This skill provides guidance for building AI agents with the Noetic framework. It should be used when creating, modifying, or composing agent patterns using Noetic's step primitives, context layers, tools, sub-harness (coding-agent) steps, generative UI (OpenUI), chat platform integration (chat-sdk.dev), and agent harness. Triggers include mentions of "agent", "react pattern", "context layer", "spawn", "tool", "loop", "sub-harness", "step.claudeCode", "Claude Code / Codex / opencode / pi as a step", "generative UI", "OpenUI", "openUi codec", "openUiSurface", "tool ui / render fragment", "chat-sdk", "noeticAgent", "Slack bot / Discord bot / Teams bot with Noetic", "getChannelStream", or any Noetic-specific API usage in the packages/core directory.
Check `.sentrux/rules.toml` for layer and boundary violations and suggest fixes. Use when the user asks "does this respect the architecture?", "any rule violations?", "check boundaries", or "are we following the layering?". Reads the `reason` field on each violation to tailor the fix suggestion.
This skill provides guidance for writing evaluations, scored tests, and optimizations for Noetic agents using the @noetic-tools/eval package. Use when creating .eval.ts files, defining scorers, running noetic-eval, or optimizing step compositions. Triggers include mentions of "eval", "scorer", "benchmark", "evaluate", "noetic-eval", "noetic test", "optimize prompts", "GEPA", "baseline", "regression", or any @noetic-tools/eval API usage.
Query OpenRouter for available AI models, pricing, capabilities, throughput, and provider performance. Use when the user asks about available OpenRouter models, model pricing, model context lengths, model capabilities, provider latency or uptime, throughput limits, supported parameters, wants to search/filter/compare models, or find the fastest provider for a model.
Run a full structural audit of the noetic repo via the sentrux MCP server. Use when the user asks "how is the codebase?", "audit the repo", "full sentrux check", or after a large refactor/merge. Returns a quality score, health breakdown, rule violations, and the top untested risky files.
Snapshot current sentrux metrics as a baseline so later changes can be diffed against it. Use when the user says "mark a baseline", "checkpoint quality here", "I'm about to refactor — save where we are". Pair with `sentrux-diff` after the work.
Compare current sentrux metrics against the most recent baseline and report what got better or worse. Use when the user asks "did I break anything?", "diff quality since baseline", "what got worse?", or at the end of a refactor. Requires a prior `sentrux-baseline` or `session_start` call in the same MCP session.