AI & ML

Move issues and external PRs through a state machine of triage roles — categorise, verify, grill if needed, and write agent-ready briefs.

Compact the current conversation into a handoff document for another agent to pick up.

LLM Operations -- RAG, embeddings, vector databases, fine-tuning, prompt engineering avancado, custos de LLM, evals de qualidade e arquiteturas de IA para producao.

Compiled CLI covering all 52 endpoints of the Anytype local API — objects, properties, tags, search, chat, files — one binary, no MCP server needed.

Use to forecast support contact volume and calculate staffing requirements for voice, chat, or async channels. Trigger for "how many agents do we need", "forecast our ticket volume", "Erlang C", workforce management or WFM planning, service level and occupancy targets, shrinkage, headcount planning for support, or when a staffing model keeps missing its service level.

Use to assemble a fair, evidence-backed coaching pack for a support agent's one-to-one from QA evaluations and conversation history. Trigger for "prepare a coaching session for X", "what areas does X need to improve", "areas of markdown for this agent", "what coaching opportunities stand out", "build a coaching agenda from these tickets", or preparing a weekly or monthly agent review.

Use to turn an ad-hoc CX reporting request into a versioned spec that can be re-run each period and actually compared across periods. Trigger for weekly or monthly QA and support reports, "same report but for last week", "generate the weekly digest", "supervisor report for each team", a report request pasted as a long prompt for the second or third time, or when two runs of the same report disagree.

Configure, troubleshoot, and use Hermes image generation — all 5 built-in providers (FAL, OpenAI, OpenAI-Codex, xAI, Krea), model catalogs, env vars, and .env location gotchas.

Embedding database for RAG and semantic search.

Expert skill for building LLM applications with the LlamaIndex framework — RAG pipelines, multi-agent orchestration, event-driven workflows, knowledge graph construction, production deployment, and evaluation. Use when working with LlamaIndex or comparing RAG and agent orchestration frameworks.

Pitfalls, examples, and edge cases for Hermes Kanban workers. The lifecycle itself is auto-injected into every worker's system prompt as KANBAN_GUIDANCE (from agent/prompt_builder.py); this skill is what you load when you want deeper detail on specific scenarios.

DSPy: declarative LM programs, auto-optimize prompts, RAG.

Karpathy's LLM Wiki — build and maintain a persistent, interlinked markdown knowledge base. Ingest sources, query compiled knowledge, and lint for consistency.

Build complex AI systems with declarative programming, optimize prompts automatically, create modular RAG systems and agents with DSPy - Stanford NLP's framework for systematic LM programming

Remove refusal behaviors from open-weight LLMs using OBLITERATUS — mechanistic interpretability techniques (diff-in-means, SVD, whitened SVD, LEACE, SAE decomposition, etc.) to excise guardrails while preserving reasoning. 9 CLI methods, 28 analysis modules, 116 model presets across 5 compute tiers, tournament evaluation, and telemetry-driven recommendations. Use when a user wants to uncensor, abliterate, or remove refusal from an LLM.

Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and reward model training. Use when need RLHF, align model with preferences, or train from human feedback. Works with HuggingFace Transformers.

Delegate coding tasks to Claude Code (Anthropic's CLI agent). Use for building features, refactoring, PR reviews, and iterative coding. Requires the claude CLI installed.

Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform

Skilled agent architecture replacing multi-agent system for RL training. Trigger when: (1) planning agent-guided training, (2) implementing tool-augmented LLM consultations, (3) comparing skilled vs multi-agent approaches, (4) designing simulate-verify loops for training, (5) implementing prompt evolution / learnable parameters, (6) understanding Claude Agent SDK integration in training, (7) debugging SkilledTrainer consultations or tool calls, (8) configuring agent safety bounds for training actions.

Multi-timeframe backtesting combining 15Min + 1Hour model signals. Trigger when: (1) multi-TF backtest, (2) combining timeframe signals in backtest, (3) validating multi-TF strategy, (4) --multi-tf CLI flag.

Model-trader version compatibility protocol: Embed version metadata in checkpoints, validate at load time. Trigger when: (1) training and live trading versions diverge, (2) models fail to load, (3) action interpretation issues.

Guard pattern for cleanup scripts: verify QC sentinel files and prompt for human review before destructive operations

Fix HOLD bias in RL reward function. Trigger when: (1) model learns to always HOLD, (2) trade rate is too low (<10%), (3) slippage penalty exceeds typical price moves.

Pattern for creating new Colab notebooks. Trigger when: (1) Creating a new notebook for experiments, (2) Adding notebook-based functionality, (3) Agent or validation notebooks, (4) Any notebook that uses GPU training infrastructure.

Frame causal questions and estimate treatment effects with assumption checks. Use when: (1) policy impact analysis, (2) A/B interpretation beyond correlation, (3) confounding diagnostics. NOT for: medical/legal conclusions without experts.

This skill should be used when the user asks to "create an agent", "add an agent", "write a subagent", "agent frontmatter", "when to use description", "agent examples", "agent tools", "agent colors", "autonomous agent", or needs guidance on agent structure, system prompts, triggering conditions, or agent development best practices for Claude Code plugins.

Build Gradio web UIs and interactive ML demos in Python. Use when creating or editing Gradio apps, Interface wrappers, Blocks layouts, event listeners, chatbots, or deploying demos to Hugging Face Spaces.

Compact the current session into a single detailed handoff message that can be pasted into a fresh agent run. Use when switching context, ending a session, or avoiding context-window loss.

Use Orca orchestration for structured multi-agent coordination: threaded messages, blocking ask/reply flows, task dispatch, worker_done/escalation waits, task DAGs, decision gates, coordinator loops, or decomposing work across agents. Use `orca-cli` instead for full ownership handoffs, including requests phrased as "hand off", "handoff", "handover", "give this to another agent", or "another worktree" when the user did not explicitly ask to supervise, monitor, wait for results, or coordinate a DAG. Use `orca-cli` for ordinary terminal control, lightweight terminal prompts, shell commands, Orca worktree management, reading or waiting on terminals, and automation of the browser embedded inside Orca. Use Computer Use for browser windows, webviews, Orca app UI, or desktop UI outside Orca's embedded browser.

Delegate coding to xAI Grok Build CLI (features, PRs).

Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference model needed, more efficient than DPO. Use for preference alignment when want simpler, faster training than DPO/PPO.

Generate a Startup Canvas combining Product Strategy (9 sections) and Business Model (costs + revenue) for a new product. An alternative to BMC and Lean Canvas that separates strategy from business model. Use when launching a new product or evaluating a startup concept.

Configure and use Honcho memory with Hermes -- cross-session user modeling, multi-profile peer isolation, observation config, dialectic reasoning, session summaries, and context budget enforcement. Use when setting up Honcho, troubleshooting memory, managing profiles with Honcho peers, or tuning observation, recall, and dialectic settings.

Axolotl: YAML LLM fine-tuning (LoRA, DPO, GRPO).

Shopify Admin & Storefront GraphQL APIs via curl. Products, orders, customers, inventory, metafields.

Agent D2 - Data Collection Specialist - Interviews, Focus Groups & Observation. Covers protocol development, question design, probing strategies, transcription conventions, and systematic observation. Absorbed D3 (Observation Protocol Designer) capabilities.