agent-architecture-auditFull-stack diagnostic for agent and LLM applications. Audits the 12-layer agent stack for wrapper regression, memory pollution, tool discipline failures, hidden repair loops, and rendering corruption. Produces severity-ranked findings with code-first fixes. Essential for developers building agent ap
autoresearch-agentAutonomous experiment loop that optimizes any file by a measurable metric. Inspired by Karpathy's autoresearch. The agent edits a target file, runs a fixed evaluation, keeps improvements (git commit), discards failures (git reset), and loops indefinitely. Use when: user wants to optimize code speed,
book-to-skillConverts books, documentation folders, and source collections (PDF, EPUB, DOCX, HTML, Markdown, RST, AsciiDoc, RTF, MOBI/AZW) into structured agent skills — extracting named frameworks, principles, techniques, and anti-patterns into a master SKILL.md plus on-demand chapter files, a glossary, a patte
checkpoint-promotionGate fine-tuned checkpoints with drift budgets, paired comparison, and forgetting checks before promotion. Use after a training run produces a checkpoint, when deciding whether a tuned model ships, or when a promoted model needs re-gating against updated goldens.
cli-for-agentsDesigns or reviews CLIs so coding agents can run them reliably: non-interactive flags, layered --help with examples, stdin/pipelines, fast actionable errors, idempotency, dry-run, and predictable structure. Use when building a CLI, adding commands, writing --help, or when the user mentions agents, t
dataset-curationPrepare, format, and validate datasets for supervised fine-tuning and preference training. Use when converting raw data into training format, applying chat templates, configuring sequence packing, generating synthetic training data, or writing a dataset card before a run.
eval-harness-firstBuild the evaluation harness that gates every fine-tuning run — golden sets, per-failure-mode graders, judge calibration, and base-model baselines. Use when starting a fine-tuning effort, when converting traces into an eval set, or when calibrating a judge against human labels.
finetuning-method-selectionDecide whether to fine-tune at all, and route to the right method (SFT, DPO/ORPO/KTO, GRPO/RLVR, continued pretraining) and base model. Use when starting any fine-tuning effort, when unsure whether RAG or prompting would suffice, or when choosing between preference-optimization and reinforcement met
grpo-rlvr-trainingTrain reasoning and verifiable-task behavior with GRPO and reinforcement learning from verifiable rewards (RLVR). Use when task success is algorithmically checkable (math, code, tool calls, structured output), when designing GRPO reward functions, or when a GRPO run diverges or reward-hacks.
harvestClone and install Claude Code skills from GitHub repositories. Use when the user provides a GitHub URL containing skills, wants to import skills from a repository, or asks to "harvest" or "install" skills from GitHub. Automatically detects skills, handles name collisions, and copies them to ~/.claud
herdr-atomicDeterministic scripts for driving herdr, the terminal workspace manager that keeps coding agents alive in persistent panes. Use this skill whenever herdr is mentioned by name, and also whenever the task calls for running a command in another pane, delegating work to a second or parallel coding agent
herdr-orchestratorRun a herdr multi-agent worker fleet as the ORCHESTRATOR — decompose the project's work, spawn and assign workers by tier (sonnet, opus, fable, codex-sol, codex-terra), answer blocked workers, and integrate results using the horch CLI. Use this skill whenever the session says you are an orchestrator
herdr-setupUse when the agents plugin (which carries the herdr fleet skills) was just installed or updated, before the first `horch` run on a machine, or when something in the fleet is missing - `horch: command not found`, `herdr is not installed`, `herdr too old`, a worker `--resume` that fails with "No conve
herdr-workerOperate as a WORKER in a herdr multi-agent fleet — a CLI session that receives tasks from an orchestrator, works independently, records progress in the session ledger, asks the orchestrator via horch when blocked, and reports a "[role] DONE" summary before its pane is closed. Use this skill whenever
llm-friendly-contextClarifies inputs, outputs, success criteria, decisions, and unresolved conditions so downstream consumers can execute without guessing. Use when writing or revising LLM-facing prompts, handoffs, planning artifacts, reviews, reports, or generated instructions.
loop-design-checkDesign a goal-oriented agent loop or review one for failure modes: spinning, Goodhart-gaming the verifier, or running a wrong answer to completion. Covers machine-decidable goals, loop types, plan/build/judge skeletons, and runaway prevention, not the mechanism that wires the loop. Use when designin
lora-qlora-recipesConfigure LoRA and QLoRA supervised fine-tuning with current best-practice hyperparameters. Use when writing or reviewing a LoRA/QLoRA training configuration, choosing rank/alpha/target modules, or deciding between LoRA, QLoRA, and full fine-tuning.
mcp-builderGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).
model-routerTriage what you're about to do and decide whether a different model should do it instead — route mechanical, well-specified work down to Haiku/Sonnet, escalate hard architecture/debugging/security reasoning up to Opus/Fable, and fan independent work out across parallel subagents. Consult this on eve
preference-optimizationAlign a fine-tuned model with preference data using DPO, ORPO, KTO, or SimPO. Use when preference pairs or thumbs-up/down feedback exist, when choosing between preference-optimization methods, or when a DPO run needs hyperparameters or debugging.
rag-architectDesign, tune and evaluate RAG pipelines with three deterministic stdlib-only Python scripts: a corpus-driven chunking optimizer, a requirements-to-design pipeline planner, and a retrieval evaluator (precision@k, recall@k, MRR, NDCG). Use when the user asks to design a RAG pipeline, choose a chunking
recsys-pipeline-architectDesign composable recommendation, ranking, and feed pipelines using the six-stage Source→Hydrator→Filter→Scorer→Selector→SideEffect framework popularized by xAI's open-sourced X For You algorithm. Use when building any system that picks "the top K items for a (user, context)" — content feeds, search
skill-creatorCreate 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 t
writing-for-agentsWriting documents for agents. Use when creating or editing skills, or modifying AGENTS.md or CLAUDE.md.