Sandjab

Convertit un répertoire de skill Claude Code en source .clio en utilisant la session Claude Code courante comme modèle — aucune clé API, contrairement à `clio import` (chemin LLM) qui exige ANTHROPIC_API_KEY. Boucle agentique gather → draft → `clio check` → fix, sans plafond de retry. USER-INVOCABLE ONLY. DÉCLENCHEURS OBLIGATOIRES : '/skill2clio', 'skill2clio', 'convertis ce skill en clio', 'importe ce skill en .clio', 'récupère ce skill en clio sans clé', 'reconstruis le .clio de ce skill', 'skill vers clio sans clé API'. Ne PAS déclencher sur : compilation .clio → cible (c'est `clio compile`), récupération verbatim d'un skill émis par clio avec sidecar (c'est `clio import --mode strict`), ou paraphrases vagues.

Summarise a piece of text into a short paragraph.

Genera un resumen conciso de un texto en un párrafo corto.

Produit un résumé concis d'un texte en un court paragraphe.

Execute flow pipeline

Execute flow foo

Execute flow echo_str

Common ESP32-S3 platform reference for all Waveshare devices in this monorepo. Use when working with PlatformIO setup, toolchain, common libraries, or adding a new device.

Development reference for Waveshare ESP32-S3-ePaper-1.54G (200×200 4-color e-paper, ES8311 audio, SHTC3, PCF85063 RTC, battery). Use when working with this device, its e-paper display (EPD_1in54g driver), power rails, audio codec, sensors, deep-sleep, or restoring the factory firmware.

Development reference for Waveshare ESP32-S3-Touch-AMOLED-1.75C (round 1.75" watch form factor). Use when working with this device, its AMOLED display (CO5300 QSPI 466×466), touch (CST9217), audio (ES8311 playback + ES7210 mic-array AEC), power (AXP2101), IMU (QMI8658), or restoring the factory firmware.

Development reference for Guition JC3636K718 (ESP32-S3, round 1.8" 360×360 + encoder + RGB ring). Use when working with this device, its display (ST77916 QSPI), audio chain (PCM5100A DAC + NS4150B amp), 13× WS2812 ring on GPIO 0, encoder, microphone, DRV2605 + LRA haptics, or restoring the factory firmware.

Development reference for Waveshare ESP32-S3-Knob-Touch-LCD-1.8. Use when working with this device, its dual-MCU architecture, display (ST77916 QSPI), touch (CST816), haptics (DRV2605), or audio (PCM5100A).

Development reference for Waveshare ESP32-S3-Touch-AMOLED-1.8. Use when working with this device, its AMOLED display (SH8601 QSPI), touch (FT3168), audio (ES8311), power (AXP2101), IMU (QMI8658), or RTC (PCF85063).

Use when asked to produce a "monographie"/monograph for a subject in this scriptorium repo with the LEAN pipeline — verify-first/write-once variant of /monograph (prose written AFTER the council, section-batched jurors, deterministic lint), same guarantees, fewer tokens, more readable prose. Triggered by /leanmonograph <sujet>.

Use when asked to produce a "monographie"/monograph for a subject in this scriptorium repo AT REDUCED COST — the frugal variant of /monograph. Same verified pipeline (single best-of HTML reference document), but cheaper models on research/verification, a 2-juror council and hard caps. Triggered by /frugalmonograph <sujet>.

Use when asked to produce, generate or build a "monographie"/monograph (a single verified reference document) for a subject in this scriptorium repo — turning a topic (a phrase to a page) into a verified HTML document. Triggered by /monograph <sujet>.

Use when asked to arrange/classer/ranger les monographies du scriptorium par domaine, (re)définir ou réévaluer la taxonomie de la home GitHub Pages, classer un nouveau thème, ou rééquilibrer les domaines. Maintient tools/taxonomy.json (seule source de vérité), lu par tools/build_site.py. Triggered by /arrange.

Use when asked which MODEL and which REASONING EFFORT each agent of a skill or multi-agent workflow should use — i.e. a per-agent (model × effort) tiering table with MINIMAL / OPTIMAL / MAXIMAL configs. Accepts the target by NAME or by PATH. Triggers: "/quifaitquoi <nom-ou-chemin>", "qui fait quoi", "quel modèle/effort pour les agents de <skill/workflow>", "tiering model x effort", "model effort par rôle d'agent".

Use when facing complex reasoning tasks — competition math (AIME, HMMT, Olympiad), STEM problems, GPQA-style hard science questions, intricate logical deduction, algorithmic / code-competition challenges, or any problem where a single chain-of-thought feels insufficient or where the initial approach is uncertain. Spawns K independent parallel reasoning agents and then performs critical sequential deliberation across their trajectories — explicitly NOT a majority vote. Do NOT activate for simple factual lookups, casual conversation, straightforward code edits with an obvious solution, or pure information-retrieval tasks.