AI & ML

caneff Bundle

Compact the current conversation into a self-contained handoff document a fresh session can pick up. Default writes it to /tmp and tells you to /clear and /pickup; a leading mode word (agent / sub / side) instead launches an independent session, a sub-agent, or a background side-agent seeded with it.

caneff Bundle

Set up a new Python repo — or convert an existing one — to my standard 2026 agent-assisted stack (uv, ruff, ty, pytest, pre-commit, GitHub Actions, justfile, AGENTS.md).

undertone0809 Bundle

Use when checking out, running, previewing, reviewing, or validating a GitHub pull request locally in a safe worktree, including PR preview URLs, local screenshots, readiness checks, logs, and cleanup instructions.

undertone0809 Bundle

Use when implementing screenshot-driven or narrow Rudder UI polish: density, alignment, spacing, labels, cards, menus, hover states, empty states, onboarding steps, redundant pages, compact workflows, screenshots, or small visible interaction fixes.

undertone0809 Bundle

Independently review Rudder changes or proposals for intent, correctness, product quality, and evidence. Use review depth from AGENTS.md; return accept, needs more evidence, or reject with actionable findings. Does not implement fixes or replace required black-box acceptance.

undertone0809 Bundle

Use when analyzing one Rudder agent run or recent run batches: run IDs, partial IDs, transcripts, run logs, execution traces, runtime failures, finalizer failures, stdout/stderr, run quality, or recent org run behavior.

undertone0809 Bundle

Use when Rudder development work needs first-principles advisor analysis plus independent reviewer rounds: proposals, UI/product decisions, architecture, release readiness, workflow changes, agent outcomes, explicit acceptance gates, repeated review, or “没有通过 review 返工”.

undertone0809 Bundle

Use when reviewing Rudder agent work, Codex sessions, PRs, commits, UI, releases, regressions, proposals, or agent outcomes for product correctness, evidence quality, scope, architecture, and handoff trust.

undertone0809 Bundle

Use when maintaining Rudder landing-page/demo screenshot workflows: seeding screenshot-ready demo orgs, capturing polished full-page app screenshots, producing screenshot manifests, or handing a seeded environment to the user for self-capture.

londopy Bundle

Find and remove AI co-author attribution from a git repo - the Co-Authored-By trailer and "Generated with" footer that Claude Code, Codex, Copilot, Cursor, Gemini and others add to commits and that put the agent in your GitHub contributor graph. Works from any agent. Use this whenever the user says Claude, Codex or an AI showed up as a contributor, asks to remove Co-Authored-By or stop adding it, asks whether commits are still being attributed to an agent, wants AI attribution scrubbed from history, wants a repo checked for AI co-authors before pushing or publishing, or asks to block pushes that carry the trailer. Also use it before you push to a repo where the user has said they want no AI attribution.

londopy Bundle

Show the merged Claude Code settings actually in effect and which file decided each key - managed policy, --settings, .claude/settings.local.json, .claude/settings.json, ~/.claude/settings.json and ~/.claude.json - and name the reasons a setting is not applying. Runs from any agent; it inspects Claude Code's files. Use this whenever the user says a Claude Code setting isn't working or isn't taking effect, asks why they are still being prompted for a tool they allowed, asks which settings file wins or where a value is coming from, asks what a settings key does or where it goes, wants their settings files checked for typos or keys in the wrong file, asks whether a project is trusted, or wants to see all their hooks, permission rules or env in one place. Also use it after you edit any settings.json yourself, to confirm the key landed where it applies.

londopy Bundle

Take a roll call of configured MCP servers - which are configured where, which will fail to connect and exactly why (missing command, dead path, unset variable, nothing listening on the port, the server's own stderr), and what each one costs in tools and context. Reads Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, VS Code, Windsurf and OpenCode config files. Use this whenever the user says an MCP server failed to connect, shows "Connection closed" or "still connecting", asks why a server or its tools are missing, asks what MCP servers they have or where one is configured, asks which servers are slow or heavy, wants MCP config checked before committing a .mcp.json, config.toml or mcp.json, asks which agent a server is configured for, or after adding or editing a server. Also use it when your own system context lists a server as failed and the user asks about it.

londopy Bundle

Take a roll call of installed skills - see which are registered, which are new, which will never register and why, and fix or load them without restarting. Works in Claude Code, Codex, Cursor, Gemini CLI, Copilot, OpenCode and any other Agent Skills host. Use this whenever the user says reload skills, refresh skills, rescan skills, re-index skills, roll call, lint my skills, audit my skills, "I just installed a skill", "I just added/edited a SKILL.md", asks whether you can see a particular skill, asks why a skill is not showing up or not triggering, asks how much context their skills cost, asks which skills another agent can see, or wants a skills repo checked before publishing. Also use it after you yourself install or write a skill mid-session, to confirm it registered.

xju2 Bundle

Create or revise prompts for AI coding agents such as Claude Code, Codex, or other software-engineering agents using OpenAI-style Harness Engineering principles. Use when the user asks for an implementation prompt, milestone prompt, issue-to-agent prompt, coding-agent task brief, or continuation prompt for an AI-driven software project. Ground the prompt in the target repository's own AGENTS.md, architecture docs, plans, ADRs, tests, and current code; emphasize outcomes and mechanically enforced invariants while preserving local agent autonomy, agent legibility, feedback loops, self-review, and improvements to the development harness.

xju2 Bundle

Bootstrap a new AI-driven or agent-first software project using Harness Engineering principles. Use when ChatGPT is asked to create, initialize, scaffold, or plan a new repository intended to be developed substantially by coding agents such as Claude Code, Codex, or similar systems. Establish repository-local knowledge, concise AGENTS.md guidance, architecture and milestone plans, deterministic setup and validation, executable acceptance criteria, mechanically enforced invariants, agent-legible observability, ADR and execution-plan conventions, and feedback loops that improve the development harness over time.

caneff Bundle

Update agent skills installed via the `npx skills` package manager without losing local edits — including following upstream renames, folder moves, and rewrites that a plain update would drop as orphans.

undertone0809 Bundle

Use when starting the current Rudder checkout as a temporary managed local preview with a stable URL, readiness check, logs, stop command, and cleanup path for manual inspection of the current branch or worktree.

undertone0809 Bundle

Use when verifying Rudder agent runtime behavior in a real local environment, especially MCP/native Rudder tools across Codex, Claude, OpenCode, Pi, or user-named runtimes. Trigger for requests like 真是/真实环境跑过吗, 排查所有 rudder tools, transcript/fallback verification, runtime MCP availability, provider matrix proof, or checking whether agents used Rudder tools instead of rudder CLI/Bash/curl fallback.

undertone0809 Bundle

Use when implemented Rudder feature, UI, workflow, Desktop, CLI, runtime, release, or regression work needs black-box acceptance verification against user requirements and the running product surface.

undertone0809 Bundle

Use when Rudder local Desktop development, Electron shell, embedded Postgres, ~/.rudder instance state, dev startup, update/install smoke, packaged boot, or release-blocking local Desktop validation fails or needs recovery.

undertone0809 Bundle

Use when reviewing a local Codex session, task, thread, or commit as a product manager or first-principles reviewer for product correctness, scope, behavior, validation, and whether the task solved the right user problem.

undertone0809 Bundle

Use when Rudder pages or product surfaces show missing, stale, sparse, empty, slow, or wrong data, including Calendar, runs, issues, dashboard counts, chat output, prod/local org data, API/UI/DB mismatches, or “这个数据从哪来”.

snapsynapse Bundle

Build a deployer-side accountability inventory of the automated and AI systems a person or team runs: one row per surface that acts on its output without a human deciding first, ending in a named person, a named fallback, and a yes-or-no on whether that person can stop it alone. The machine harvests candidates and judges their reach; it never fills a human field. Use whenever the user says "name the human", "accountability inventory", "who owns this automation", "which of my automations has nobody on it", "audit my scheduled jobs for an owner", "what runs without me", or asks who is accountable for a cron, workflow, launch agent, hook or agent. Also use when the user has a name-the-human config or inventory JSON and wants it harvested, judged, validated, rendered or diffed against a prior run.

caipe-io Skill

Audit and update all documentation moving parts for ai-platform-engineering. Checks release blog posts, features page, agent docs, homepage version strings, Docusaurus version config, and sidebar completeness. Fixes what is stale and reports what needs manual attention. Use after cutting a release, adding a new agent, or updating platform features.

rajudandigam Bundle

agent-inspect skill (website mirror)

rajudandigam Skill

Local evidence debugger and trajectory-test toolkit for TypeScript AI agents. Use when capturing framework-faithful traces, asserting TraceContract/TraceFacts, packaging Evidence v2, or inspecting local runs over read-only MCP (get_trace_facts).

undertone0809 Bundle

Control Rudder's built-in Browser through the separate run-scoped rudder-browser MCP, but invoke this skill only when the user explicitly mentions or selects Rudder's built-in Browser, Rudder Browser, or $browser. Use it then for opening or showing pages, inspecting visible or interactive state, navigating, clicking, typing, localhost testing, responsive checks, screenshots, dialogs, downloads, uploads, exports, console logs, clipboard, page assets, or listing tabs already open in Rudder's built-in Browser. Do not invoke it for generic web browsing, URL visits, web research, localhost testing, screenshots, or page interaction when the user has not explicitly chosen Rudder's built-in Browser.

undertone0809 Bundle

Expert advisor for when a build, UI, workflow, spec, or implementation feels wrong but the user cannot yet express the right product, design, engineering, or evaluation critique. Use before more implementation to turn vague dissatisfaction, weak AI-built results, traces, benchmarks, or eval evidence into a grounded first-principles scenario analysis, explicit criteria, realistic options, corner-case coverage, and a recommended next move.

undertone0809 Bundle

Use when Rudder needs performance or architecture optimization: slow pages, skeleton loops, query cache misses, refetch storms, large-org over-fetching, payload budgets, expensive API/DB paths, hot files, boundaries, measurement, or ZStudio-scale proof.

luandryperez Bundle

Genera una minuta formal de una llamada de seguimiento Vauxoo-Cliente a partir de la transcripción y la publica como Google Doc nativo (marca Vauxoo) en una carpeta de Drive indicada. Antes de generar, verifica automáticamente que el prompt local siga sincronizado con el artículo fuente de Knowledge (vauxooprod, id 160) y avisa si hay diferencias. Triggers: "crear minuta", "generar minuta", "minuta de la llamada", "acta de reunión con el cliente", "CrearMinuta". Invocación: /crear-minuta --transcript <link o ID de Google Doc con la transcripción> --folder <link o ID de carpeta de Drive> [--proyecto "Nombre del proyecto/cliente"] [--grabacion <link del video de la reunión>] /crear-minuta --check-prompt → solo compara el prompt local contra Knowledge #160, no genera minuta El link de grabación es opcional como argumento: si no se da, el skill intenta detectarlo automáticamente dentro del propio Google Doc de transcripción (las notas de Gemini/Meet suelen traer su propio link "Grabación").

caipe-io Skill

Search the internal knowledge base for runbooks, architecture documentation, ADRs, best practices, and troubleshooting guides using RAG. Use when looking for internal documentation, deployment procedures, architecture decisions, or operational runbooks.

caipe-io Skill

Run CAIPE integration tests — streaming comparison, metadata validation, and optionally full agent routing tests.

mabbit143 Skill

Self Improving Agent

verbatra Skill

Operate a verbatra i18n project through the verbatra stdio MCP server, the one a verbatra mcp process serves over stdio. Use when an MCP client is connected to a verbatra server and translation status has to be read, a key is missing or stale in a target locale, a translation has to be corrected or re-run, a glossary term has to be added, or the review queue and token usage of the last run have to be reported. Also use when a provider-spending tool appears to be absent from the tool list. Covers every registered tool, the spend boundary that keeps some of them off the default list, and what each result means.

verbatra Skill

Operate a verbatra i18n project from an open Verbatra Studio dashboard tab through its WebMCP browser tools. Use when a browser agent is driving the Studio dashboard and translation status has to be read, a key is missing or stale in a target locale, translation values have to be scanned in bulk, who last changed a locale file has to be found, or a translation has to be corrected in place. Also use when the Studio tools are absent from the tab, or a provider-spending tool is missing from the set. Covers the whole tool set, the two gates that decide which tools register, and the traps specific to driving a browser tab.

mabbit143 Bundle

Use when a Head of Ops, Knowledge Manager, or TPM-Internal needs to author, validate, or clean up company SOPs and internal runbooks (procurement intake, vendor offboarding, incident-comms cascade, employee onboarding, expense reimbursement, system-access provisioning, customer-escalation playbook) — including 5W2H completeness checks (Who-What-When-Where-Why-How-HowMuch), cross-link and orphan-page validation across a sprawling Notion/Confluence/Obsidian wiki, KB ingestion + hygiene reporting, ops onboarding doc generation, and runbook step verification (named owner, expected duration, observable success signal, rollback path, escalation contact). Pairs Kaoru Ishikawa's 5W2H method, Atul Gawande's *The Checklist Manifesto*, ISO 9001, ITIL v4 Service Operation, FDA 21 CFR Part 211, and Google SRE Workbook runbook discipline with deterministic stdlib-only Python tools that score completeness, detect anti-patterns, and emit prioritized cleanup lists. Distinct from `engineering/llm-wiki` (Karpathy-style personal