Skills dans ce dépôt
SkillsMP a collecté 61 skills depuis Ansh2508/assistance-system. Ouvrez un skill pour examiner sa source et ses détails.
Ansh2508/assistance-systemAffichage de 21 skills collectés sur 61.
How to point the installed `graphify` skill (a separate, upstream-managed tool at ~/.claude/skills/graphify) usefully at VELTH and veOS specifically, and which of this project's other skills should consume or feed its output. Load this alongside `graphify`…
Langue du texte source : anglais
The mandatory check before ANY claim that a VELTH fix/regression "did/didn't reach production," "is/isn't live," or "was/wasn't deployed." Load it the moment a deploy-status question comes up — from a bug report, a "was it deployed" question, or before…
Langue du texte source : anglais
The terminal node every VELTH task must pass before any success claim leaves the session. Load it at the MOMENT OF REPORTING — before writing "done", "fixed", "green", "passed", "verified", "working", "all tests pass", "ready to commit", or any status summary…
Langue du texte source : anglais
Wiring-verification and live AI-quality-testing discipline for veOS — how to prove a feature flag/env var actually reaches the running process, how to prove a deploy actually shipped new code (configuration drift detection via verify-deployed-image.sh), how…
Langue du texte source : anglais
When a research task asks for coverage across multiple countries, languages, or fields - go genuinely deep on every single one named, not a token citation per item. Use before dispatching or accepting cross-domain/cross-country research, and before reporting…
Langue du texte source : anglais
For risky architectural decisions - read primary sources in full, validate the claim yourself at your own scale, then encode the decision structurally (not as a comment or convention). Use before committing to a design based on a paper, benchmark, or best…
Langue du texte source : anglais
How to translate a completed, validated research finding (a paper, a benchmark, a simulation result) into an actual code or architecture decision. Distinct from research-simulate-encode, which covers finding and validating the research itself (its steps 1-2);…
Langue du texte source : anglais
Before reporting any AI-generated deliverable (code, research, a report) as complete, check it against the original request line by line - don't let it pass just because it superficially covers every topic asked about. Use before saying "done" on anything an…
Langue du texte source : anglais
Anshu's operating system for verifying AI-produced work when he cannot read the code fluently. Use this skill whenever he is about to accept, gate, merge, stage, or sign off on anything an AI produced — code, tests, migrations, documents, research claims, or…
Langue du texte source : anglais
Run all validation, test, and build commands. Use before reporting work done.
Langue du texte source : anglais
Use when the user wants to design, redesign, shape, critique, audit, polish, clarify, distill, harden, optimize, adapt, animate, colorize, extract, or otherwise improve a frontend interface. Covers websites, landing pages, dashboards, product UI, app shells,…
Langue du texte source : anglais
Check and update legal references in YAML files. Use when verifying ADR versions, DGUV updates.
Langue du texte source : anglais
Build Risk Vault YAML files (v2.0 RiskObjects). Use when creating vertical configs, hazard categories, or legal banks.
Langue du texte source : anglais
Verify velth LinkedIn drafts (carousel PDFs, captions, first comments) for content correctness before publishing. Checks every legal citation, norm reference, number, and date against the canonical fact sheet and primary sources; flags outdated law, wrong…
Langue du texte source : anglais
The strategic + technical architecture reference for VELTH's fine-tuning / continual-learning infrastructure build — the weight-level layer (LoRA, DPO, adapter registry) that sits above the already-built context_store (prompt-level) and…
Langue du texte source : anglais
Use BEFORE handing Cowork any task that involves git writes, verifying a fix by running tests, or "fix everything autonomously" scope. Load when a Cowork report claims a file is broken/corrupted, claims git state is corrupted, cannot run pytest/docker, or…
Langue du texte source : anglais
How to WRITE a Cowork dispatch — the prose and structure, not the workflow (velth-spec) or the topology (velth-graph-engineering). Load before writing any dispatch of meaningful size. Encodes real, hard-won lessons across the full working relationship (not…
Langue du texte source : anglais
Design the TOPOLOGY of any VELTH multi-actor task — which nodes exist, which transitions are legal, where humans gate, where determinism replaces judgment. Load this BEFORE decomposing any non-trivial task and BEFORE writing any Cowork prompt with more than…
Langue du texte source : anglais
Schreibt den Begleittext (Caption) für einen velth LinkedIn Carousel Post auf der Unternehmensseite, plus Erstkommentar mit Quellen. Lädt außerdem die Regeln für Upload und Buffer Draft. Immer laden, wenn ein Carousel Post getextet, überarbeitet oder als…
Langue du texte source : allemand
Use when merging origin/main (or any long-diverged branch) into a VELTH migration/feature branch with real conflicts - especially the repo-layer migration branches. Load BEFORE running `git merge`, and again before resolving each conflicted file. Encodes the…
Langue du texte source : anglais
The strategic decision lens for every non-trivial VELTH decision — architecture, roadmap, product scope, partnership, pricing, expansion, hiring. Load BEFORE deciding, not after building. Encodes the moat ladder VELTH is climbing, the data properties that…
Langue du texte source : anglais