一键在 Manus 中运行任何 Skill

nucleo-anima-debug

星标39

分支2

更新时间2026年6月28日 21:49

Debug and develop NucleoOS firmware/ANIMA logic on the PC without flashing — the host harness compiles the real C and runs it locally. Use when working on ANIMA (the offline NL assistant), retrieval cascade, NLU, knowledge graph, or any firmware logic, and when deciding how to verify a change. Lists the gate suites to run before claiming a fix.

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

indecenti

indecenti/NucleoOs

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

Host-first debugging for NucleoOS / ANIMA

Default: do NOT iterate by flashing. The host harness compiles the real ANIMA firmware C (MinGW gcc) and runs it on the PC — no device, no WiFi, no cable. Flash only to confirm. Don't jump to flash→WiFi for logic work; that's the slow path and a user pref.

npm run anima -- "che cos'è nucleoos"   # one-shot query against the real cascade
npm run anima                            # interactive REPL
npm run anima:build                      # force rebuild the host exe
npm run anima:sweep                      # batch over tools/anima-host/queries.txt

Gate suites (run the relevant ones before claiming a fix)

npm run anima:gate — the 12-gate regression suite (0-hallucination, skill-routing, …). This is the pre-flight; it must be green before any flash.
npm run anima:route — dispatch/routing snapshot (catches cascade-order regressions).
npm run anima:reason — math/reasoning layer.
npm run anima:halluc / anima:meta / anima:boundary / anima:realistic — anti-hallucination stress (485-phrase suite + metamorphic harness).
npm run anima:typed / anima:typednl — typed knowledge graph.
npm run anima:local — the browser WASM cascade (policy + contract + capability + parity).
Per-subsystem (non-ANIMA): npm run <ir|link|mesh|eth|hw|skill|radio|logviewer>:test, npm run validate (registry/manifests), npm run i18n:gate.

Live device debug (when host isn't enough)

HTTP debug without flashing: /api/logs (no auth), /api/diag (consolidated). Pair with the 6-digit PIN shown on screen → session cookie → /api/fs. The IP/PIN drift per environment — read them fresh; don't hardcode. delete is NOT recursive. See docs/debugging.md.

Pitfalls

Host dim is 256, device dim is 192 — vector dims differ between host and device.
Don't rebuild the L1 index casually: k-means isn't reproducible and the reshuffle breaks gates (see memory anima-l1-index-rebuild / anima-index-provenance-guard).
L1 query stack: httpd task needs ≥24 KB or it panics.
Online tier is LLM-only by design (cloud Grok/Claude → browser WebLLM, never the offline cascade); the "Grok" teacher is actually configured as Groq (Llama) by default.

同仓库更多 Skills

同仓库

nucleo-release-dual

indecenti/NucleoOs

Build and ship NucleoOS to BOTH Cardputer units — the original and the ADV — in one pass. Use when the user wants to release/build/OTA/flash for both boards, the ADV and non-ADV Cardputers, or two devices at once. Encodes the key fact that it's ONE universal binary (runtime board auto-detect), so you build once and fan out the shipping to each unit, then verify per board.

2026-07-0939

nucleo-release

indecenti/NucleoOs

Build, gate, and ship NucleoOS to the Cardputer — firmware via OTA and the web/SD payload via the device file API. Use when the user asks to release, deploy, flash, OTA, sync the SD, or push a web app to the device. Covers the one-command release, firmware-only and SD-only paths, and the gotchas (PIN, .gz shadowing, never /MIR).

2026-07-0939

nucleo-native-app

indecenti/NucleoOs

Scaffold and implement a NucleoOS native firmware app (C++) correctly — anti-flicker, poll handler, RAM discipline, input routing, registration. Use when creating or fixing a native app in firmware/components/nucleo_app/.

2026-06-2839

nucleo-new-app

indecenti/NucleoOs

Scaffold a new NucleoOS web app correctly — manifest, www assets, registry entry, ANIMA launch aliases, i18n, icon — so it shows up, launches by name, and passes validation. Use when adding a new app to the OS, creating an app folder, or wiring an existing app folder into the registry.

2026-06-2839

nucleo-verify

indecenti/NucleoOs

How to verify a NucleoOS change actually works before claiming done or releasing — the full preflight matrix (registry/manifest validation, i18n, gzip parity, API spec drift, ANIMA gate, app tests) plus running the device simulator for web changes. Use when finishing a change, before a release, or when asked to verify/test that something works.

2026-06-2839

name	nucleo-anima-debug
description	Debug and develop NucleoOS firmware/ANIMA logic on the PC without flashing — the host harness compiles the real C and runs it locally. Use when working on ANIMA (the offline NL assistant), retrieval cascade, NLU, knowledge graph, or any firmware logic, and when deciding how to verify a change. Lists the gate suites to run before claiming a fix.

Host-first debugging for NucleoOS / ANIMA

npm run anima -- "che cos'è nucleoos"   # one-shot query against the real cascade
npm run anima                            # interactive REPL
npm run anima:build                      # force rebuild the host exe
npm run anima:sweep                      # batch over tools/anima-host/queries.txt

Gate suites (run the relevant ones before claiming a fix)

npm run anima:gate — the 12-gate regression suite (0-hallucination, skill-routing, …). This is the pre-flight; it must be green before any flash.
npm run anima:route — dispatch/routing snapshot (catches cascade-order regressions).
npm run anima:reason — math/reasoning layer.
npm run anima:halluc / anima:meta / anima:boundary / anima:realistic — anti-hallucination stress (485-phrase suite + metamorphic harness).
npm run anima:typed / anima:typednl — typed knowledge graph.
npm run anima:local — the browser WASM cascade (policy + contract + capability + parity).
Per-subsystem (non-ANIMA): npm run <ir|link|mesh|eth|hw|skill|radio|logviewer>:test, npm run validate (registry/manifests), npm run i18n:gate.

Live device debug (when host isn't enough)

Pitfalls

Host dim is 256, device dim is 192 — vector dims differ between host and device.
Don't rebuild the L1 index casually: k-means isn't reproducible and the reshuffle breaks gates (see memory anima-l1-index-rebuild / anima-index-provenance-guard).
L1 query stack: httpd task needs ≥24 KB or it panics.
Online tier is LLM-only by design (cloud Grok/Claude → browser WebLLM, never the offline cascade); the "Grok" teacher is actually configured as Groq (Llama) by default.