Loading...
Loading...
Found 6 Skills
Plan Fireworks post-training from a task-shaped goal — recommend a method, cookbook entry, data plan, and eval from what the user already said, ask only decision-critical questions, and hand off to configure when ready to train. Use when the user describes a problem in plain language, is unsure which method or dataset fits, needs public dataset ideas, or wants to explore before committing. Read-only — does not create jobs or upload data. For explicit train requests with known model and data, use configure; for failures use debug.
Compatibility redirect and shared reference carrier for the Fireworks training skills. Use research to choose method, data, evaluation, and a cookbook entry; use configure to plan and run training; use debug for stuck, failed, or low-quality runs. Keep this skill installed so the three entry skills can load its progressive references.
Diagnose Fireworks training and deployment issues, including stuck or failed jobs, error messages, poor quality, checkpoint and resume problems, LoRA adapter load mismatches, and deployments that are ready but not serving. Use for systematic read-only triage before retrying. Use configure for a new or approved retry run; use research when the starting method, data, or cookbook example is unclear.
Plan, run, and monitor Fireworks training from a coding agent. Covers managed SFT, DPO, ORPO, and RFT plus Training API serverless and dedicated workflows, complete cost and parameter confirmation, active monitoring, checkpoints, deployment, resume, and teardown. Use for explicit train, fine-tune, deploy, or resume requests. Use research when method, data, or evaluation is unclear; use debug for a stuck or failed run.
Emit particles (fire, smoke, snow, fireworks, magic, trails) from an entity in a Decentraland SDK7 scene with the ParticleSystem component. Do NOT use for procedural entity motion (see animations-tweens), GLTF model effects (see add-3d-models), or 2D UI effects (see build-ui).
Run a bounded red-teaming autoresearch loop to generate LLM guardrail training data. You (the agent running the skill) are the attacker and the judge: you craft attacks and label every response. The only model the harness calls is the target under test, over any OpenAI-compatible API (OpenRouter, Moonshot/Kimi, Fireworks, Ubicloud, OpenAI, or custom), with its key in `.red-team/.env`. Every attempt (pass and fail) is written to JSONL ready for fine-tuning guardrails in Llama Guard format. Use when asked to red-team or stress-test an LLM for harmful content, jailbreaks, prompt injection, or backdoor/trigger behavior; to build a safe/unsafe dataset; or to mine novel attacks at scale. Complements the hacker skill (code/infra exploitability): this skill targets model behavior and produces data.