and promote to global memory, keep as ephemeral, or delete. Addresses the dual-memory tension in a layered memory system. [VAD] Structured walkthrough of `~/.claude/projects/.../memory/`. Classifies each entry as promote/keep/delete based on type and stability. Writes promotions to global memory. Writes a review report. [WHEN] Use when: promote memory, dual-memory, weekly memory review, memory promotion, review project memory, "is this memory stable?", "should this go global?", inbox cleanup of memory. [LANGUAGE] Configurable. [EXPERTISE] Memory architecture, ephemeral-vs-persistent classification, three-criteria evaluation (recurrence, generality, stability).

Aphorism database for citable lines. Distinguishes between user aphorisms (own, go to TELOS WISDOM.md) and external aphorisms (from other authors, go to a separate aphorism database). Supports search, categorisation, citation format, and retrieval for chronicles/scripts. [WHAT] Quote database where each aphorism has: text, author, source, domain tag, captured-on date. User-own aphorisms go to `~/.claude/memory/USER/TELOS/WISDOM.md`, external ones to `~/.claude/memory/KNOWLEDGE/Sources/aphorisms.md`. [WHEN] Use when: aphorism, quote, citable, "save this line", quotable, wisdom-quote, "who said X", find a quote about Y, "add to the quote database". Auto-activates after writing/editing if a line stands out. [LANGUAGE] Configurable. [EXPERTISE] Aphorism curation, quote attribution, citation format, retrieval.

Audio pipeline: transcription (Whisper), TTS (ElevenLabs), audio extraction (ffmpeg), and YouTube transcript via MCP. [WHAT] Audio skill for transcribe, TTS, voice-over, podcast workflow. Whisper for transcription (default OpenAI Whisper, configurable per-language), ElevenLabs for TTS, ffmpeg for extraction, YouTube-transcript MCP for video. [WHEN] Use when: transcribe, audio, TTS, voice-over, podcast, YouTube transcript, generate speech, read aloud, dictation audio, mp3, wav, m4a. [LANGUAGE] Configurable. TTS voice configurable per project.

Umbrella skill for reading a text artefact (chronicle, memo, report, transcript) and extracting candidates to the Ogmios KNOWLEDGE area in one operation. Classifies into 7 entity types: People, Organizations, Places, Events, Concepts, Sources, Research. The user chooses per candidate: create KNOWLEDGE node / archive to vault / skip / save in _inbox. [WHAT] Complement to atomize-zettel (Cards/Zettelkasten in vault). Atomize works against vault Cards/, extract-entities works against the Ogmios KNOWLEDGE area. [WHEN] Use when: extract entities, identify people/organizations/places, who is mentioned in this text, mining from text to KNOWLEDGE. [LANGUAGE] Configurable. [EXPERTISE] Entity extraction, NER classification, KNOWLEDGE-area routing.

Fact-checking and quality review of texts. [WHAT] Verifies claims, checks sources, identifies style issues. Can be used standalone or piped after /chronicle. [WHEN] Use when: fact-check, verify, check, quality check. [EXPERTISE] Fact-checking, source verification, style review.

Image generation and image editing. Two engines: GPT-Image-1 (OpenAI) and Gemini Nano Banana Pro (Google). Seven prioritised workflows: blog cover, presentation illustration, concept sketch, geopolitical map, meme/satirical image, infographic prompt, image edit/inpainting. [WHAT] Image skill via `~/.claude/scripts/call-image-api`. Default: Nano Banana Pro for photorealism and maps, GPT-Image-1 for precision edits and text-in-image. Output saves to a configured pictures path. [WHEN] Use when: generate image, illustration, AI image, cover image, presentation image, infographic, image edit/inpainting.

Meta-skill for upgrading your Ogmios fork against future PAI versions (Daniel Miessler upstream). Structured process: clone PAI repo → diff against your config → classify ISCs (Isolated Strategic Changes) → build PRD → execute in phases → audit-preserve. [WHAT] Process skill that documents how to upgrade your Ogmios fork against upstream PAI releases without losing your customisations. [WHEN] Use when: PAI upgrade, miessler new release, upgrade fork from upstream, "new PAI version released". NOT for: feature requests on your own fork.

Structured incident investigation grounded in the Toyota Production System, Kaoru Ishikawa, James Reason's Swiss Cheese model, Dean Gano's Apollo method, and Google SRE blameless-postmortem culture. Five methods: 5-whys, fishbone, postmortem, fault tree, Kepner-Tregoe IS/IS-NOT. [WHAT] RCA for failure analysis: why did transformation projects fail, OSINT cause chains, policy evaluation. Core axiom: the proximate cause is where analysis begins, not ends. People are never root causes — if a human could make the error, the system permitted it. A cause is "root enough" when it's actionable. [WHEN] Use when: root cause, RCA, 5-whys, fishbone, postmortem, fault tree, "why did this happen", "why does X fail", IS/IS-NOT, defect investigation, pre-launch risk. NOT for: structural loops (use systems-thinking), axiom decomposition (use thinking-firstprinciples). [LANGUAGE] Configurable. Non-blameful tone (the SRE-style "blameless").

Deep analysis of unfamiliar situations, domains, or contexts via a structured three-phase protocol: Map, Drill, Act. [WHAT] Builds quick situational understanding in three interactive phases. The user signals unfamiliarity, the assistant asks precise questions (Map), drills deeper in the core domain (Drill), delivers actionable analysis (Act). Ends with a structured insight document saved to the vault's unsorted area. [WHEN] Use when: help me understand, I don't know enough about, what should I know about, I'm meeting X and know nothing, situation analysis, unknown domain, new situation, quick context before a meeting, prep before a decision. [LANGUAGE] Configurable. Output matches input language. [EXPERTISE] Domain orientation, structured questioning, three-phase analysis, decision prep.

Structural analysis of complex systems grounded in Donella Meadows, Peter Senge, Jay Forrester, Russell Ackoff, and the Santa Fe Institute tradition. Five workflows: Iceberg, CausalLoop, FindArchetype, FindLeverage, ConceptMap. [WHAT] Systems-thinking skill for resilience work, democracy, hybrid threats, policy work. Core axiom: "behaviour is generated by structure — events are visible, structure is not." At Extended+ effort on recurring behaviours or cross-system coupling, systems-thinking is the structural lens, not optional. [WHEN] Use when: systems thinking, causal loop, feedback loops, archetypes, leverage points, iceberg model, unintended consequences, fix the system, structural cause, "why does this keep happening?", recurring problems, systemic questions, second-order effects. NOT for: incident causal chains (use root-cause-analysis). [LANGUAGE] Configurable. Output structured with mermaid diagrams where relevant.

Methodological foundation for financial OSINT analysis. [WHAT] Follow-the-money, ownership-structure analysis, network analysis, sanctions checking. Frameworks for financial investigations with public-data tools. [WHEN] Use when: follow the money, ownership structure, beneficial owner, sanctions, oligarch, money laundering, shell company, offshore, UBO. [EXPERTISE] Corporate intelligence, sanctions checking, network analysis.

Network analysis of groups of people or organizations per OSINT methodology. Maps connections, hierarchy, common nodes, and influence channels. [WHAT] Maps relationships between persons/organizations. Identifies central nodes, bridge persons, clusters, board crossovers, collaboration patterns, financial dependencies. Produces a network map (mermaid graph or table) plus analytical commentary. [WHEN] Use when: network analysis, "map the network", "connections between", SNA, social network analysis, "who is connected to", group analysis, influence network, ideological cluster, network mapping, network OSINT. NOT for: single person (use person-osint), corporate structure (use financial-osint), location analysis (use geolocation). [LANGUAGE] Configurable. [EXPERTISE] SNA principles (centrality, bridge nodes, clusters), corporate-registry data, board crossovers.

Person OSINT — professional person investigation using systematic methodology. [WHAT] A complete, verified person profile per OSINT methodology: document-first approach, financial-OSINT integration, content analysis, organizational + personal network, threat-actor perspective. [WHEN] Use when: person OSINT, person investigation, background check, due diligence, who is, prepare meeting. [LANGUAGE] Configurable. [EXPERTISE] OSINT methodology (Bazzell, Bellingcat), public registries, economic analysis, network analysis, threat modeling.

Domain expertise and institutional understanding for intelligence analysis. [WHAT] Understands how institutions, systems, and domains actually work — not just on paper. Validates and grounds analysis. [WHEN] Use when: implications analysis, "what does this mean for us?", validation, decision processes, institutional context, system mapping. [EXPERTISE] Institutional knowledge, decision processes, system mapping, implications analysis. Core principle: reality is more complex than the org chart.

Methodology for broad horizon-scanning and signal detection. [WHAT] Defines WHAT to scan and HOW results are valued. Execution happens via the intelligence-researcher agent. [WHEN] Use when: horizon scanning, signal detection, weak signals, peripheral scanning, early warning, trend spotting. [EXPERTISE] Signal detection, peripheral scanning, weak signals, trend analysis. Workflow: SIGNAL = methodology | Agent = research | STRUCTURE/CONTEXT/SYNTHESIS = analysis

Rigorous analysis, hypothesis testing, and method criticism for intelligence work. [WHAT] Takes raw material, builds testable hypotheses, assesses uncertainty, systematically challenges assumptions. Built-in devil's advocate. [WHEN] Use when: hypothesis testing, source criticism, ACH, premortem, red team, structured analysis, evidence valuation, alternative hypotheses. [EXPERTISE] Hypothesis testing, source criticism, uncertainty assessment, cognitive bias detection. Core principle: hypotheses are tested, not confirmed.

Integration, narrative, and final production for intelligence analysis. [WHAT] Weaves together SIGNAL, STRUCTURE, and CONTEXT into coherent, decision-relevant products. Creates briefs, situation reports, scenarios. [WHEN] Use when: analysis brief, situation report, scenario, strategic summary, final product, communicate to decision-makers. [EXPERTISE] Narrative integration, synthesis, decision-relevant communication, scenario writing. Core principle: complexity should become graspable without being oversimplified.

Smart routing from the Ogmios KNOWLEDGE area to a configured Obsidian vault. Reads a KNOWLEDGE node (or direct input), analyses entity type, picks the right vault template + target folder, creates a file with correct frontmatter, updates the KNOWLEDGE node's `vault_link`. [WHAT] Routing skill from Ogmios → vault. Seven entity types map to vault MOCs. No vault structure changes — uses the user's existing templates and MOCs. [WHEN] Use when: "save it to the vault", "archive this", "move to the vault", "save in the vault". [LANGUAGE] Configurable. Strict vault convention.

After finished text work, identify concepts, people, places, and ideas worth atomic notes (Cards), link in existing cards, and suggest updates to cards where the new material casts new light on already-established ideas. [WHAT] Scan recently produced/edited text against `Cards/_index.md`. Return three lists: NEW CANDIDATES (concepts worth their own card), EXISTING TO LINK IN (cards that exist but aren't linked), EXISTING TO UPDATE (cards where the text adds something substantive). The user chooses per row. Closes "the loop" between writing and the knowledge graph. [WHEN] Use when: atomize, "suggest atomic notes", "link in cards", "update cards", after a finished chronicle/memo/report, after the ALGORITHM VERIFY phase on text-producing work. Use PROACTIVELY as a silent last step in the Algorithm's VERIFY phase when produced text is >800 words. [LANGUAGE] Configurable. Output: short, scannable list. No AI filler patterns. [EXPERTISE] Atomic notes, Zettelkasten weave, concept extraction, Cards index lookup,

Smart note creation in an Obsidian vault. Routes intent to the correct template for common note types: Person, Meeting, Concept, Research, Project, Article, Daily, Recipe, MOC, Zotero, generic Note. [WHAT] Create or update notes in your Obsidian vault. Routes intent to the correct template. Handles YAML frontmatter, MOC classification, wikilinks, metadata updates. [WHEN] Use when: note, person card, meeting note, concept note, daily note, new project, save article, recipe, create note, update metadata, fix metadata (single file). [LANGUAGE] Configurable. [EXPERTISE] Obsidian templates, YAML frontmatter, MOC classification, wikilinks.

Vault-wide operations on an Obsidian vault. Health diagnostics, bulk metadata standardization, file organization, plugin guidance. [WHAT] Vault-wide operations: health diagnostics, orphan/dead-end resolution, bulk metadata standardization, tag migration, file sorting/moving, unresolved link fixing. Dataview query help. Plugin configuration guidance. [WHEN] Use when: vault health, vault maintenance, bulk metadata, fix all files, sort unsorted, orphans, dead ends, unresolved links, move files, dataview query, plugin help, cleanup, organize vault. [LANGUAGE] Configurable. [EXPERTISE] Obsidian CLI, vault graph analysis, bulk operations, Dataview, plugins.

Email skill for writing, editing, and replying to email with the right register, length, and tone. Four registers: collegial (close colleagues), institutional (agency/board), external (media/researchers), private (friends/family). Auto-detects recipient and context. [WHAT] Voice-aware email skill: right opening/closing per relationship, right length (short by default, long when necessary), right language (matches recipient), avoid AI patterns, strip em-dashes, preserve the user's voice in email. [WHEN] Use when: write email, draft email, reply, email to, "write to X", "polish this email", "why is this so long". NOT for: memo (use policy-brief), op-ed (use op-ed). [LANGUAGE] Configurable; matches recipient.

Plan and structure lectures, keynotes, workshops, and panel appearances. Output: slide outline + speaker notes, ready to export to pptx or html-slides. [WHAT] Three-act structure, slide outline (one idea per slide), audience adaptation, time management, story integration. Markdown outline + speaker notes, exportable to pptx or html-slides (reveal.js, marp). [WHEN] Use when: lecture, presentation, keynote, workshop, panel, seminar, slides, slide deck, talk structure, conference. NOT for: cue cards (use talking-points), pptx file (use /pptx). [LANGUAGE] Configurable. Matches input. [EXPERTISE] Presentation design, storytelling, academic lectures, workshops, pptx/html-slides bridge. [DISTINCTION] talking-points = WHAT to say orally | presentation = slide flow | pptx = the file itself

Create social media posts for BlueSky, Mastodon, and LinkedIn in the user's voice. Platform-adapted short format (300-3000 characters). [WHAT] PLATFORM-ADAPTED SHORT FORMAT — for each platform's conventions. Geopolitical commentary, threads (BlueSky/Mastodon), professional tone (LinkedIn), archive.org links. The user's voice at short range. [WHEN] Use when: social media, BlueSky, Mastodon, LinkedIn, post, thread, "BlueSky post", "LinkedIn post", "thread about", quick comment, "share on". NOT for: chronicle (use chronicle), op-ed (use op-ed), read-aloud script (use teleprompter). [LANGUAGE] Configurable. Matches platform and audience. [EXPERTISE] Short-form analysis, social media conventions, archive linking, thread structure. [DISTINCTION] chronicle = long form (8000+ chars) for blog | social-post = short form (300-3000 chars) for specific platforms

Create talking points and cue cards for radio, TV interviews, panel debates, and oral appearances. Format: core message first, one-liners, bridging for hard questions. [WHAT] ORAL DELIVERY — optimised to be SAID. Core messages first, one-liners with depth, bridging for hard questions, counter-argument handling, imagery. Cue cards, not finished text. [WHEN] Use when: talking points, radio interview, TV interview, media interview, panel debate, speech, dinner speech, opening speech, "prepare interview", "before TV", "before radio", "cue cards", "interview prep". NOT for: slides (use presentation), read-aloud script (use teleprompter), written debate (use op-ed). [LANGUAGE] Configurable. [EXPERTISE] Oral communication, broadcast media, rhetoric, bridging, message discipline. [DISTINCTION] chronicle = written chronicle | op-ed = written argumentation | presentation = slide flow | talking-points = CUE CARDS for ORAL interview/panel

Create read-aloud scripts for teleprompter. Plain prose without headings or bullet lists, calibrated for reading time at a natural reading speed. [WHAT] READ-ALOUD SCRIPT — text optimised to be spoken. Format: continuous prose, natural breathing pauses, short to medium sentences, no headings, no bullets. Length calibrated to target time at ~150 words/min as default. [WHEN] Use when: teleprompter script, voiceover script, video script, podcast script, "script for X minutes", read-aloud manuscript, YouTube script. NOT for: cue cards (use talking-points), slides (use presentation), op-ed (use op-ed). [LANGUAGE] Configurable. Output: plain prose ready to read directly. [EXPERTISE] Teleprompter writing, oral rhythm (~150 wpm), natural breathing pauses, scene-building without visual structure, word-counting for length calibration.

Create a NEW chronicle, blog post, or short opinion piece in the user's personal voice. Format: ≤2 pages, narrative structure. [WHAT] Low-affect expertise, narrative force, varied rhetoric, coherent arc. Anti-AI writing: human text, never generic. Creates text from scratch in chronicle format (opening, build, landing). [WHEN] Use when: chronicle, blog post, short opinion piece, short reflection, "write a chronicle about", "blog about". NOT for: op-ed/debate article (use op-ed), polishing existing text (use voice-editor), social-media post (use social-post). [LANGUAGE] Configurable. Reads voice profile from memory/USER/VOICE.md or skill argument. [EXPERTISE] Chronicle writing, voice fidelity, prose, anti-AI principles, narrative structure ≤2 pages.

Write op-eds and opinion articles for major publications. Thesis-driven argumentation, not narrative. [WHAT] Argumentative structure: thesis first, evidence, solution, call to action. Publication adaptation per length and tone. Co-author handling. Reply and final reply formats. Sharp policy stance. [WHEN] Use when: op-ed, debate article, opinion piece, "argue for", "write an op-ed", reply, final reply, letter to the editor. NOT for: chronicle/blog (use chronicle — narrative, not thesis-driven), policy memo (use policy-brief). [LANGUAGE] Configurable. Sharp and thesis-driven, not the user's narrative chronicle voice. [EXPERTISE] Op-ed conventions, thesis formulation, publication adaptation, evidence handling, co-author handling, reply/final-reply.

Create memos, policy briefs, and decision support for government agencies and decision-makers. [WHAT] Administrative language, evidence-based argumentation, structured sections, reference handling, recommendation formulation. Multilingual. [WHEN] Use when: memo, policy brief, decision support, response to consultation, internal note, policy analysis, white paper, briefing. [LANGUAGE] English (default) or other — match input or ask if unclear. [EXPERTISE] Policy writing, government documents, evidence synthesis.

Create or upgrade Claude Code skills per Anthropic's latest spec and Ogmios PAI conventions. A living tool that tracks the Anthropic pattern's evolution. [WHAT] Two modes: (1) CREATE a new skill with SKILL.md, frontmatter, [WHAT]/[WHEN]/[EXPERTISE] format, NOT-distinctions, progressive disclosure. (2) UPGRADE an existing skill — diagnose against the spec, generate a diff, apply. [WHEN] Use when: create skill, new skill, upgrade skill, refactor skill, anthropic update, skill audit. NOT for: social posts (use social-post), documents (use document-factory). [LANGUAGE] English for meta-language. [EXPERTISE] Anthropic Agent Skills spec, Ogmios PAI conventions, YAML frontmatter, progressive disclosure, semantic-triggers/SKILLS-INDEX sync, validation.

Creative ideation skill for breaking established thinking patterns and generating unexpected connections. Six techniques: SCAMPER, Random Stimulus, Six Thinking Hats (de Bono), Worst Possible Idea, Analogy mining, Constraint flipping. For when the user needs to escape a rut. [WHAT] Port of the Miessler BeCreative pattern. Six distinct creative techniques that force a perspective shift. Output is not a finished idea but RAW MATERIAL for further work. The skill is generative, not convergent. [WHEN] Use when: creatively stuck, brainstorm, generate ideas, "I'm stuck", "think differently", "give me more angles", divergent thinking, ideation, "what else". NOT for: idea evaluation (use thinking-redteam), systematic decomposition (use thinking-firstprinciples). [LANGUAGE] English and other languages. Output follows the source language.

Quickly create CLI tools: bash/TypeScript/Python scripts in ~/.claude/scripts/ with the right shebang, exec bit, argparse pattern, --help, and dry-run default for destructive ops. Convention: bun for TS, uv for Python. [WHAT] Build skill for the personal script surface. Output: a working CLI with consistent --help, --dry-run, --verbose flags. Logging to ~/.claude/logs/[scriptname].jsonl when relevant. [WHEN] Use when: create CLI, build script, "I need a tool that...", "a command to...", command-line tool, short utility. NOT for: larger app (use engineering), one-off-bash (write inline). [LANGUAGE] Code is always English. Help text and comments may be in other languages if preferred.

Evaluation skill for AI output: apply measurable criteria to text/code/ research/policy. Supports four eval types: rubric-based (1-5 scale), pairwise (A vs B), checklist (binary), and red-team pass (uses thinking-redteam). Output: score + reasoning + improvement suggestions. [WHAT] Structured way to answer "is this good enough?". Uses domain rubrics: chronicle quality, memo rigor, academic precision, OSINT credibility. Eval is a feedback mechanism, not a gate. [WHEN] Use when: evaluate, score, judge, "is this good enough?", "rate this", quality check, compare A vs B. NOT for: fact-checking (use fact-check), academic peer review (use academic-opponent). [LANGUAGE] English and other languages; matches input.

Content-adaptive wisdom extraction from text/podcast/video/article/paper. Detects wisdom domains in the content and builds custom sections rather than static headers. 5 depth levels from Instant (1 section) to Comprehensive (10-15 sections + themes). [WHAT] Port of the Miessler ExtractWisdom pattern. Spicy/contrarian takes are required. Voice follows the user's conversational style — bullets should read like "tell me what you just saw", not a press release. [WHEN] Use when: extract wisdom, analyse podcast/video/article, key takeaways, what's interesting here, distill, summarise interview, "extract wisdom from". NOT for: knowledge archiving (use archive-to-vault), atomic notes (use atomize-zettel). [LANGUAGE] Source-matching by default; English if explicitly requested.

Apify platform for platform-specific scraping beyond WebFetch/BrightData. Support for 5 priority platforms: LinkedIn (beyond MCP), Instagram, YouTube (comments + meta), Google Maps, Facebook (public). Uses APIFY_TOKEN from `~/.claude/.env`. [WHAT] Apify skill for OSINT and intelligence monitoring. Complements the scraping skill for platforms where tiers 1-4 are insufficient. Cost awareness: each run costs credits — start narrow. [WHEN] Use when: apify, deep LinkedIn scraping, Instagram scraping, YouTube comments, Google Maps reviews, Facebook public page, "platform-specific scraping", "deep-scrape Instagram profile". NOT for: regular site (use scraping tier 1-3), single LinkedIn post (use linkedin MCP). [LANGUAGE] English and other languages.

Browser automation via Playwright MCP for tasks requiring an interactive web session: login, click flows, screenshots, form-fill, multi-step scraping. Complement to the scraping skill (tier 4). [WHAT] Playwright skill for: fetching content behind login, filling forms (government services, subscriptions), screenshots for evidence preservation (OSINT, source verification), interactive scraping of heavy SPAs. Headless by default; headed mode when needed. [WHEN] Use when: browser automation, playwright, click on, fill form, screenshot of site, login-required scrape, "log in to X and fetch", multi-step flow, "fetch behind paywall via subscription". NOT for: simple GET requests (use scraping tier 1-3), background jobs (Playwright is synchronous). [LANGUAGE] English and other languages.

Web scraping skill with a 4-tier strategy: WebFetch (fast, static), searxng (search-driven), BrightData scrape_as_markdown (anti-bot, JS-rendered), Playwright (interactive, login-required). Picks the right tool per target URL. [WHAT] Scraping for OSINT, intelligence monitoring, source verification. Default: start cheapest (WebFetch), escalate as needed. BrightData for Cloudflare / anti-bot-protected sites. Playwright for login or heavy SPAs. [WHEN] Use when: scrape, fetch URL, crawl, extract content, read site, capture article, mirror page, extract markdown, anti-bot. NOT for: web search (use searxng or exa), archiving (use Wayback). [LANGUAGE] English and other languages. Output content matches source language.

Structured deliberation with 4 advisors illuminating a question from different perspectives. Inspired by Daniel Miessler's Council pattern. Visible transcript of the debate. [WHAT] Convene a council of 4 distinct perspectives (expert roles fitted to the question) that debate, challenge, and build on each other's arguments. Produces a visible transcript with conclusion and action recommendation. [WHEN] Use when: council, deliberate, multiple perspectives, weigh arguments, structured debate, "I need to think this through" [EXPERTISE] Structured deliberation, multi-perspective analysis, dialectical reasoning

Decomposition from fundamental principles. Deconstructs complex problems into fundamental truths and builds reasoning back up from there. [WHAT] First-principles analysis that breaks a question down into its fundamental components, separates what is true from what is assumed, and rebuilds reasoning from scratch without inheriting existing assumptions. [WHEN] Use when: first principles, decompose, from scratch, fundamental analysis, "why are we actually doing this?" [EXPERTISE] Fundamental decomposition, assumption challenging, Socratic method