Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, passive voice, negative parallelisms, and filler phrases.
Adversarially red-team OUR OWN claim/finding/draft BEFORE we publish or build on it — not the citations (that's verify-claims), but the ARGUMENT and framing. Use before shipping any Crucible verdict, flagship claim, hypothesis, or post, and whenever the owner says "red-team this", "stress-test the claim", "is this groundbreaking or textbook?", "would this survive a sharp critic?", or "are we overclaiming?". Runs a skeptic-led multi-lens panel (prior-art, strongest counter, weakest link, the missing 6th lens, textbook/overclaim check) and returns a verdict (PUBLISH / REFRAME / KILL) with the specific fixes. This is the systematized version of the audit that killed 5 of 6 candidates in the breakthrough scan and caught the nudging confound — it is meant to find reasons we are WRONG, before someone else does.
Use when someone asks to run Storm Research, use the storm-research skill, run the STORM method on a topic, says "storm research this" / "storm report on X" / "give me a STORM briefing on X", or wants a multi-perspective, citation-verified HTML research briefing on a topic. Runs a 4-phase pipeline: five expert lenses (Practitioner, Academic, Skeptic, Economist, Historian) -> contradiction map -> synthesized HTML report -> adversarial peer review + primary-source verification. Best for topics where multiple viewpoints and fact-checked claims matter; overkill for a simple factual lookup.
Adversarially verify every factual claim and cited number in a piece of writing against its PRIMARY source before it goes public. Use BEFORE publishing any Crucible entry, post (render_post / render_crucible), or gated outreach reply/press — anything with a number, a named study/paper, a statistic, a quote, a dollar figure, or a date. Also use when the owner says "verify this", "check the claims", "is this 100%", "fact-check before we post", or before any `approve` of an outreach/press action. Returns a verification banner (N checked / X false / Y corrected / Z demoted), per-claim verdicts, and a claim-safety guide (assert / caveat / avoid), then applies the corrections.
SEO + AI-search (AEO/GEO/LLMO) optimizer for the Agora storefront (dancenitra.github.io/agora) — a bilingual EN/SK static research blog on GitHub Pages. Synthesized from 37 SEO YouTube transcripts (vault `04 Resources/raw/YouTube Transcripts - SEO Ranking/`) into an actionable pre-publish checklist + site-wide technical audit, tailored to OUR stack and rules (static HTML via tools/render_post.py, owner anonymity, verified-numbers/cite-prior-art discipline, owner-gated outreach). Use when: publishing/rendering a post, the owner says "seo"/"rankuj"/"optimize for ranking"/"SEO friendly"/"rank #1", auditing the site for SEO, or before any `render_post.py --piece`. Two modes: (A) per-post optimization, (B) site-wide technical audit. Goal: rank #1 on Google AND be the cited source in ChatGPT/Perplexity/AI-Overviews for our niche.
Open-world idea generator that ESCAPES the same-research loop. Instead of reading Agora's own canon (which recycles), it anchors every run in EXTERNAL material — a rotating random draw from the 36 books in the vault's `04 Resources/Books/` plus COLD (un-mined) vault domains — and runs the IDEAOGENESIS operators (Gap, Inversion, Scale, Constraint-Drop, Fusion, Exaptation, Via-Negativa, Secret, Anomaly→Paradigm, Against-Method) to force cross-book collisions. Use when the owner asks to "generate ideas", "nové nápady", "forge ideas", "open world ideas", "go farther / something new", when a "Forge ideas" inbox task appears, or on a schedule. The books are the anchor; YOU (Claude) make the leap. Replaces the deprecated idea-forge.
Process an inbound outreach/correspondence reply and brief the owner in Slovak (their point + our answer + how we use it), then optionally draft a GATED reply. Use whenever a "Correspondence reply by <user>" inbox task appears, the Envoy flags a new reply/reaction, or the owner asks to "spracuj outreach", "čo prišlo", brief them on a GitHub comment/reply, or before any `approve` of an outreach/press action.
Build and extend the Dungeon OS 3D world (the live Three.js renderer in agora-game-server/static/index.html). Use whenever the task touches the dungeon scene, agent avatars in 3D, tiles/walls/props, lighting, the isometric camera, post-processing/bloom, sprites/labels, or "the game looks flat/cheap/laggy/should look presentable". NOT for the LLM agent loop (mcp_server.py), the brain (server/agora), or the React dashboard.