AI & ML

brenordv Skill

Scope-discipline review lens for a drafted plan. Reads the original user prompt and the plan, and flags scope creep, gold-plating, speculative generality, and work that solves problems the ask never raised. The review panel's one lens that argues for less. Never writes plans or code. Use in Stage 1 of the full-work workflow, or whenever planned work needs checking against what was actually asked.

rhenter Skill

Use for Django-specific model, ORM, view, middleware, migration, admin, transaction, or application-structure work.

ionden Bundle

Use when asked to audit, optimise, slim down, or fix Claude Code subagent definitions (.claude/agents/*.md) — set a proper `tools` allowlist (an agent with no `tools` field inherits every tool on each launch), cut bloated or duplicated system prompts, tighten descriptions so they trigger reliably, and right-size the model. Triggers: "optimize my subagents", "optimise my agents", "audit my subagents", "subagent optimizer", "my agent uses too many tokens", "fix the tools list", "review .claude/agents", "make my agents cheaper".

sohaibseg Bundle

Build and sharpen a project's domain model. Use when discussing codebase terminology, writing or editing a CONTEXT.md, or recording or editing an ADR.

sohaibseg Bundle

Prepare and transfer actionable context from a long Codex session into a fresh session or task. Use when the user asks for a handoff, context transfer, session restart, continuation in a new task, compact briefing, recovery package, or a durable summary that lets another agent resume without rereading the full conversation.

padosoft Bundle

Use this skill when measuring whether a model-backed feature works — building or changing an evaluation harness, a golden dataset, a metric, a scoring report, a regression gate on prompt or model changes. Also when the user asks how to know a prompt change made things better, how to stop a model upgrade silently regressing, what to put in a test set, how to score free-form output, or wants red-team coverage. It covers datasets and reports as versioned artifacts, the isolation an evaluation run needs, cohorts, and what must never leak into a report. Do not use it to write application tests (padosoft-test-integrity), to choose a model, or to design prompts.

padosoft Bundle

Use this skill when creating, editing or reviewing an Agent Skill of the padosoft/skills repository, when the user wants to turn a recurring workflow, a checklist or a set of guidelines into a reusable skill, or when they ask where a skill belongs (profile, scope, package) or why the repo CI is failing on catalog, profiles or manifests: it guides the whole creation with scaffolding, the Padosoft conventions and the automated checks. Do not use it to write the technical domain content (that is the job of the skill you are creating) nor to install existing skills (padosoft-skills-router handles that). It also keeps the provenance of the work — dates, customers, people, ids, credentials — out of anything that gets published.

stbenjam Skill

Use when reviewing API compatibility before approving a pull request

stbenjam Skill

Review release evidence for missing approvals and risky changes. Use when validating a release candidate.

stbenjam Skill

Missing Description Skill

stbenjam Skill

Use when modifying agentic contextual building blocks like skills (SKILL.md), slash commands, agents, hooks, plugins, marketplaces, and instruction files (CLAUDE.md, AGENTS.md, GEMINI.md, QWEN.md, Cursor, Copilot, Cline, or Kiro rules), tool configuration such as an OpenCode `opencode.json`, Muse Code hooks (`.muse/hooks.json`), a Grok Build `.grok/` project layer or a Google Antigravity customization root (`.agents/`, `.agent/`, `_agents/`, `_agent/` — hooks, MCP config, rules, registries and plugins), and committed project memory (`.agents/memory/`). Run skillsaw on the files you touched, apply autofixes, resolve remaining violations with `skillsaw explain` guidance, and re-lint until clean before considering the work complete.

stbenjam Bundle

Analyze upstream specs (agentskills.io, Claude Code plugin/marketplace format, OpenAI Codex plugins/marketplace, OpenClaw, MCP and MCP Registry, CodeRabbit, APM) for changes, identify gaps in skillsaw's rule coverage, and create or update PRs to close those gaps. Use when performing periodic maintenance on the skillsaw linter.

padosoft Bundle

Use this skill when designing or changing the corpus a model retrieves from — the document store, the chunk table, identifiers, statuses, the promotion of knowledge into it, the graph of relations between documents. Also when the user reports that a document could not be deleted, that the same file was indexed twice, that retrieval grounds answers on an outdated or unreviewed page, that two projects collided on an identifier, or asks how a model should be allowed to write into the knowledge base. It covers the idempotency anchor, canonical typing, the human gate on promotion, trust ordering at retrieval, and the delete path. Do not use it for permissions and ingestion trust (padosoft-rag-ingestion-security) or for choosing an embedding model.

padosoft Bundle

Use this skill when writing or reviewing Laravel/PHP code — a controller, a FormRequest, a service, a job, a migration, an Eloquent query, a model event, a queued or bulk operation — and whenever a symptom shows up: a controller that grew into the business logic, an N+1 found in the logs, a bulk command that runs out of memory, a job that retries something it should not, a soft-deleted row reappearing in a count, a model event that fires on a mass update and does not. It applies to any Laravel version: the rules are about the shape of the application. Do not use it for a security review (padosoft-laravel-security-review), for what goes in a log (padosoft-logging-discipline), or for infrastructure and deployment.

padosoft Bundle

Use this skill when code runs at the edge in front of an origin — a Cloudflare Worker or equivalent that proxies, renders or caches on behalf of a backend. It covers who owns a control when two layers could implement it, cross-site request forgery validated at the edge, the two-hop client-address model, header forwarding and what must never reach the origin, private caching and key design, cross-origin policy, cookies, server-side rendering at the edge, secrets and bypass tokens, and edge rate limiting. Also when the user says a request is blocked and nobody knows by which layer, a header is being spoofed, a cached page showed another user's data, a bypass works with only a user agent, or asks whether a control is really on. Do not use it for the origin application's own review, for CDN configuration, or for edge performance tuning.

padosoft Bundle

Use this skill when building or reviewing something that calls a model or hosts an agent — an LLM adapter, a tool-calling loop, an MCP server, a prompt-driven feature, a spend budget, a trajectory or replay store, a similarity or scoring threshold, an agent-generated artifact that a human is meant to approve. Also when the user reports a runaway bill, prompts or customer data ending up in logs or traces, an agent doing more than intended, a "confidence score" being treated as a decision, or generated output being accepted as evidence. Do not use it for prompt wording and model choice, for training or fine-tuning, or for building the skills themselves (padosoft-skill-creator covers that).

padosoft Bundle

Use this skill whenever an agent is about to write something it has not verified — a call to a helper whose signature it inferred from the name, an enum case that sounds plausible, a config key, a column, a file path mentioned in a plan but never opened, a docblock explaining a mechanism nobody read, a commit message asserting what the code does. Also when the user says the agent invented an API, hallucinated a method, confidently described behaviour that does not exist, or asks how to stop it happening. It gives the rule (ask, or log the doubt — never the silent third option), what must be verified, and how autonomous mode changes the answer without removing the obligation. Do not use it for verifying facts about the outside world, for prompt engineering, or for reviewing code somebody else wrote.

stbenjam Skill

Review release evidence for missing approvals. Use when validating a release candidate.

stbenjam Skill

Walk a deployment through staging, canary, and the full fleet. Use when releasing a service to production.

stbenjam Skill

Reference for the Claude API and Anthropic SDK — model ids, pricing, parameters, streaming, tool use. TRIGGER — read BEFORE opening the target file whenever the prompt names Claude or Anthropic in any form, or the task is LLM-shaped with the provider unstated. SKIP only when another provider is being worked on.

dioad Bundle

Activate when the user wants to measure how well a Go project is tested, find weak spots in the test suite, or act on a prior test-quality report. Combines mutation testing (gomutants) with per-test coverage-contribution analysis to produce a quantitative scorecard, a test contribution map (redundancy / consolidation candidates via greedy set-cover), and an LLM-actionable findings doc (doc/test-quality-findings.md) covering weak assertions, coverage gaps and risk. Phase 2 works through findings one commit at a time.

naw103 Bundle

Coordinate parallel coding agents with Foremerge's local Git-compatible CLI and MCP server. Use when an agent should publish intent before editing, claim semantic scope, detect duplicate or incompatible work, coordinate on durable conflicts, publish a ChangeSet, run a trusted named verification check, accept validated work, or record its Git integration.

austin-jc Skill

Scaffold a new skill or plugin in the agent-skills marketplace repo. Use when the user asks to add, create, or scaffold a skill or plugin in this repository, or asks how this repo is structured.

stbenjam Skill

Check release evidence against policy. Use when reviewing a release candidate.

stbenjam Skill

Move a service off retired Anthropic and OpenAI model ids onto their current replacements. Use when a provider announces a retirement or an API call returns a deprecation notice.

stbenjam Skill

Use when scanning response schemas for incompatible API changes

stbenjam Skill

Walk through a demo release review end to end. Use when showing the release workflow to a new team member.

chrisadkin8 Bundle

Turn a /research note into an implementation spec inside the repo it changes, grounded in the code with file:line citations, then check it and give it a cold review. Use when the user runs /spec, or asks to write a spec, plan or prompt file for changing or enhancing a repo, especially from a research note. Accepts a research note path or a description of the change; "finish <spec>" re-checks and verifies a spec after editing; "spike <spec>" runs its spike questions in sandboxed headless sessions and folds the answers back in.

chrisadkin8 Skill

Give a markdown file one adversarial cold read by an agent that never saw this conversation, then relay what it found. Use when the user runs /cold-review, or asks for a cold, fresh-eyes or adversarial review of a document - a walkthrough, runbook, README, design doc, spec or research note. Accepts the path to a markdown file; "prompt <path>" writes the review prompt for the user to run in a fresh session instead of launching an agent.

newblackai Bundle

Give any AI agent or codebase the fashion studio of The New Black AI (thenewblack.ai) — 47 workflows that generate and edit fashion images and videos (design, product-to-model, virtual try-on, fabric, sketch, video, HD), read and write the account's creations and tech packs, publish to Shopify and social accounts, and talk to the account's own AI agents — by REST API, by the `tnb` CLI (npx @thenewblack/cli) or by MCP. Use when the user wants to connect a system, store, script, folder of photos or agent to The New Black AI, mentions the TNB API, the tnb CLI, MCP connector, tnb_live keys, or AI fashion generation by API.

yousefomar724 Bundle

Edit video in the real CapCut desktop app on Windows through computer use — driving the UI the way a person does (grid screenshots, keyboard shortcuts and the UI Automation tree wherever CapCut exposes it) instead of generating video from code. Builds and applies user-approved CapCut procedures one demonstrated capability at a time and writes each exact working procedure back into this skill. Use this whenever the user wants anything done inside CapCut or CapCut Pro — splits, trims, removing pauses or filler words, bad takes, jump cuts, push-ins, auto captions, text animations, B-roll, 9:16 reframes, vocal isolation, background removal, camera tracking, transitions, color, export — or asks to map CapCut's controls, train or teach the agent on CapCut, or continue the CapCut capability ladder. Not for code-rendered video (HyperFrames, Remotion, ffmpeg pipelines) or CapCut web/mobile.

grisuno Bundle

LazyOwn RedTeam Framework — penetration testing and C2 via MCP

grisuno Bundle

Use when running LazyOwn inside a Hermes agent session. Provides Hermes-optimized MCP tools with compact output, checkpoint resume, dynamic rules, and native delegation planning.

zhengqiuyang Skill

Authors test suites for MCP servers using the mcp-test tool: functional YAML assertions, prompt-injection security probes, false-positive tuning, and CI gating. Use when building or maintaining an MCP server that needs more than manual poking in an inspector — especially before exposing it to agent clients you do not control.

zofe Bundle

Design and implement a state machine (workflow) on a Rapyd Admin model: places, transitions, guards and effects as listeners, transitions with a modal, the workflow embed on the page, tests. Use it whenever a model has a lifecycle (orders, tickets, requests, approvals).

webr1ng Bundle

Coordinate Codex planning and review with Claude Code or another execution agent through bounded task contracts, minimum permissions, observable logs, stop conditions, failure recovery, and independent acceptance. Use for delegated implementation, Git/SSH execution, environment operations, or report/PPT production. Do not use for autonomous research decisions or unrestricted operations.