ai-agents-skills

Use when the user asks for a multi-agent discussion, panel review, multi-agent review, or multi-agent research session with role selection, round control, and template-based orchestration.

Use when drafting, validating, or normalizing bounded cross-agent task/result packets for parent-controlled handoffs. This is a packet-contract skill, not a runtime delegation broker.

Send email over SMTP using only the Python standard library, with plain-text and HTML bodies, file attachments, cc/bcc, reply-to, a dry-run preview, connection verification, and redacted config inspection.

Runtime helper for autonomous-research-loop ledgers. Use to initialize, append, validate, inspect, or smoke-test autonomous research loop state files without network, package installation, provider CLI calls, or live agent spawning.

Run bounded autonomous research iterations with evidence gates, recovery ledgers, and optional cross-agent handoffs. Use when the user asks to continue research autonomously, run a research loop, integrate autonomous agent loops, or keep improving a research workflow without repeated prompts.

Use for external DOI/ISBN/title resolution, manifest creation from pasted text, and paper retrieval after the local library-first workflow does not satisfy the request or the user explicitly opts out of library use.

Use when a research task benefits from an explicit phased workflow with structured source handoff across search, analysis, and writing, and when preserving citations across phases matters.

Use for review-only requests for papers or books when the user did not explicitly ask for annotation. Handles the normal single-agent review flow.

Use when starting a nontrivial research task to frame scope, success criteria, evidence plan, and the right downstream workflow before expensive browsing or multi-agent work begins.

Use when a research draft or report exists and needs a pre-final review for unsupported claims, ambiguity, scope drift, or missing evidence before delivery.

Use when a task fails, a user corrects the assistant, a capability is missing, or a recurring better pattern should be logged and considered for canonical ai-agents-skills integration.

Use when the user wants research, source gathering, current-information lookups, cross-source synthesis, or extraction from URLs/PDFs/videos using web search, page inspection, local tools, and sub-agents.

Runtime engine for url-to-screenshot. Use to detect a browser, capture a URL to a PNG over CDP, verify a captured PNG, or run the offline self-test of the deterministic core, without network, package installation, or live browser launch in the smoke path.

Use when the user wants to capture a web page (an http or https URL) to a clean PNG screenshot, in viewport or full-page mode, with cookie-consent dismissal, timeouts, SSRF-safe URL admission, and blank-output verification, across Linux, macOS, and Windows. The executable engine ships as the url-to-screenshot-runtime skill.

Use when the user asks to draw, refactor, extract, compile, or review a TikZ/PGF figure, especially structural diagrams such as flowcharts, DAGs, trees, commutative diagrams, finite graphs, automata, or research-derived summary figures. Prefer this skill when the output should follow a structure-first workflow like figure brief to spec to render to check to compile to review, and when document-facing output should use adjustbox width fitting.

Use only when the user explicitly mentions both annotation and review for a paper task. Produces annotated review outputs and supports an explicit add-to-Zotero step when the user asks for it.

Use when preparing optional AxiomMath AXLE MCP setup for manual formal-proof assistance.

Use when the user wants to search, retrieve, send, add, update, sync, export, convert, or clean books from the managed Calibre library runtime.

Use when the user wants to extract arXiv IDs or DOIs from research or RSS digests and turn them into getscipapers requests or manifests.

Use when the user wants to parse, convert, chunk, or structurally analyze PDFs, DOCX, PPTX, HTML, images, audio transcripts, or similar documents with Docling. Prefer this skill for local document parsing before ad hoc text extraction.

Use when the user wants a minimal Lean-style theorem skeleton, namespace wrapper, or generated formal statement stub.

Use at the start of computationally intensive local tasks to detect CPU, memory, disk, and optional accelerator availability before planning execution.

Use when the user wants a quick sanity check for a finite graph claim, construction, or encoding using the lightweight OpenClaw verifier.

Use when preparing optional LeanExplore MCP setup for Lean declaration search and formalization support.

Use when deciding whether a research claim should enter the optional Lean formalization lane.

Use when checking whether a Lean artifact can safely support a research claim.

Use when the user wants animated math (handwritten-style equation writing, equation morphing between derivation steps, and emphasis) rendered with Manim, as a silent video clip that can stand alone or be spliced into a slides-to-video deck. The free, optional companion to slides-to-video for math lectures.

Use when a research or engineering task needs automatic heavy-compute routing through the local broker for Modal-backed remote CPU, high-memory CPU, or GPU execution.

Use when the user wants a local research digest from tracked topics or wants to manage tracked research topics.

Use when the user wants RSS-based research/news digests, feed management, or feed health checks.

Use when the user needs SageMath for graph theory, combinatorics, algebra, spectral computations, or mathematical verification beyond what local Python tools can do.

Use when the user wants to turn prepared slides (PNG, PDF, or PPTX) into a narrated, captioned video in a chosen language and presenter role, using only free tools. A three-phase human-in-the-loop flow (analyze, draft transcript, render) gates rendering behind an explicit transcript approval.

Use when selecting, ranking, or validating submission venues for an existing scholarly manuscript or draft venue shortlist. A deliverable venue recommendation requires comparator-paper evidence for every ranked venue; bibliography overlap and offline placeholders are discovery signals only. Do not use for generic draft review, rewriting, paper retrieval, paper download, Zotero mutation, or one-off venue facts.

Use when the user explicitly wants Vietnam Thu Quan / vnthuquan / vietnamthuquan.eu ebook discovery, metadata, mirror checks, categories, formats, archive inspection, or controlled downloads through the local vnthuquan package.

Use when the user asks to send, get, retrieve, find, share, add, or search for a paper. This is the live OpenClaw Zotero workflow adapted for Codex and should take priority over external paper retrieval.

Use when drafting, rewriting, polishing, or revising prose while preserving author intent by tracking claims, evidence, caveats, and revision deltas.

Use when the user mentions OpenProse or prose workflows, wants explicit multi-agent research and synthesis, or wants a reusable orchestration pattern. In Codex, emulate OpenClaw OpenProse using spawn_agent, structured decomposition, and workspace artifacts.

Use immediately before calling a research answer done, final, or complete to verify evidence coverage, dates, remaining gaps, and delivery readiness.

Route VNU eOffice requests to an existing vnu_eoffice package or CLI: monitor updates, list latest incoming/outgoing documents, search by keyword, download attachments, and send requested files through Telegram.

Use for a clarity-only pass that must not change behavior — simplifying, renaming, de-duplicating, or restructuring code, configs, research scripts, or prose. Gates on understanding the target before touching it and re-verifies after each change so behavior stays fixed.