Total 53,207 skills, AI & Machine Learning has 8898 skills
Showing 12 of 8898 skills
Measures the task accuracy of text models served by MAX using standard benchmarks such as GSM8K, MMLU, HellaSwag, ARC, AIME, GPQA, TruthfulQA, WinoGrande, and BABILong. Use when benchmarking a served model, comparing it with model-card or reference scores, verifying that a new MAX model produces correct answers, or running repeatable dataset evaluations against a MAX OpenAI-compatible endpoint.
Debug silent corruption when a MAX model loads, compiles, serves, and generates tokens but output disagrees with a reference implementation. Use whenever parity debugging stalls on scalar taps, the model returns gibberish or wrong greedy tokens, logit cosine is high but argmax differs, or generation is coherent then diverges — during an architecture port, a quantization bring-up, a multi-GPU conversion, or after a MAX upgrade. Triggers on "parity failure", "silent corruption", "logits match but tokens diverge", "top-1 mismatch", "greedy divergence", and "model serves but generates garbage". Not for crashes on load or pre-serve scaffolding (use import-model). Mandates reference-vs-MAX tensor-dump comparators first, verify fixes numerically before recompiling, and serve-vs-pipeline bisect when dumps match but text diverges.
Benchmark a model served on MAX with the `max benchmark` command: measure throughput (tokens/sec), latency (TTFT, TPOT, inter-token latency), and GPU utilization by driving load against a running `max serve` endpoint. Use this whenever the user wants to benchmark, load-test, or measure the performance of a MAX model, get tokens-per-second / TTFT / TPOT numbers, run a concurrency or request-rate sweep, compare latency vs throughput, size a deployment, or produce benchmark JSON, even if they don't say "benchmark" by name. Also use when a `max benchmark` run fails to connect or reports zero/garbage numbers.
Use this skill whenever the user asks to create, improve, audit, or split prompts for AI video generators (Seedance, Kling, Veo, Runway, Luma, Pika, Sora, any image-to-video system). The skill also covers storyboards, shot lists, director treatments, dynamic montage, multi-clip story structure, camera direction, lighting, blocking, pacing, character continuity, dialogue, and sound design. Trigger even when the user says things like "придумай сцену для видео", "разбей на склейки", "сделай раскадровку", "улучши промпт для Kling", "переведи сценарий в промпты", "как снять X в AI-видео", or shares a prompt and asks to fix it.
Write one model-agnostic video prompt spec, then compile it to whichever video model you can actually call. Use for cross-model prompt work, model comparison matrices, reusing one brief across providers, or when the target model is not yet available and the work must proceed on another one.
Perform a systematic diagnostic scan of an AI workflow across 5 quality dimensions — prompt quality, context efficiency, tool health, architecture fitness, and safety — producing a scored report with prioritized remediation actions.
Orchestrate continual learning by delegating transcript mining and AGENTS.md updates to `agents-memory-updater`.
Use when a user wants to set up, configure, install, or reconfigure the opencode Fusion agent team - a strong main/build agent that plans and reviews but cannot edit files, delegating all edits to a cheaper sidekick subagent, plus an explore search agent and optional research/design/reviewer/vision specialists. Triggers include "set up fusion", "configure fusion", "install fusion", "fusion setup", "undo fusion" / "remove fusion", changing which models the main, sidekick, or explore agents use, or naming a subscription to start from a ready-made profile - e.g. "set up fusion with my OpenCode Go subscription" (also OpenCode Zen, ChatGPT Plus/Pro, GitHub Copilot). Writes the global opencode config under ~/.config/opencode/.
Iteratively refine code via requested rounds of parallel subagent review and orchestrator-applied fixes, with bounded adaptive continuation enabled by default for productive runs. Use when the user says "let's do three rounds", "fix/review loop", "back-and-forth review", "iterative review", or asks to harden a change with multiple rounds of codex/claude review.
Read-only multi-agent review of a GitHub Pull Request, with the synthesized report posted back as a PR comment so the author is notified. Use when the user wants to review a GitHub PR (github.com or GitHub Enterprise) and post a structured review back to the PR conversation. Auto-detects the PR from the currently checked-out branch when no locator is supplied. Requires `gh`, `uuidgen`, `jq`, and `uv` or `python3` on PATH. Activates the `review-anvil` engine in read-only mode and orchestrates the shell helper for posting.
Connect to MAXIA AI-to-AI marketplace on Solana. Discover, buy, sell AI services. Earn USDC. 13 MCP tools, A2A protocol, DeFi yields, sentiment analysis, rug detection.
Multi-agent review-and-improve loop for a GitHub PR you have checked out — posts a "starting" PR comment cc'ing the original author, runs requested rounds plus any adaptive continuation, applies fix commits to the local branch after each round, pushes everything back to the PR, then edits the starting comment in-place with the synthesized report (or a failure summary). Auto-detects the PR from the currently checked-out branch when no locator is supplied. Use when the user wants to "improve a PR", "review and commit fixes", "iterate on my PR", or "review and push back" against a checked-out PR branch. Requires `gh`, `uuidgen`, `jq`, and `uv` or `python3` on PATH. Activates the `review-anvil` engine in per_fix mode.