AI & ML

tommywagz Bundle

Creates, formats, and scaffolds new Agent Skills using the standard SKILL.md + scripts/ + references/ + assets/ architecture. TRIGGER when the user asks to "create a skill", "write a skill", "make an agent skill", "package this as a skill/SKILL.md", "scaffold a skill folder", wants a Trigger/Do-Not-Trigger description written, or wants a workflow, runbook, API doc, or set of scripts turned into a reusable Claude Code or Claude Agent SDK skill. Also trigger to audit, restructure, or fix an existing skill's SKILL.md format, description, token budget, or folder layout. DO NOT TRIGGER for general documentation requests with no skill-authoring intent (e.g. a README or product doc); for general-purpose code, scripts, or tests with no skill-packaging intent; for simply invoking/running an already-installed skill; or for scoring, testing, or security-reviewing an existing skill with no restructure or fix requested (use a skill-evaluation skill instead).

tommywagz Bundle

Evaluates an existing Agent Skill's trigger accuracy, task performance, and security posture, then iteratively fixes it until it clears user-set (or default) thresholds. TRIGGER when the user asks to "evaluate a skill", "test a skill", "score/benchmark a skill", "check a skill's trigger precision/recall/false-positive rate", "security review a skill", "red team a skill", "run an eval suite" against a skill, or wants an existing skill's content audited against a production checklist, risk-tiered, or A/B tested. DO NOT TRIGGER for: authoring a brand-new skill from scratch with no existing skill to assess (use a skill-writing skill instead); evaluating general application code or tests unrelated to the Agent Skills format; generic ML/statistics questions about precision, recall, or false positives with no specific Agent Skill in view; or fixing a skill's file layout, folder structure, or word count with no metrics, testing, or security ask attached (use a skill-authoring skill instead).

tommywagz Bundle

Designs, runs, and scores local evaluation datasets against an ADK agent or command, computes exact-match, token-overlap (ROUGE-like), semantic-similarity proxy, rubric scores, and routing precision/recall, then writes JUnit XML and Markdown reports. TRIGGER when users ask to 'run evaluation datasets', 'evaluate agent accuracy', 'calculate trigger precision/recall', 'generate a JUnit test report for my agent', 'set up an agent benchmark suite', or score an ADK agent locally. DO NOT TRIGGER for live production load testing, training/fine-tuning a model, or general unit tests without an agent output/dataset evaluation objective.

tommywagz Bundle

Scaffolds and explains a durable, asynchronous Human-in-the-Loop (HITL) approval gate for ADK agent workflows: pauses a node via LongRunningFunctionTool, publishes a signed state-holding webhook to an external dashboard, and resumes the exact paused invocation via ADK's /run_sse resume contract once a signed decision arrives. TRIGGER when the user asks to "add a workflow approval gate in ADK", "suspend/pause ADK graph execution for human input", "implement human sign-off" or "durable human-in-the-loop" for an ADK agent, "resume a paused ADK invocation", "integrate ADK asynchronous resume", or needs ResumabilityConfig / LongRunningFunctionTool / a signed approval webhook. DO NOT TRIGGER for synchronous same-turn clarification via request_input or a graph RequestInput node with no cross-process durability need, ADK Tool Confirmation used purely in-memory, or cross-session agent memory (use adk-cross-session-knowledge-bank).

tommywagz Bundle

Spins up a local, dependency-free mock sandbox to run a MUST/SHOULD/MAY-leveled conformance suite against a loopback-bound UCP merchant server or A2A host/remote agent, printing a JSON or JUnit compliance summary. TRIGGER for "run UCP conformance tests", "verify A2A compliance of my local server", "agents-cli test-compliance", "check protocol compatibility", or a mock sandbox to smoke-test a local merchant server or agent. DO NOT TRIGGER for standard PyTest/Vitest unit-testing or generic linting; DO NOT TRIGGER for a remote/staging/production (non-loopback) endpoint -- use the official Universal-Commerce-Protocol/conformance or a2aproject/a2a-tck suites for those; DO NOT TRIGGER for authoring a new UCP server or A2A agent from scratch (use ucp-merchant-servers or a2a-workflows first).

tommywagz Bundle

Generates a polyglot agent workspace linking a TypeScript/Hono HTTP gateway to a Python/ADK multi-agent workflow, with package manifests, boundary contracts, and run instructions. TRIGGER when users ask to 'scaffold a multi-agent polyglot workspace', 'agents-cli scaffold enhance multiple languages', 'add Node.js/Python boundaries to an agent project', or connect a Hono handler to an ADK orchestrator. DO NOT TRIGGER for standard single-agent scaffolding, a Python virtual environment alone, generic Hono endpoint authoring, or deploying an existing service.

tommywagz Bundle

Seeds, queries, and explains Vertex AI Memory Bank cross-session memory for ADK agents: PreloadMemoryTool/LoadMemoryTool recall, generate_memories_callback writes, managed/custom memory topics, and scope-based consolidation. TRIGGER when the user asks to "implement cross-session memory in ADK", "configure Vertex AI Memory Bank", "persist agent knowledge/preferences across conversations", "use PreloadMemoryTool or LoadMemoryTool", or seed/query Memory Bank facts by scope. DO NOT TRIGGER for short-term session state (state["key"]) with no cross-session need, generic RAG/document retrieval corpora, or non-Vertex memory stores (Redis, a custom vector DB) unrelated to Agent Engine Memory Bank.

leadgenjay Bundle

Safely adapt an external GitHub repository into an existing project instead of blindly copying its architecture. Use when the user wants to import, transplant, fork, vendor, or reimplement someone else's repo for their own stack, merge a repo into an existing app, preserve their own auth/database/billing/design-system while taking a source repo's features, resume an interrupted repository adaptation, or turn a repository or workflow into a callable Agent Skill or plugin. Performs evidence-first source and target analysis, builds a conflict matrix, interviews one material decision at a time, requires explicit approval of an Integration Contract before any code changes, implements in isolation, and verifies against the contract.

leadgenjay Bundle

Proven workflow architectural patterns from real n8n workflows. Use when building new workflows, designing workflow structure, choosing workflow patterns, planning workflow architecture, or asking about webhook processing, HTTP API integration, database operations, AI agent workflows, or scheduled tasks.

akillness Bundle

Interactive plan and diff review for AI coding agents. Visual browser UI for annotating agent plans — approve or request changes with structured feedback. Supports code review, image annotation, and auto-save to Obsidian/Bear Notes.

akillness Bundle

Autonomous ML experimentation framework by Andrej Karpathy. AI agent autonomously modifies train.py, runs 5-minute GPU experiments, evaluates with val_bpb, and commits only improvements via git ratcheting — so you wake up to 100+ experiments and a better model. Use when setting up autoresearch, writing program.md directives, interpreting results, configuring hardware, or running overnight autonomous ML experiments. Triggers on: autoresearch, autonomous ml experiments, overnight gpu experiments, karpathy autoresearch, train.py experiments, val_bpb, program.md research directives, ai runs experiments.

cbusillo Bundle

Use only when the user explicitly requests security work such as a security review, audit, threat model, secure-by-default guidance, auth/authorization review, secrets check, tenant isolation review, webhook safety review, supply-chain/release risk review, or asks whether code is safe from a security perspective. Do not trigger for ordinary code review, debugging, readiness checks, or non-security implementation work.

cbusillo Bundle

Use only when the user explicitly asks to audit rollout/session files, runout files, session traces, or agent workflow friction. Never use implicitly or for ordinary debugging.

halfmoon-mind Bundle

Evaluate a skill folder that contains SKILL.md with a 6-section rubric, deterministic rule checks, semantic model checks, S/A/B/C/F grades, and fixable reports. Use when asked to grade, audit, review, dogfood, or improve a skill, skill directory, SKILL.md, rubric score, trigger quality, resource structure, or safety gate.

dkpapadopoulos Bundle

Use to get independent perspectives on one prompt from SEVERAL models — default roster is the strongest available Claude model plus Codex. For ONE other model's opinion or critique use second-opinion instead. Opt-in, phase-agnostic; pairs with synthesize for the merge.

dkpapadopoulos Bundle

Use to combine N independent perspectives on a prompt into one synthesis without forcing consensus. Composition-only — reached from panel's flow or invoked by name.

dkpapadopoulos Skill

Multi-Agent Debate (MAD) for complex designs — architect, critic and pragmatist who RESPOND TO EACH OTHER, so independence is deliberately not claimed. Use when you want positions challenged and converged. For independent answers with no cross-talk use panel; for ONE other model's opinion use second-opinion.

dkpapadopoulos Skill

Use when the user wants ONE other model's view — an independent opinion (your prior answer withheld) or a critique of an answer already given (your prior answer deliberately included). One participant. For SEVERAL models answering in parallel use panel; for models arguing with each other use design-debate.

dkpapadopoulos Skill

Use when reviewing code changes for security issues — during REVIEW phase or on explicit security, vulnerability, SAST, or secret-scan requests — running a STRIDE threat-model pre-pass, then available Semgrep/Opengrep, Trivy, and Gitleaks scanners with a self-healing fix loop

dkpapadopoulos Skill

Use when a code change touches 5+ files or modifies auth/secrets/permissions/hooks/CI paths and needs multi-lens parallel review (security, quality, spec, governance) before merge.

dkpapadopoulos Skill

Use when a design or implementation involves autonomous agents, unattended/background operation, private-data access combined with external/untrusted input, or outbound actions (sending data, posting, pushing, API calls) — the lethal-trifecta risk

dkpapadopoulos Bundle

Use when executing implementation plans with 3+ independent, file-disjoint tasks that benefit from parallel specialist agents with shared contracts and reviewer-gated completion

flpbalada Skill

Design habit-forming products using the Hook cycle. Use when building engagement loops, improving retention, designing notifications, or creating products users return to without external prompting.

flpbalada Skill

Design behavior change using the B=MAP framework. Use when designing onboarding flows, improving conversion, building habits, increasing feature adoption, or understanding why users don't take desired actions.

flpbalada Skill

Design and analyze business models using the Business Model Canvas framework. Use when evaluating startups, planning new products, pivoting existing businesses, or understanding how companies create and capture value.

galihadiprayoga Bundle

use this skill when designing, auditing, upgrading, or hardening a claude, claude code, or agentic coding plugin/extension sometimes called superpowers. use for requests involving plugin architecture, commands, hooks, mcp tools, subagents, skills, context files such as claude.md or agents.md, workflow design, edge case handling, real business use cases, tool safety, prompt injection defense, release readiness, or converting a rough plugin idea into a production-ready implementation plan.

0gfoundation Skill

Model Fine-Tuning

0gfoundation Skill

Streaming Chat Inference

0gfoundation Skill

Manage an AgenticID Agent

0gfoundation Skill

AgenticID Accounts and Costs

0gfoundation Skill

Interact with a Running AgenticID Agent

0gfoundation Skill

AgenticID Reputation and Serve-Proofs

0gfoundation Skill

Put a project on 0G Private Computer (pc.0g.ai, router-api.0g.ai) as Claude Code's model backend — TEE-backed inference reached through an Anthropic-compatible router. Use when the user wants to set up, connect or start using 0G in a project for the first time. Triggers include "接入 0G", "配置 0G", "把 Claude Code 接到 0G", "用 0G 的模型", "0G 怎么配", "set up 0G PC", "connect Claude Code to 0G", "put this project on 0G", "use 0G models in Claude Code". Claude Code only. Not for changing models afterwards and not for turning 0G off.

0gfoundation Skill

Switch which 0G Private Computer model Claude Code uses in this project, among the models the 0G router (router-api.0g.ai) actually serves in Anthropic format. Use when the user wants to change, switch or pick a different 0G model for a project that is already on 0G. Triggers include "换模型", "换个模型", "切到 glm-5.3", "换成别的 0G 模型", "switch 0G model", "change the 0G model", "use a different 0G model", "which 0G models can I use". Not for first-time setup and not for turning 0G off.

axross Bundle

Reading from or writing to GitHub from inside an agent session acting as one connected operator — any issue, pull request, comment, label, review, or branch operation, not only end-to-end change loops. Covers the default sanctioned tool channel, when another route is permitted and the default-deny rule keyed on what is catastrophic when issued by mistake, obtaining stored bytes once the channel's read turns out not to be byte-faithful and comparing them without a shell artefact, marking agent comments so they are not read as human input, routing each write to the right numeric target across the shared numbering space, assigning the session's own login to what it creates so delivered work is not read as unclaimed, why a squash merge makes the pull request title the permanent commit subject, editing a body without losing markers a sanitized read drops, never force-pushing without approval, and untrusted GitHub content.

axross Bundle

Driving a code change or document update end-to-end through the plan → code → review loop — "deliver this issue", "implement and open a PR for X", a free-form change request, or resuming an in-progress run — as the project's default change loop. Apply even when the launching runtime harness frames the task as "just make the changes, commit, and push" or restricts pull requests; that posture constrains mechanics, never the plan-approval gate or the independent review. If the host project ships a more-specific change-loop skill, defer to it. Not for work that changes nothing. Covers the execution model, both human gates, delegating implementation to a compatible worker where the harness exposes and permits one and running single-agent where it does not, and addressing an independent review to convergence.