Browse Skills

Fine-Tuning Expert prepares JSONL datasets, configures LoRA/QLoRA adapters, sets hyperparameters, and trains custom or adapted LLMs. Use it for parameter-efficient fine-tuning with Hugging Face PEFT, OpenAI fine-tuning, instruction tuning, RLHF, DPO, and deploying quantized fine-tuned models.

Context Preservation proactively extracts critical values from tool results into working notes before automatic context pruning destroys them. Use it when an agent must retain URLs, API fields, IDs, and analysis conclusions across long tasks.

Error Recovery guides an agent through a decision tree for tool call failures: retry transient errors once, fix and retry structural ones, or escalate if blocking. It switches tools after repeated failures and logs failed tasks to a database.

Filesystem Search teaches agents when to use terminal_rg, terminal_glob, and terminal_exec for filesystem search: content grep with ripgrep, filename globs, and predicate queries for mtime/size/type.

Hive Image Generation creates and edits images from a text prompt using OpenAI gpt-image-2 through the Hive image service, billed to Hive credits. Use it to generate logos, illustrations, banners, product shots, and edit or restyle existing images from reference files.

Hive Note-Taking maintains a free-form scratchpad of decisions, extracted values, and open questions so context pruning doesn't lose anything you still need. Use it to record working data and reasoning between agent tool calls.

Terminal Tools Foundations teaches agents how shell tools behave: foreground/background auto-promotion, output handles, semantic exit codes, destructive warnings, and the bash-only policy. Use it whenever shell tools are available to avoid misreading grep exit 1 or losing truncated output.

Worker Delegation models a goal as a tracker table and writes a playbook that fans out independent row-level work to parallel workers via run_playbook. Use it in colony mode before fan-out to pilot the worker skill, then let convergence retry and resume unfinished rows.

Writing Hive Skills scaffolds a new Agent Skill for a Hive agent that conforms to the Agent Skills specification (SKILL.md with YAML frontmatter, optional scripts/references/assets directories). Use when the user asks to create, scaffold, add, or package a new skill for a Hive agent.

Think applies a ten-stage loop (observe, listen, think, connect, feel, accept, create, grow) to consequential or ambiguous decisions, architecture reviews, postmortems, and tradeoff analyses. Use it when a task requires deep reasoning, challenging assumptions, or verifying a high-stakes conclusion.

Wiki Retrieve builds and queries a vault-local contextual BM25 retrieval index with optional Nomic cosine reranking for finding relevant passages. It requires explicit consent for remote egress and falls back to BM25 when reranking is unavailable.

ChatGPT App Builder uses the mcp-use TypeScript framework to build, modify, debug, or review ChatGPT Apps and MCP servers.