claude

portable

The to-go bag: the working-conventions loader and its surfacing contract, the skill-library loader, the Web Tools app instrument, the tracker operator, and the rest of the portable skill set; the roster, one row per piece, is docs/portable.csv, held to disk by node/test/portable-manifest.test.mjs, so this description does not enumerate it. It carries the whole skill library, html-style included, and the agents under agents/, which are declared here one file at a time because a declared list replaces the default scan. Also carries the hooks that skills/hooks/hooks.json registers, which run on their own: the plugin refresher (SessionStart), which moves this plugin to the tip of main each session, the AskUserQuestion guard (PreToolUse), which refuses that tool in favor of a question asked in prose, the cloud turn reporter (UserPromptSubmit, PreToolUse, Stop, StopFailure and SessionEnd), which publishes bounded prompt, tool-activity and response signals to a declared sessions store, the session recorder (Stop), inert unless a checkout declares a sessions store, the session dispatcher (SessionStart), which runs every checkout's own .claude/hooks/session-*.sh whatever the project root is, the MCP failure hint (PostToolUseFailure), which turns a -32003 approval wall into the sandbox-traps connector-vs-builtin diagnosis at the moment it fires, the PR-subscribe hint (PostToolUse), the reading-column guard (PreToolUse), which refuses a Tailwind class that narrows text to a reading column, the governing-docs warner (PostToolUse), which reminds the model that documentation edits need specific permission, the program-files hint (PostToolUse), which names the program that governs an edited file, its charter and the skill that reviews a change against it, the default-skill directive (SessionStart), which tells a session the contract did not reach to invoke it, the sessions-store directive (SessionStart), which tells a session whose record store is not checked out to attach and clo

2 on GitHubby mehrlanderupdated 3 days ago
/

Plugin portable from marketplace web-tools in mehrlander/web-tools. Commands read from the repo's manifest — 3 days ago.

This plugin runs code on your machine — its hooks fire automatically on events like tool use. The exact commands are listed below; read them before installing.

What it installs

Skills (62)

  • apple-shortcuts-actions

    Compose Apple Shortcuts actions and hand them over as a tappable link that drops wired action cards onto the user's clipboard, ready to paste into any shortcut. Use when they want to build, generate, or share a shortcut or a sequence of actions, when they ask for actions they can paste, or when they

  • bake-page

    Turn a CDN-driven HTML artifact into one self-contained file: Tailwind and daisyUI compiled and pruned, Phosphor icons inlined as SVG, Alpine and other runtime JS inlined. Use when a page needs to render inline (preview box, data URL, iPhone) without CDN loads, or when the user asks to "bake" a page

  • caveman

    Rewrite a span of text, or a set of authored values such as a spreadsheet column of descriptions, so it says the same thing in fewer words. Use when the user hands over writing and asks for it caveman, or asks to tighten, compress, shorten or cut the filler from it, and when a column of descriptions

  • claude-errands

    Take and do the errands Codex or Gemini on the laptop left for Claude (claude-errand requests in web-tools-private): list them, claim the oldest one for this machine, do what its prompt asks, and close it with a report. Use when the owner says "look for errands left for Claude", "do your errands", "

  • claude-markdown

    How the Claude app displays Markdown in a chat reply. Use before showing proposed wording or any other change in a reply, laying out a table in a reply, or giving the reader text to copy, and when a reply displayed wrong on the reader's screen.

  • concept-index

    Build a repo's declared vocabulary and check a piece of writing against it: which repo files it names without a link, and which terms of art it uses as though the reader already knows them. Use when the user asks what a reply assumed, says "get out of the weeds", asks for a glossary or entity list f

  • content-registry

    Operate a repo's epistemic content registry (data/design/content.csv): classify what each artifact is (supplied, mechanical, human-, model-, or hybrid-authored) and which corpora it belongs to, scaffold draft rows, verify locators, and extract corpora. Use when the user asks to classify repo content

  • default

    The context every session runs on: the portable conventions from mehrlander/web-tools, the conversation rules (docs/CONVERSATION.md), the Qualified writing rules (docs/QUALIFIED-WRITING.md) and the doc-craft rules (skills/doc-craft/SKILL.md), shipped in the plugin. Load them at the start of any sess

  • design-sprint

    Develop a page concept in three phases adapted from Google's Design Sprint Kit: establish the substance, explore competing ways to communicate it and select one, then refine and build. Use when a page or artifact is requested with the intention and topic known but the form left open, or when the use

  • doc-audit

    Summoned when someone wants a document's wording checked sentence by sentence before approving it, to see which sentences are weak or wrong and the shortest fix for each. Not for shortening a document; use reduction-panel.

  • doc-craft

    Revise documentation so each edit serves its critical objective and keeps the whole system recognizable, navigable, and maintainable.

  • document-visual-mapping

    Map a reference document (PDF manual, rulebook, instruction packet) into a standoff register that drives an interactive reader: a split-viewer, swipe deck or step list that opens a page and highlights one phrase. Use when the user wants to walk, slice or highlight a source document in a UI, build an

  • drop-link

    Mint a link that opens GitHub's new-file form on the session's working branch with the filename prefilled, so the user can paste long content from their phone straight onto the branch without it riding through chat context. Use when the user says 'drop link', 'give me a drop link', asks for a placeh

  • excel-com

    Drive a live Excel from Windows PowerShell over COM: build a formatted sheet, read the active workbook, walk tables and pivots. Use when writing PowerShell that touches Excel, when a script must reach the running application rather than a file on disk, or when Excel COM throws something that points

  • file-retrieval

    Retrieve files from a configured corpus through a fixed-behavior tool instead of improvised grep/Read. Two commands: corpus_search.py (find hits with ranked snippets in a fixed schema) and read_doc.py (read one document whole). Use when the user says "find the files about X", "search the repo/docs/c

  • gold-set

    Build a gold set, an independently labeled sample that scores a mechanical classifier or extractor for correctness, by a measurement fan-out of blind readers, an adversarial skeptic on every disagreement, and a scoring script, then decide from the scorecard whether the program should change. Use whe

  • html-style

    The binding house style for HTML pages built here, covering what a page looks like rather than how it is built. Read it BEFORE laying out any page and BEFORE any general design, dashboard, or charting skill, which recommend what these rules forbid, such as stat cards, explanatory prose on the page,

  • in-flight

    Check whether any branch, PR, or tracker task is already working on something before starting it, and report which claims have gone stale. Use when the user asks what is in flight, what is live, whether anything is already working on a topic or a set of files, whether a branch exists for this, what

  • inspire-excellence

    Helps construct a high-ambition prompt that hands over a standard and a direction rather than a feature spec. Use when building a prompt meant to push work toward excellence: one-shotting a webpage or artifact to a high bar, handing over a blank or near-blank subject and wanting something interestin

  • ios-clipboard

    Copy and paste on iOS Safari. Use when building web apps that need clipboard access on iPhone or iPad, especially in data URL contexts (bookmarklets, paste-and-run pages) where the modern clipboard API is unavailable.

  • ios-sheet-drag

    Keep a drag interaction working on iPhone when the page is opened in a sheet, the presented card the user dismisses by dragging down. Use when a splitter, slider, resize handle, drag-to-reorder, or canvas pan closes the sheet or refuses to move on iOS, when a scrollable pane dismisses the sheet on o

  • load-skill

    Fetch a named skill by URL, from Marcus's library at mehrlander/web-tools/skills/ or another declared source, and follow it in the current session. Only for a session without the portable plugin, which already carries every skill in that library as /portable:<name>, or for a skill from another sourc

  • look-markers

    Hand over a look link: a page link whose fragment scrolls to one element, tints it, shows a one-line reason, and can ring a control to tap or run a walkthrough of such steps. Use unprompted whenever a reply would tell the reader where to look or what to tap on a page, and when asked to point at, hig

  • mobile

    Apply when the user invokes this skill explicitly. Marks the session as mobile-originated, for context and later review. Input may be transcribed. Typos and odd phrasing should be read accordingly. No other change needed.

  • notes

    Leave or read notes: short, signed, dated observations addressed to a subject (a repo, branch, file, pull request, task, snag, or another note), kept in the estate's notes store and never edited. Use when a session has an observation about a work item that belongs to no other record, when the owner

  • office-scripts-shape-ui

    Build interactive UI inside Excel using Office Scripts bound to shapes. Use when the user wants a click-to-render card, a collapsible inspector, a master-detail view, or any pattern where clicking a shape reads a cell and displays structured info. Also use when fighting Office Scripts API quirks: ce

  • outlining

    Use when asked to outline a document, lay out its structure or pull its key points; to inventory the claims a text makes, presupposed ones included, before editing or fact-checking it; to read a reply, document or transcript as the questions it answers, for a question margin or to see what a convers

  • page-design-loop

    Develop a page through inspectable object models, populated sketches and versioned revisions. Use when the user wants to generate or compare information structures and page directions, operate explicit design actions, or revisit a page's underlying model. A specified layout can be implemented direct

  • passage-reading

    Summoned when someone wants a dense passage read closely, a paragraph or a few, to see how its sentences connect: what each refers back to, introduces and claims. Not for summaries or a whole document's claims; use atomic-decomposition.

  • peeves

    The owner's pet peeves: a numbered list (peeves.csv beside this file) of things the owner has corrected an assistant for, again and again, across hundreds of sessions, in writing, chat replies, pages, code, process, git and naming. Read it before handing over prose, a page, a script or a PR body, an

  • phosphor-icons

    Selecting and using Phosphor Icons in web projects. Use this skill whenever creating HTML artifacts, web UIs, dashboards, or browser-based tools that need icons, even if icons aren't explicitly requested. Also use when the user asks "what icon should I use for X" or needs to find the right icon name

  • power-query

    Patterns, gotchas, and best practices for Power Query (M language). Use whenever the user is writing or debugging Power Query / M code, or putting queries into a workbook.

  • powershell-ise-theming

    How to share WPF theme resources (a Theme.xaml ResourceDictionary with keys like GhostButton, PillButton, MutedLabel) across PowerShell ISE add-on tools and forms without disturbing the ISE's own UI. Use this whenever the user wants their vertical ISE tools, add-ons, or UserControls to pick up their

  • programs

    Manage a declared program, which organizes work and related resources around an ongoing concern such as surfacing. Reports what changed in its governed files, what its measures show, its open tasks and the snags on its files, then works a proposed change through its reviewers to the owner. Use when

  • reduction-panel

    Run a fan-out reduction panel over a document: one cheap reader per section proposing telegraphic and tightened versions plus a cut list, one stronger judge over the whole file and every pass, then a restore round guarded against the author's own defensiveness. Use when a document should get shorter

  • reorient-me

    Summarize where a long conversation has gotten to, formatted for re-entry rather than record-keeping. Use when someone asks to be caught up or reoriented inside a thread they have lost track of: reorient me, where are we, where did we land, catch me up, recap where we have gotten, I have lost the th

  • repo-review

    Read a repo's own state and report what stands out, at one of three depths, named by argument: rounds (a quick read of what is moving, stalled, or emerging), SOR (state of the repo: an on-demand audit of every layer with ranked findings), or sweep (a parallel fan-out where each agent reads a differe

  • report-distiller

    Distill research documents, reports, white papers, and academic papers into a structured summary following a specific analytical framework. Use this skill whenever a user shares a document (PDF, paste, or link) and asks for a summary, distillation, overview, analysis, or review, especially for resea

  • review-panel

    Convene a general review panel over a document, page, plan or change: five reader seats defined by how they search (conservator, radical editor, source historian, owner proxy, counterexample hunter), each blind to the others, then a closer who verifies and weighs their findings and states the decisi

  • roles-sheet

    Write or revise a roles sheet, a one-page document that divides one process's work among the roles that do it, as a card per role with a row per kind of work. Use when asked for a roles and responsibilities sheet or a who-does-what page, or to add, split or reassign a role on one.

  • rundown

    Read the owner's open Claude Code sessions and write a rundown: each session's state and a message the owner can paste to move it. Run when the owner asks for a rundown.

  • sandbox-traps

    Diagnose a Claude Code web sandbox failure that impersonates a different problem. Use when a tool call returns MCP error -32003 or unexpectedly requires approval, when git log looks truncated or history appears rewritten, when a check reports that a pinned commit is missing from the checkout, when g

  • scour

    Fan out agents to acquire source material that does not exist locally, mapping who says what about a subject across the open web. Use when the user wants to survey a field, find the actors or institutions in some arena, gather primary sources at breadth, build a corpus to delve into later, or says \

  • screenshot-review

    Review a rendered screenshot before handing it over or drawing a conclusion from it, using a self-read against a failure checklist and a fresh-eyes reader (a context-free Haiku subagent) run blind first, then informed. Use when a headless screenshot is about to be sent into chat, when a shot is the

  • session-review

    Review a Web Tools development session and its repository or PR context. Use when the user asks for $session, gives a bare claude/* branch or a screenshot of a Claude Code session, or asks to check back on a previously reviewed one.

  • sessions

    Log this session. Attach and clone the session archive (the checkout whose .web-tools.json declares `sessions`) beside the project root when nothing has, so the portable plugin's Stop hook can find it. Use when the session-start directive says to, when the user asks whether this session is being rec

  • shape-text

    A reference on using structure in text to convey information. Consult it when working structure into or out of a document, chat answer, or HTML page, alongside the stated aim, for example surfacing structure latent in flat prose, calming an over-formatted draft, or auditing whether existing structur

  • shortcut-links

    Give the user a tappable link to install or run an iPhone shortcut, copy actions, open Settings, or run a diagnostic that reports its result. For building shortcut actions, see apple-shortcuts-actions.

  • show-diff

    Summoned when proposing a change to an existing document the user will review, a skill file, a piece of prose, a config, and you want the edit shown rather than described in words or dumped as a block to paste into a separate tool. Also when the user asks to see an edit as a diff or a tracked change

  • skill-prefs

    Standing preferences for how a skill should be written. Use whenever creating, drafting, editing, or improving a skill, alongside the stock skill-creator rather than instead of it. Read this before writing the SKILL.md. It governs register, scope, output, and trigger strategy, the places where skill

  • source-anchored-writing

    Generates documents where every factual claim is visibly anchored to a specific source, or explicitly flagged as unsupported, in the manner of a heavily contested encyclopedia entry. Fires only on an explicit request such as "include sources," "add footnotes," "cite where this came from," "I need to

  • source-anchored-xlsx

    Build Excel workbooks where every typed number sits next to an image of its source, laid out the same way, and formatted tables draw from those numbers by formula. Use when figures are transcribed, combined or reconciled from a PDF, report, slide deck or image into a spreadsheet that must be checkab

  • source-anchoring

    Doctrine for output that does not ask to be believed: every element traces back to a trusted source, or is flagged as not. Parent of source-anchored-writing (prose) and source-anchored-xlsx (workbooks); route to those for mechanics and apply this directly to anything else (deck, dashboard, generated

  • succinct-js-idioms

    Succinct JavaScript idiom preferences for code generation. Use this skill whenever generating, reviewing, or refactoring JavaScript, including browser scripts, artifacts, and Node 22+ targets. Nudges toward concise, well-supported APIs and platform primitives.

  • succinct-text

    Revise expository writing toward succinctness by establishing topic, altitude, and sequence, condensing sentences, and proving losslessness with a coverage crosswalk.

  • tasks

    Operate a repo's project tracker: propose and file a task, claim one, update or close it, assess the tracker as a whole, refine the backlog, and regenerate the board, following the docs/TRACKER.md schema and the rule that task files and board.md commit straight to main (not a feature branch). Carrie

  • tend

    Cultivate a workspace toward recognized objectives: investigate branches, pull requests, trackers and snags, store each conclusion as a finding, and bring the owner a prepared user call only where the next step needs their word. Run when the owner invokes /tend, or when a session receives a /tend pr

  • tree

    Render a repository file tree in chat as a linked markdown table: folders and files as tappable GitHub blob/tree links, depth carried by an inline code-span connector prefix (the classic box art) or by invisible braille indentation, filetype icons as anchors, an optional gloss column. Table-first be

  • userscripts

    Maintain, test, and ship Safari userscripts and bookmarklets in Web Tools. Use when editing userscripts/lib/launcher.js, updating userscript stubs or bookmarklets, diagnosing GM storage or loader cache updates, configuring repository capture intake, or verifying userscript builds.

  • wayback

    Read the Internet Archive's Wayback Machine for pages a site no longer serves or never listed, by enumerating its captures under an address and reading them. Use when the user wants past versions of a page, postings or documents a site has taken down, everything the archive holds under a path, or sa

  • web-tools-app

    Use the Web Tools app, the hosted page that browses any GitHub repo and moves files between repos: mint a browse link, mint a stage link (the 🗂️ #stage= fileset transport), author a repo's .web-tools.json manifest (landing, pins, stage.files, stage.targets), or run a cross-repo file transfer. Loads

  • windows-powershell

    Windows PowerShell 5.1 (Desktop Edition) code generation with strict version constraints. Use when writing PowerShell for Windows environments, especially government/enterprise where 7.x is unavailable. Enforces 5.1-compatible syntax, prevents use of ternary operators, null coalescing, pipeline chai

Subagents (9)

  • closer

    The review panel's last seat: weighs every other seat's findings without having authored any, separates convergent findings from single-seat ones, rejects what does not hold, and states the decision the owner faces. Summon after the panel's readers have returned.

  • conservator

    Review-panel seat that assumes existing details may encode hidden requirements. Summon on a proposed change, rewrite or deletion to find what it would lose: constraints, edge cases and decisions the new version drops without saying so.

  • counterexample-hunter

    Review-panel seat that tries to falsify the subject's claims: finds the case, input or file where a stated rule, finding or guarantee does not hold. Summon on an analysis, a convention, a test's claim of coverage, or any 'always' or 'every'.

  • house-style-critic

    Reviews one HTML page against the HTML house style (the html-style skill), rule by rule, from a context that did not build the page. Summon after building or changing a page and before handing it over, with the page's source path and, where one exists, a screenshot at desktop and at phone width. Ret

  • novice-reader

    Review-panel seat that reads without insider context: a capable reader new to this repository and this conversation. Summon to find where a document, page or reply loses a reader: undefined terms, references to things never introduced, steps that assume knowledge.

  • owner-proxy

    Gives the owner's likely view of a message, a document or a decision, drawn from what the owner has said and recorded. Summon before handing the owner substantive work, or to get the owner's perspective on content or a decision before asking the owner. Does not edit.

  • peeves-critic

    Reviews a piece of work (a reply, a document, a page, a commit message, a PR body) against the owner's pet peeves, the numbered list in the peeves skill, and reports each one it trips with the peeve's id and the exact place. Summon before handing over prose or a page, or when the owner asks whether

  • radical-editor

    Review-panel seat that searches for large simplifications: whole sections, steps or files that could go, structures that could merge, a shorter shape for the same job. Summon when a document, page or design feels heavier than its job.

  • source-historian

    Review-panel seat that tracks provenance and precedent: where a claim, rule or design came from, what earlier decisions it repeats or contradicts, and whether the record still supports it. Summon when a subject cites history, restates a convention, or changes something with a past.

Hooks (23)

  • SessionStart

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/refresh-plugin.sh"

  • SessionStart

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/session-dispatch.sh"

  • SessionStart

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/invoke-default.sh"

  • SessionStart

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/invoke-sessions.sh"

  • SessionStart

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/environment-report.sh"

  • SessionStart

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/account-skills.py" --hook

  • PreToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/reading-column-guard.sh"

  • PreToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/send-later-guard.sh"

  • PreToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/ask-question-guard.sh"

  • PreToolUse

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session-activity.py"

  • PostToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/pr-subscribe-hint.sh"

  • PostToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/wake-quiet-hint.sh"

  • PostToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/warn-governing-docs.sh"

  • PostToolUse

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/program-files.sh"

  • PostToolUseFailure

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/mcp-fail-hint.sh"

  • Stop

    bash "${CLAUDE_PLUGIN_ROOT}/hooks/session-record.sh"

  • Stop

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/reply-check.py"

  • Stop

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session-activity.py"

  • UserPromptSubmit

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session-activity.py"

  • UserPromptSubmit

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/default-files-hint.py"

  • UserPromptSubmit

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/context-size-hint.py"

  • StopFailure

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session-activity.py"

  • SessionEnd

    python3 "${CLAUDE_PLUGIN_ROOT}/hooks/session-activity.py"

View source on GitHub

More from mehrlander/web-tools

All 2 plugins →

h

claude

The house policies as slash commands: /h:<shortcut> shows the entry of that name. Every skill is user-only and generated from docs/house-policies.md by node/build/policy-skills.mjs.

18 skills

2

Similar plugins

Token usage tracking, model cost attribution per agent, budget alerts, and optimization recommendations

24 skills · 1 command · 1 agent · hooks

ruvnet/ruflo 81,589

container

claude

Teaches Claude container's command surface and how it maps to Docker, Lima, Colima, and Podman on macOS

1 skill

apple/container 51,329

Build, train, and deploy AI models with deep AWS AI/ML expertise brought directly into your coding assistants, covering the surface area of Amazon SageMaker AI.

19 skills

anthropics/claude-plugins-official 47,076

HADS (Human-AI Document Standard) — semantic tagging convention for docs that work for both humans and AI models — plus the grounded-vault pattern for provenance-linked, drift-checked knowledge stores.

2 skills

wshobson/agents 42,461

An authenticated hosted MCP server that accesses your PlanetScale organizations, databases, branches, schema, and Insights data. Query against your data, surface slow queries, and get organizational and account information.

19 skills · 1 MCP server

anthropics/knowledge-work-plugins 35,601

diffusers

claudecodex

Conventions and task skills for contributing to diffusers

4 skills

huggingface/diffusers 34,915

Maintain portable?

This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.

claude & codex plugins2 plugins