skill-engineer

Removes AI writing patterns and injects human quality (specificity, burstiness, precise emotional language) into drafted content. Includes an optional personal-brand mode (voice reference + deterministic eval gate) you can adapt to your own voice. Use when editing drafts so they don't sound like AI. NOT for generating new content, fact-checking, SEO scoring, or research.

Review-only quality grader for reader-facing analytical deliverables (HTML reports, executive briefs, strategy decks, decision documents). Modes: spec (returns authoring template), spec-judge (grades the storyline before writing), review pass 1 (argument structure), review pass 2 (readability), review pass 3 (humanization — verifies the caller's humanize edit landed: zero em-dashes, no AI tells, human texture); each pass gates the next. Returns structured violation reports; the CALLING skill applies fixes. Rule of thumb: if a human will read it, it goes through this review; machine artifacts, agent-read repo docs, and working drafts do not. Invoke via 'run write-report on X', 'get an authoring spec', or 'judge this storyline'. Do NOT use for SEO review (use seo-review-loop), code review, fact-checking, or short chat replies. Full mode/strictness/doc-type details in the skill body.

Renders complex information as a single self-contained HTML file — dashboards, analytical reports, comparison tables, timelines, flow/sequence diagrams, and data explainers — using Antony Evans's brand tokens (forest green / off-white / amber, Space Grotesk + Inter). Triggers on "build/render/make this as HTML", "create a dashboard", "show me a visual report", "turn this into a one-pager", "render with diagrams", or when another skill says "use html-output for the deliverable". Also activates when the artifact has 3+ dimensions of data, needs diagrams or interactive export buttons, or would be clearly more legible in a browser than in markdown. Does NOT trigger for short chat replies, markdown notes for the Obsidian vault, code files, slide decks (use pptx), spreadsheets (use xlsx), PDFs (use pdf), logos or images (use generate-image), or plain text answers. Output is one styled .html file with optional Mermaid or Chart.js via CDN when justified.

Review-only multi-mode quality grader for reader-facing analytical deliverables (typically HTML reports with prose + charts/diagrams inline). Supports four modes: (1) `spec` returns a doc-type-specific authoring template the caller fills in; (2) `spec-judge` grades a filled-in authoring spec (storyline) BEFORE the caller writes the document — catches broken governing observation, MECE-overlap, Rumelt goal-not-strategy at the storyline stage and saves 60-70% of the iteration cost of post-hoc structure fixes; (3) `review pass: 1` audits argument STRUCTURE (Minto Pyramid, MECE, SCQA, Rumelt kernel, dot-dash storyline, issue-tree integrity) of the finished document; (4) `review pass: 2` audits READABILITY (action titles, so-what, specificity, code/jargon discipline, active voice, Frankenstein detection). Structure pass gates the readability pass. **Reviews reader-facing artifacts — HTML deliverables, executive reports, gate decisions — not machine artifacts or working markdown drafts.** The rule of thumb: if a hu

Produces a ranked shortlist of venture concepts for disrupting an industry via a 6-phase workflow (Industry Deconstruction → Value Chain Pain Audit → Enabling Conditions Scan + capability seeds → Framework Application → Idea Generation via eight structural moves → Stress Test). Runs two generation lanes: an incumbent-anchored lane (Blue Ocean ERRC, Aggregation, Decoupling, Counter-positioning) and a first-principles lane (Phase 3 capability seeds + Move 8 capability-first + Thiel's Secrets reframed as testable bets, never truth-gated). The incumbent:capability mix is allocated per-industry from Phase 1-3 signals and ratified by the human at Gate 2 (not hardcoded), with a ≥1-per-lane floor. Every concept ships as a bet (load-bearing hypothesis + cheapest validation test); the human gate is test-worthiness, not conviction. Built against a disruption-dataset.yaml from inherited analyze-industry outputs or fresh librarian research. Three approval gates. Reimagination-specific bar test (≥3 non-obvious concepts, ze

Structural demand-side analysis for a defined industry: JTBD (Christensen/Ulwick — functional + emotional + social), JTBD-based segmentation (NOT demographic), substitution risk per segment (named cross-category candidates + switching cost + likelihood), WTP drivers per segment, and named leading demand signals. Output: demand.md — surfaces disruption threats before they show in share data. Sub-skill of analyze-industry, also invocable standalone. CONDITIONAL: orchestrator SKIPS when ALL three hold — market mature (>20yr) AND demand volatility <±2% AND no substitution threat flagged in Five Forces; standalone caller decides. Triggers: 'analyze demand for [industry]', 'JTBD for [industry]', 'customer segmentation for [industry]', 'substitution risk for [industry]', 'WTP drivers in [industry]'. Do NOT use for product-level JTBD, single-company positioning (use build-company-model), market sizing (use size-market), or supply-side competition (use map-competitive-arena).

Produces a senior-analyst industry-structure brief by orchestrating MBB sub-skills (BCG Strategy Palette, McKinsey G3 sizing, Porter Five Forces with complementors and AI-as-force, Bain profit pools on Porter value chain; Phase 2 adds arenas, S-curve, Helmer 7 Powers, JTBD). Three-layer output to 08-knowledge/world-model/industries/[slug]/. Quick or Deep mode. Three approval gates. Fresh-context bar test. V/C/A/I provenance. Pyramid + SCQA synthesis shipped as a quality-reviewed HTML report (rendered via html-output, audited by analysis-quality-review). Concludes with where-to-play/how-to-win. Triggers on 'analyze the [industry] industry', 'industry brief for [industry]', 'is [industry] structurally attractive', 'industry analysis for [PE deal/market entry]', 'industry-structure analysis'. Do NOT activate for single-company analysis (use build-company-model), DTC category go/no-go (use assess-category), decision pressure-testing (use stress-test), or buyer/investor universe (use building-buyer-shortlists).

Produces a forward-looking industry trajectory analysis: dual S-curve (market-adoption + technology lifecycle with diagnostic signals), Three Horizons overlay on G3 sub-segments (consumes size-market when present), discontinuities catalog (regulatory/tech/behavioral with year-range timing windows), Helmer Power Progression (which 7 Powers are buildable now / Year 3 / Year 5 / closed), and base/bear/bull scenarios with ≥3 named swing variables for the 5-year horizon. Output: trajectory.md with V/C/A/I tags. Sub-skill of analyze-industry but invocable standalone. Triggers on 'industry trajectory for [industry]', 'where is [industry] on the S-curve', 'three horizons for [industry]', 'scenario analysis for [industry]', 'discontinuities facing [industry]'. Do NOT activate for company-specific moat assessment (use assess-moat-sources), market sizing (use size-market), or competitive arena mapping (use map-competitive-arena).

Assesses which of Helmer's 7 Powers (scale economies, network economies, counter-positioning, switching costs, branding, cornered resource, process power) protect incumbents in a defined industry, how durable each is over the analysis horizon, and which Power combination the winning archetype must hold. Output: moat-sources.md with per-Power present/absent + intensity, benefit + barrier decomposition for each present Power, durability rating (erosion timeframe + vector), buildability for a new entrant or roll-up, winner-archetype Power profile, durability risks, and reconciliation against value-chain-profit-pools.md if present. Sub-skill of analyze-industry; invocable standalone. Triggers on 'assess moat for [industry]', '7 powers analysis of [industry]', 'how durable is the moat in [industry]', 'moat durability [industry]', 'will the moat hold'. Do NOT use for single-company moat (use build-company-model), customer-lock-in only (use analyze-demand), or share analysis (use map-competitive-arena).

Diagnoses competitive environment for an industry using BCG's Strategy Palette (Reeves) — Classical / Adaptive / Visionary / Shaping / Renewal — via the three-question discipline (predictability, malleability, harshness). Output: strategic-environment.md classifying the environment, evidencing the 3 dimensions with V/C/A/I tags, flagging ambidexterity, emitting a sub-skill routing matrix that weights Phase 1+2 sub-skills heavy/light/skip. Sub-skill of analyze-industry, invocable standalone. Runs FIRST in orchestrator chain — wrong diagnosis = wrong toolkit. Triggers on 'diagnose strategic environment for [industry]', 'strategy palette for [industry]', 'is [industry] classical or adaptive', 'classify the [industry] environment', 'Reeves environment for [industry]'. Do NOT activate for industry-attractiveness (use map-five-forces), market sizing (use size-market), single-company moat work (use assess-moat-sources), or DTC category go/no-go (use assess-category).

Produces a Porter strategic-group map overlaid with McKinsey Arenas framing for an industry. Identifies ≥3 strategic groups via 2-axis visualization, profiles each (size, members, basis of competition, profitability), names a winner archetype (consolidator / innovator-specialist / platform-orchestrator / vertically-integrated / asset-light / none-emerging), and assesses mobility barriers between adjacent groups. Output: competitive-arena.md with map, profiles, archetypes, mobility-barrier matrix, arenas overlay, V/C/A/I tags, next_skills YAML. Sub-skill of analyze-industry; standalone-invocable. Triggers on 'strategic group map for [industry]', 'competitive groups in [industry]', 'who competes with whom in [industry]', 'winner archetypes in [industry]', 'competitive landscape for [industry]'. Do NOT activate for industry forces (use map-five-forces), market sizing or whole-market arenas (use size-market), firm-specific moat (use assess-moat-sources), demand analysis (use analyze-demand).

Produces a Porter Five Forces analysis extended with complementors (sixth force, Brandenburger/Nalebuff) and AI-as-named-force (cost-structure impact + new-entry vector + data-intermediary position) for a defined industry. Output: five-forces.md naming THE governing force that determines this industry's profit distribution in one causal sentence, plus per-force intensity assessment with V/C/A/I-tagged evidence and explicit profit-pool cross-reference. Sub-skill of analyze-industry but invocable standalone. Triggers on 'five forces analysis for [industry]', 'Porter analysis for [industry]', 'industry-structure analysis for [industry]', 'is [industry] structurally attractive'. Do NOT activate for single-company competitive positioning (use map-competitive-arena or build-company-model), market sizing (use size-market), or moat assessment of a specific company (use assess-moat-sources).

Produces a paired Porter value-chain decomposition + Bain/Gadiesh-Gilbert profit-pool overlay for a defined industry. Output: value-chain-profit-pools.md with value-chain stages (inbound logistics through service + support activities), absolute EBIT/economic profit (not revenue margin) per stage, horizontal-bar profit-pool visualization in markdown, identification of which stage has structural profit concentration and why, plus V/C/A/I provenance tagging. The pairing is deliberate — value chain alone is incomplete; profit pool alone lacks structural context. Sub-skill of analyze-industry but invocable standalone. Triggers on 'value chain analysis for [industry]', 'profit pool analysis for [industry]', 'where does the money sit in [industry]', 'where do profits concentrate in [industry]', 'value chain and profit pools'. Do NOT activate for single-company P&L decomposition (use build-company-model), customer-level economics (use analyze-demand), or competitive share analysis (use map-competitive-arena).

Produces a granular market sizing for a defined industry/sub-segment using McKinsey G3 granular-growth decomposition + arenas qualification screen, with top-down and bottom-up triangulation. Output: market-sizing.md with TAM/SAM/SOM, ≥3 sub-segment growth-rate decomposition, explicit de-averaging statement, and V/C/A/I provenance tags on every numeric claim. Sized to PE-CDD speed. Triggers on 'size the [industry] market', 'TAM SAM SOM for [industry]', 'how big is the [industry] market', 'market sizing for [industry]', 'granular market sizing'. Sub-skill of analyze-industry but invocable standalone. Do NOT activate for single-company revenue forecasting (use build-company-model), customer-segment demand analysis (use analyze-demand), or competitive share analysis (use map-competitive-arena).

Creates robust long-form task prompts and agent briefs using production prompt engineering patterns (layered XML architecture, trust boundaries, anti-rationalization rules, numeric anchors). Invoke when the user needs to create an agent system prompt, write a complex task prompt, design a multi-agent workflow prompt, or produce prompts that are clearer, more reliable, and easier to debug. Does NOT activate for: short single-turn prompts, simple rewrites, asking Claude to explain prompt engineering concepts, or editing existing prose that is not a prompt.

Debriefs a developer after an AI-assisted coding session by inspecting git diffs or commits, explaining the actual architecture decisions, design patterns, tradeoffs, caveats, and learning concepts with file/function references, then generating a practical quiz and follow-up study notes. Use when the user says they shipped code with AI, vibe coded, wants to learn what changed, wants a post-session code debrief, or asks to review a diff/commit for learning. Do NOT use for generic code review, bug fixing, refactoring, or architecture planning when there is no concrete diff or commit range to inspect.

Generates a structured morning brief and writes it to the daily review file. Pulls from: current priorities (goals summary), active task board, latest competitor digest, and content pipeline status. Fires automatically on weekdays. Output is the morning intention section of your daily review file.

Guides a founder interactively from "no idea" through idea generation, rapid validation, and commitment to a business — synthesizing PG, YC, Lean Startup, Mom Test, JTBD, and gstack office-hours methodology updated for the AI era. Produces founder profiles, scored idea candidates, validation sprint results, conviction scorecards, and 30-day action plans. Use when the user wants to find a business to start, explore startup ideas, figure out what to build, get unstuck on what to work on, or brainstorm business directions. Do NOT use when the user already has a specific idea to evaluate (use assess-category or stress-test instead), wants a business plan written, or needs a pitch deck reviewed.

Creates MARP presentation decks (.md files rendered to PDF/HTML/PPTX via marp CLI) with custom CSS themes, SVG inline charts, dashboard components, and speaker notes. Use when asked to create slides, build a deck, make a presentation, generate MARP output, design a theme, edit existing slides, or convert content into a slide format. Also handles read/summarize requests on existing MARP files. Do NOT use for standalone image generation, static infographics, or PowerPoint requests with no MARP involvement.

Assesses US regulatory compliance likelihood for any consumer product before listing for US sale. Accepts a CJ product PID/URL, Amazon ASIN, or product name/description as input. Classifies the product into a regulatory category (children's toys, electronics, cosmetics, supplements, pet products, apparel, kitchen/food-contact, or general consumer goods), determines which US standards apply, gathers market evidence, and produces a structured verdict (🟢/🟡/🟠/🔴). Saves report to a compliance-reports directory. Do NOT use for non-US markets or to produce legal opinions.

Use when the user provides a DTC or ecommerce store URL and asks for a teardown, breakdown, brand analysis, competitor teardown, investor memo, store audit, deep dive, or 'what's going on with [brand]'. Produces an investor-grade markdown teardown report covering brand, market, unit economics, supply chain, channel mix, marketing, reviews, agentic-commerce readiness, risks, and a falsifiable verdict. Triggers: 'dtc teardown', 'brand teardown', 'store teardown', 'competitor teardown', 'analyze this store', 'investor memo on [brand]', 'break down [store url]'. Do NOT use for SEO-only audits, design-system extraction, lead-gen scraping, or general web scraping with no brand/investor focus.

Audits and evolves brand positioning, voice consistency, and messaging for an existing brand. Works across personal brands and DTC/ecommerce store brands. Runs a structured audit across positioning, voice, visual direction, and content pillars, then produces a scorecard with specific fix recommendations. Also develops reusable messaging frameworks (positioning statements, value propositions, pillar definitions). Use for brand health checks, positioning refreshes, and messaging framework development. Do NOT invoke for building a brand from scratch (use /build-brand), content creation, or product copy.

Develops a full go-to-market marketing plan for a new online business, product, or idea — covering beachhead market, ICP, competitive landscape, Dunford positioning with messaging, channel strategy with buyer journey mapping, budget allocation, pricing, content strategy with retention, 90-day launch timeline, PMF definition, first customer motion, and AARRR + GEV KPI framework with measurement setup. Produces a structured markdown plan saved locally. Activate when: "build a marketing plan", "create a GTM strategy", "how should I market my new product/brand/startup", "plan the launch for", "what channels should I use to launch", "write a marketing plan for". Do NOT activate for: auditing existing marketing performance, writing individual content pieces, running specific campaigns, or competitive analysis without a launch goal.

Autonomously optimizes any Claude skill (SKILL.md) or CLAUDE.md file through a closed hypothesis→test→evaluate→keep/discard loop, producing an updated skill file and an iteration dashboard (baseline score, each hypothesis, keep/discard decision, improvement delta). Use when a user wants to improve a skill's output quality against measurable criteria, run autonomous evals on a skill, or set up a self-improving optimization loop for a content, writing, or routing skill. Do NOT use for building new skills from scratch, one-time debugging sessions, or tasks where evaluation criteria cannot be expressed as binary true/false conditions — use skill-engineer-master instead.

Three-phase strategic decision analysis combining verbalized sampling (Stanford arXiv 2510.01171), customizable analytical lenses, and a structured decision brief. Surfaces tail-distribution insights — the non-obvious, suppressed analyses that mode-collapsed prompting misses. Adapted from @alex_prompter's Model Council article (April 2026) for single-Claude use. Works on any high-stakes decision: founder pivots, DTC category selection, brand positioning, workstream prioritisation, pricing, hiring.

Activate when implementing or modifying API endpoints, authentication or authorization logic, database queries or ORM calls, user input handling, session/token management, file uploads, webhooks, or any feature that stores or transmits user data — gates implementation with threat surface analysis, secure defaults verification, attacker's eye pass, and structural control check before any code ships; do NOT activate for read-only documentation, configuration review, or infrastructure changes unrelated to user data flow

Diagnoses bugs from raw error logs, stack traces, or CI failure output — triages and clusters errors, localizes root cause (file → function → line), enters plan mode with a structured fix + test proposal for human approval, writes a typed pytest regression test using mocks first, then implements a targeted code fix and verifies the test turns green. Trigger when: user pastes logs and wants the underlying bug fixed with tests that prevent recurrence; user says "here are the logs, fix it"; CI output shows failures needing code changes and test coverage; stack traces are provided with a request to fix and test. Do NOT trigger for: general code review without logs, refactoring without a specific bug, performance work, test coverage requests with no associated failure.

Parses raw customer feedback (app reviews, support tickets, interview transcripts) into categorized severity-ranked themes, scans the codebase to identify affected files and components, and proposes 3 structural edit options per theme (minimal/refactor/architectural) with rationale tied directly to the original feedback. Trigger when: user provides raw customer feedback and wants to know what to change in the code; or an agent holds user complaints needing file-level engineering proposals. Do NOT trigger for: internal PR/code review comments (use receiving-code-review instead), stack-trace bug reports (use systematic-debugging instead), or PRD drafting.