0xUrsanomics
UserAn open framework for a persistent, self-regulating AI operations system on a coding-agent CLI: tiered memory + SSOT, security-first gates, a knowledge pipeline, multi-agent protocols, session survival, and a zero-dependency cockpit.
Categories
Indexed Skills (50)
ai-seo
Optimize content to get cited by AI search engines. ChatGPT, Perplexity, Google AI Overviews, Claude, Gemini, Copilot. Use when you want your content to appear in AI-generated answers, not just ranked in blue links. Triggers: 'optimize for AI search', 'get cited by ChatGPT', 'AI Overviews', 'Perplexity citations', 'AI SEO', 'generative search', 'LLM visibility', 'GEO' (generative engine optimization). NOT for traditional SEO ranking (use seo-audit). NOT for content creation (use content-production).
anti-ai-slop
Eliminates AI writing tells from all agent output and from any external text submitted for review. Based on Wikipedia's Signs of AI Writing catalog. This skill should ALWAYS be active as a background filter on everything the agent writes. Also trigger explicitly when the user says clean this up, remove AI slop, de-slop this, make this sound human, strip AI tells, edit for AI writing, humanize this, does this sound like AI, review this text, or pastes text and asks if it reads like AI. If the user asks to write, draft, rewrite, or edit anything, this skill's rules apply automatically.
autoresearch-system
Karpathy loop for the entire agent architecture. simulate end-to-end workflows, score, find the weakest link, fix, iterate.
content-humanizer
Makes AI-generated content sound genuinely human. not just cleaned up, but alive. Use when content feels robotic, uses too many AI clichés, lacks personality, or reads like it was written by committee. Triggers: 'this sounds like AI', 'make it more human', 'add personality', 'it feels generic', 'sounds robotic', 'fix AI writing', 'inject our voice'. NOT for initial content creation (use content-production). NOT for SEO optimization (use content-production Mode 3).
marketing-psychology
When the user wants to apply psychological principles, mental models, or behavioral science to marketing. Also use when the user mentions 'psychology,' 'mental models,' 'cognitive bias,' 'persuasion,' 'behavioral science,' 'why people buy,' 'decision-making,' or 'consumer behavior.' This skill provides 70+ mental models organized for marketing application.
forest-synthesis
Weekly cross-corpus synthesis across accumulated daily-forest digests. Cross-day pattern extraction invisible to single-day reading. Use weekly not daily.
office-hours
Force assumptions into the open before code is written. 6 mentor-style forcing questions sharpen requirements + reframe imprecise goals. Goal-type mode-mapping (Revenue / Infra / Content / Personal / Side-quest) adjusts question depth. Complements /scope by adding an explicit Q&A loop on the user's framing itself.
plan-ceo-review
Product-strategist mode review on a draft plan. 4 scope-modes (EXPAND / SELECTIVE / HOLD / REDUCE) chosen explicitly upfront and committed-to throughout. Catches scope drift, surfaces 10x opportunities, hardens rigor against silent failures. Includes Landscape Check (search-before-building) + Taste Calibration (codebase reference patterns) + 5 Prime Directives.
plan-design-review
Designer's-eye plan review. Rate each design dimension 0-10, explain what would make it a 10, then fix the plan to get there. 7 passes (info architecture / interaction states / journey / slop risk / system alignment / responsive+a11y / unresolved). Backed by 12 cognitive patterns + 3 Laws of Usability + Goodwill Reservoir heuristics from established design canon.
mcp-builder
Build high-quality MCP (Model Context Protocol) servers that let LLMs interact with external services through well-designed tools. Use when creating MCP servers to integrate APIs or services in Python (FastMCP) or Node/TypeScript (MCP SDK).
seo-aeo-audit
Optimize for search engine visibility, ranking, and AI citations. Use when asked to improve SEO, optimize for search, fix meta tags, add structured data, or improve AEO and AI visibility.
hunt
Finds root cause of errors, crashes, unexpected behavior, and failing tests before applying any fix. Not for code review or new features.
think
Turns rough ideas into approved plans with validated structure before writing code. Covers new features, architecture decisions, and value judgments about whether to build, keep, or remove something. Not for bug fixes or small edits.
write
Strips AI writing patterns and rewrites prose to sound natural in Chinese or English. Only activates on explicit writing or editing requests. Not for code comments, commit messages, or inline docs.
autoresearch
Autoresearch v2. Meta-Harness-informed self-improving skill loop. Upgrades from Karpathy hill-climbing to trace-informed diagnosis: full execution trace access, counterfactual diagnosis after regressions, additive-only safety valve, multi-candidate filesystem. Runs on any skill that produces scoreable output. Also runs on CLAUDE.md routing logic itself. If you can score it, you can autoresearch it.
debugging-discipline
Evergreen debugging + restart heuristics + cognitive-bias antidotes. Tier-2 (recall-fetched on-demand, NOT always-loaded). Lifted in condensed form from gsd-build/get-shit-done debugger-philosophy + universal-anti-patterns (audit 2026-04-28).
decision-council
Forces 5 AI advisor personas to argue about a decision, anonymously peer-review each other, and synthesize a verdict with anti-false-consensus guardrails. Auto-detects high-stakes decisions from context. Trigger on: pricing, deal terms, partnerships, market entry, investments, exit strategies, go/no-go calls, trading decisions, DeFi/protocol entry, market entry/exit, futures positions, training program changes, competition prep, injury decisions, career moves, big purchases, financial planning, macro economic events (central-bank decisions, FX, inflation), geopolitical developments affecting markets, major AI/Web3/tech releases that change capability or competitive landscape, or any question where being wrong costs money, health, or reputation. Also trigger on "run council", "stress test", "what am I missing", "devil's advocate", "sanity check", or when the user leans toward an answer already. Do NOT trigger for brainstorming, writing, code, factual lookups, routine daily choices, or low-stakes ops.
memory-offload
Manages the agent's context overflow by offloading dense operational state to an Agent-Memory namespace in your knowledge graph and pulling it back when needed. This skill should ALWAYS be active as a background process. Auto-triggers when conversations get long or context-heavy. Also triggers on manual commands like "offload this", "save to memory", "dump context", "remember this for later", "store this in the graph", "memory dump", "context dump", "save state". On the PULL side, this skill triggers at the START of every conversation to check for relevant Agent-Memory pages, and mid-conversation when a topic matches stored context. Also triggers when the user says "pull my context", "load memory", "what do you have stored", "check your notes", "refresh context".
negotiation
Tactical negotiation support with game theory, emotional regulation, EV analysis, and structured counter-offer frameworks. Activates on any deal negotiation, pricing discussion, contract term pushback, partnership terms, or when the user is in a live back-and-forth with a counterparty. Trigger phrases: "negotiate", "counter-offer", "he's pushing back", "what should I say", "how do I respond to this", "they want to change terms", "they won't pay", "run negotiation analysis", "BATNA", "walk away?", "am I getting screwed". Also auto-detect: the user forwarding counterparty messages and asking for help responding, the user sounding frustrated/angry about a deal conversation, pricing objection patterns.
premortem
Run a premortem on any plan, launch, product, hire, strategy, or decision. Assumes it already failed 6 months from now and works backward to find every reason why. Produces a revised plan with blind spots exposed. MANDATORY TRIGGERS: 'premortem this', 'premortem my', 'run a premortem', 'what could kill this', 'future-proof this', 'stress test this plan', 'what am i missing here', 'find the blind spots'. STRONG TRIGGERS: 'what could go wrong', 'am i missing anything', 'poke holes in this', 'where will this break', 'devil's advocate this'. Do NOT trigger on simple feedback requests, factual questions, or LLM Council requests. DO trigger when someone has a plan or commitment where the cost of being wrong is high.
schedule
Create a scheduled task that can be run on demand or automatically on an interval. Use when the user says 'schedule X', 'remind me in N minutes', 'every weekday at 9am', or wants to convert a session task into a recurring/one-shot cron job. Outputs a create_scheduled_task tool call with cronExpression OR fireAt fields filled correctly.
skill-creator
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.
sycophancy-guard
Always-on background filter preventing sycophantic belief drift. Based on Chandra et al. 2026 proving even ideal Bayesian users spiral from sycophantic AI, including factual sycophants cherry-picking true info. ALWAYS active on every response like anti-ai-slop. Explicit triggers include "drift check", "am I spiraling", "check my assumptions", "red team this", "belief audit", "sycophancy check", "are you just agreeing with me", "challenge this", "what am I wrong about". Monitors confirmation bias, selective evidence, escalating agreement without new evidence, unverifiable strategic claims without pushback. Auto-triggers a Decision Council on detected drift. Logs belief positions to your knowledge graph. The immune system against RLHF sycophancy over extended single-user interactions.
task-extractor
Identifies tasks, action items, and assignments from raw text, assigns them to team members with @mentions, sets deadlines, and posts to an #action-items topic in your team chat. Use this skill whenever raw text contains action items, someone says "[name] to handle", "need to", "should", "follow up", "deadline", "by [date]", "TODO", or a message router classifies a message as task-action. Also trigger when processing meeting notes that contain action items, when the user says "assign this to", "create a task for", "what needs to be done", or "action items from". This skill handles the full task lifecycle from extraction through assignment and notification.
test-before-bulk
Mandatory gate before any batch operation that writes to a CRM, a knowledge graph, a spreadsheet, a vector index, or any other durable store. Test 3-5 first, fix the SKILL not a one-off script, then bulk-execute with throttling and a kill switch.
voice-profiler
Extract voice patterns from session history and update Voice-Profile.md
agent-dispatch
Domain-aware subagent dispatcher. Maps task → domain (research / build / draft / verify / ops) → recommended subagent_type + tool boundaries + briefing scaffold. Adapted from a phase-based activation pattern.
branching-workflow
When and how to use the coding-agent CLI's `/rewind` to abandon a bad branch and restart from a prior turn. Documents the recovery primitive for the layered eval architecture's Layer 3 (universal recovery toolkit) and the branch-and-retry path. Use when an output is going off the rails, when a Critic verdict says BLOCK, or when a path-of-least-resistance failure has corrupted the conversation context. Pairs with /save (extract insight first), /scope (re-frame after rewind), and the Critic skill (verdict-driven branch trigger).
code-review
Staff engineer code review for scripts, tools, and infrastructure changes
critic
Tool-grounded Critic skill (layered eval architecture, Layer 2). Dispatched as a subagent on high-stakes outputs. Grounds the verdict against tool-fetched evidence, not self-reflection. Returns SHIP / SHIP_WITH_FIXES / BLOCK with fixes + reason + confidence. Pairs with the stake classifier (Layer 1) and branching-workflow (Layer 3). Used by content drafts, regulatory-style takes, partner outreach drafts, and proposals.
daily-forest
Daily 21:30 (local) skill. Synthesize today's Signal/ harvest into a forest-level read. movers / silences / tone shifts / cross-stitches / anomalies. Indexed back to vector_brain so the synthesis substrate compounds over time.
deep-research
Multi-phase structured research pipeline with citation tracking, evidence persistence, and knowledge-graph-native output. Use when the user asks for "deep research", "comprehensive analysis", "literature review", "research report", "state of the art on", "structured research pipeline", "multi-source investigation", or wants a citation-tracked report saved to the knowledge graph. Different from a single-pass web-search lookup (quick answers). this skill is for multi-phase work that needs evidence persistence, claim-level verification, and entity cross-references in the knowledge graph. 4 mode tiers (quick / standard / deep / ultradeep) with 3-8 phases each. Outputs to `pages/Research___YYYY-MM-DD___[topic-slug].md` with auto-created `Entity/` and `Source/` cross-link pages.
dreaming
Automated memory consolidation. 3-phase process that promotes high-quality memories and decays stale ones
eod-summary
End-of-day digest pulling today's wins, decisions, status changes, follow-ups, new artifacts, tomorrow's queue from session.jsonl + tasks + tracker files. User-invoked via /eod or auto-fire at end of day.
filing
Give received artifacts a canonical home AND an index entry so recall can reach them. Intake, placement, indexing. Not a filesystem beautification pass.
goals
Active-goal registry. Tier-1 north star steering. Maintains memory/state/active_goals.json with goals + milestones + requirements. Slash command for list/new/done/pause/abandon/touch/note/show + milestone CRUD + req CRUD. Drift-audited weekly. v2 adds work breakdown + ASCII progress bars.
graph-hygiene
Logseq knowledge-graph maintenance lint. scans 7 categories (stale pages, orphans, missing properties, dead links, missing compiled-truth/timeline structure, tag inconsistency, duplicate near-names), emits severity-tagged findings. Never auto-deletes. flags for /review only. Read by graph-hygiene-weekly cron.
idea
Capture ad-hoc thought into the active project's Inbox section. Falls back to side-quests if no active project. Lightweight quick-drop primitive. no friction, no clarifying questions.
inline-tags
Append 1-3 hashtag tags to the end of every TG reply for native TG search. Domain + project + status taxonomy. Replaces full TG-Topics migration without rebuild cost.
intel-analyzer
Analyzes dropped links, articles, research papers, and raw text to determine whether the content is significant signal or noise. Produces a structured significance score across 6 dimensions (source credibility, verifiability, novelty, actionability, relevance, technical depth), delivers a verdict (HIGH SIGNAL / MODERATE SIGNAL / LOW SIGNAL / NOISE), and auto-ingests high-scoring content into the knowledge graph as Intel/ pages with cross-links. Use this skill whenever the user drops a URL, forwards an article, shares a PDF, or says "analyze this", "signal check", "is this legit", "worth reading?", "parse this", "signal or noise?", "check this link", or any variation of asking whether content is significant. Also trigger when the user pastes raw text from a social feed or chat and wants a quality assessment. This is the front door to the knowledge graph: only scored, verified content enters.
loop-experiment
Run a Karpathy-loop / autoresearch-style optimization experiment on any task with measurable fitness. Bounds the loop with the stack's discipline (fitness-function gating, branch/worktree isolation, anti-slop, audit logging). Use when running prompt evolution, code optimization, content A/B testing. anywhere with a stable measurable metric and cheap iteration.
loop-library
Catalog of repeatable agent-loop workflows (goal -> iterate w/ verification -> stopping condition -> output). Each loop is a reusable SOP for autonomous iterative work with quality gates. Run via /loop (self-paced or interval), a loop-until-done runner, a fan-out orchestrator, or manually.
meta-memory-review
Periodic self-optimization of the memory STRUCTURE (not its content). Review tier layout, dedupe, prune, and re-index the memory system.
prd
Structured PRD/spec generator. Transforms vague ideas (skill / infra / process / feature) into requirements, success metrics, build phases, and go-or-park gates BEFORE implementation starts. Lifted from wwwazzz/senior-pm-prompt, adapted to the stack (helicopter-check + ship-discipline native).
progressive-disclosure
When a chat reply exceeds the wall-of-text threshold, attach as a .md file with a 3-line summary in chat instead. Reduces mobile reading friction + token tax on future sessions reloading the transcript.
reality-feedback-create
Layered eval architecture, Layer 4 auto-create. Fires from the Stop hook chain after critic_dispatch. Appends an entry to the reality_feedback.sqlite ledger when the classifier verdict is high-stakes, with surface inference + outcome window + Critic metadata. Operator labels the outcome later via weekly review (skills/reality-review-weekly/SKILL.md). Outcomes feed a Layer 1 v2 classifier's training set over months.
reality-review-weekly
Layered eval architecture, Layer 4 weekly review. Cron-fired weekly. Queries reality_feedback.sqlite for entries where outcome IS NULL AND timestamp + window has elapsed. Surfaces N due entries via a chat message for operator labeling. Outcomes feed a Layer 1 classifier's training set over months. Decision gate at month 1 - >=50% labeling rate then continue Layer 4; else DROP.
recall
Semantic recall across the agent's full knowledge base via the vector_brain index (knowledge-graph pages, knowledge files, memory, reviewed outputs, persona memory). Trigger whenever the user asks 'what do we have on X', 'find me info about Y', 'search memory for Z', 'do we have a note on W', 'recall X', 'remember anything about Y', 'what do I know about Z', or whenever the current conversation needs grounding against prior work you don't have in active context. Use this BEFORE guessing or saying 'I don't have context on that'. the vector brain likely does. Returns ranked semantic matches with similarity scores.
redact
Per-entry redaction primitive for memory/ + outputs/ files. Strips matching content, replaces with [REDACTED YYYY-MM-DD reason] marker, logs to logs/redact-log.jsonl. Use when a credential, PII, or other sensitive content slipped into a tracked file.
vendored-deps
Track manually-installed GitHub dependencies (tools/binaries/repos adopted OUTSIDE a package manager) for upstream updates we would otherwise silently miss.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.