README

ctrlshiftbryan-skills

Bryan's personal Claude Code plugins and skills.

Plugins

Plugin	Components	Description
`ralph`	commands · skill · agent	Install and run the Ralph Wiggum agentic loop (claude/codex) in any project — adds `once.sh`/`afk.sh`, package scripts, and prompt/state scaffolding.
`manual-chrome-review`	skill	Verify a running web app in a real Chrome session — user drives keyboard/mouse, Claude inspects DOM/network/WebSocket/server state.
`pr-review-html`	skill	Generate a single-file interactive HTML code-review artifact for a GitHub PR (collapsible diffs, severity chips, per-finding checkboxes, feedback-prompt builder).
`review-address`	skill	Reply to every PR review comment (GitHub Copilot, bots, humans) with code fixes or reasoned push-backs, committing and posting an inline reply to each.
`figma-fetcher`	agent	Auto-delegating Haiku sub-agent that offloads Figma MCP read calls out of the main conversation, caching design payloads to `tmp/figma/`. Read-only.
`pr-explainer`	command · skill	Install the AI PR-explainer GitHub Action into a repo — nag workflow + check/publish scripts, an orphan `ai-docs` branch, GitHub Pages, and a sticky 🔴/🟡/🟢 'explainer' comment + status gate linking an AI HTML walkthrough.
`post-review-as-bot`	skill	Post a code review to a GitHub PR as inline comments attributed to a GitHub App `[bot]` account — mints an installation token, validates comments against the diff, submits one atomic review.
`codex-delegation`	skills ×3	Delegate work to Codex CLI (gpt-5.5) — `codex-implementation` (scoped changes via `codex exec`), `codex-review` (independent diff review), `codex-computer-use` (browser/simulator/screenshot verification).
`questionnaire`	skill	Batch 3+ clarifying questions into a local HTML form — recommended answers pre-checked, per-question comments, and a "Copy prompt" button that serializes everything into one paste-back prompt.
`batch-grill-me-html`	skill	Conduct a manual, dependency-aware design interview in HTML rounds — each form contains the complete current decision frontier, recommended answers, optional sandboxed HTML mockups for visual choices, Plannotator injection, and a paste-back prompt.
`worktrees`	skill	Bryan's git worktree conventions (`gw`/`gwct`/`gwb`/`gwl`) — sibling `<repo>-worktrees/<branch>` layout, create from main repo on main, remote-branch checkout with tracking, merged-only cleanup that also deletes the branch.
`zoom-out`	skill	Manually zoom out one abstraction level and map the relevant modules and callers using the project's domain glossary vocabulary.

Install

Claude Code (full plugin — commands + skill + agent)

/plugin marketplace add ctrlShiftBryan/ctrlshiftbryan-skills
/plugin install ralph@ctrlshiftbryan-skills

skills.sh (skill only — natural-language trigger)

npx skills add ctrlShiftBryan/ctrlshiftbryan-skills

Layout

ctrlshiftbryan-skills/
├── .claude-plugin/
│   └── marketplace.json     ← marketplace listing
├── claude/
│   ├── CLAUDE.md            ← versioned copy of my global ~/.claude/CLAUDE.md
│   ├── statusline.sh        ← custom status bar script (setup in claude/README.md)
│   └── README.md
└── plugins/
    └── ralph/
        ├── .claude-plugin/plugin.json
        ├── commands/        ← /ralph:install, /ralph:once, /ralph:afk
        ├── skills/          ← install-ralph (natural-language trigger)
        ├── agents/          ← ralph-runner subagent
        ├── assets/          ← scripts + templates copied into target projects
        └── scripts/         ← install.sh

Adding new plugins

mkdir -p plugins/<name>/.claude-plugin && cd plugins/<name>
Write .claude-plugin/plugin.json (name, version, description, author).
Add commands/, skills/, agents/, scripts/, assets/ as needed.
Add an entry to .claude-plugin/marketplace.json under plugins[].

Public skills only. A skill that needs a specific app installed — its CLIs, its server, its repo layout — belongs in that app's own repo, not here. Example: the bm-* browser-mux skills live in the private ctrlShiftBryan/browser-mux repo under skills/ and install with npx skills add from there.

Rendered live from ctrlShiftBryan/ctrlshiftbryan-skills's GitHub README — not stored, always reflects the source repo.

13 Plugins

Name	Description	Category	Source
pr-explainer	Install the AI PR-explainer GitHub Action into a repository (manual, user-invoked setup). Drops the nag workflow + check/publish scripts, bootstraps an orphan ai-docs branch, enables GitHub Pages, and wires up the repo-specific config so every PR gets a sticky 🔴/🟡/🟢 'explainer' comment + commit-status gate linking an AI-generated HTML walkthrough published to Pages. Invoke explicitly with /pr-explainer:install or by asking to install/set up/add the pr-explainer action — it does not trigger automatically from general PR, CI, or code-review conversation.	—	./plugins/pr-explainer
ralph	Install and run the Ralph Wiggum agentic loop (claude/codex) in any project. Adds ralph/once.sh + ralph/afk.sh, package.json scripts, and the prompt/state-file scaffolding.	—	./plugins/ralph
manual-chrome-review	Verify a running web app in a real Chrome session by combining user-driven keyboard/mouse input with Claude-driven DOM, network, WebSocket, and server-side inspection. Triggers on 'manual chrome review', 'verify in browser', 'manually test feature', 'live browser session', 'walk through my app in chrome', 'review the new feature'.	—	./plugins/manual-chrome-review
pr-review-html	Generate a single-file interactive HTML code-review artifact for a GitHub PR. Fetches the diff via gh, performs an honest severity-coded self-review (critical/major/minor/nit/praise), and renders a portable artifact with collapsible per-file diffs, colored inline annotations, severity filter chips, per-finding checkboxes, and a 'Create feedback prompt' modal that aggregates checked items into a paste-ready follow-up prompt. Triggers on 'review this PR', 'build a PR review', 'review PR #N', 'code review artifact', 'interactive review', 'feedback prompt for a PR'.	—	./plugins/pr-review-html
review-address	Reply to every PR review comment — from GitHub Copilot, other review bots, and human reviewers — with code fixes or reasoned push-backs. Fetches all comments via the GitHub API, triages each one, fixes code, commits, and posts an inline reply to every comment. Triggers on 'address the review', 'reply to the comments', 'handle the PR feedback', 'address Copilot's review', 'respond to review comments'.	—	./plugins/review-address
figma-fetcher	Offload Figma MCP read calls to a Haiku-powered sub-agent to cut token cost. Auto-delegates whenever a figma.com URL appears or any Figma read tool (get_design_context, get_screenshot, get_metadata, get_variable_defs, get_code_connect_map) would be called. Fetches all five tools in one shot, caches the result at tmp/figma/{fileKey}__{nodeId}.json plus a sibling .png, and returns a small markdown digest to the main conversation. Read-only — write tools (use_figma, create_new_file, generate_figma_design) stay in the main agent. Requires the official Figma MCP server.	—	./plugins/figma-fetcher
post-review-as-bot	Post a code review to a GitHub PR as inline comments attributed to a GitHub App (a "[bot]" account) instead of the human running the command. The agent writes the review JSON; the bundled, app-agnostic script validates each comment against the PR diff (dropping unplaceable ones into a notice), composes a Copilot-style body, mints an installation token from the app's private key, and submits one atomic COMMENT review. Bot identity is supplied entirely via env vars. Triggers on 'post a review as the bot', 'post the review as a GitHub App'.	—	./plugins/post-review-as-bot
codex-delegation	Delegate work to Codex CLI (gpt-5.5) from Claude Code — three skills covering the delegation surface: codex-implementation (scoped code changes via `codex exec`, Claude reviews the diff), codex-review (independent code review of uncommitted changes, branch diffs, or commits via `codex review`), and codex-computer-use (local runtime verification needing browsers, simulators, screenshots, or app launching via `codex exec` with full access). Triggers on 'use codex', 'have gpt-5.5 do it', 'get a codex review', 'have codex verify this in the browser/simulator'.	—	./plugins/codex-delegation
questionnaire	Batch 3+ clarifying questions into a local HTML questionnaire — recommended answers pre-checked, radio/checkbox options with an 'Other' free-text, per-question comment boxes, and a 'Copy prompt' button that serializes all answers into one paste-back prompt. 1-2 questions stay in chat, one per turn; at ~3+ the skill builds the form instead of serial grilling. Triggers on 'ask me your questions', 'what do you need to know', 'make a questionnaire', or whenever several clarifying questions accumulate during planning or grilling.	—	./plugins/questionnaire
batch-grill-me-html	Run an explicit, dependency-aware design interview in HTML rounds. Maps the problem as a decision tree, researches facts instead of asking the user, opens the complete currently-unblocked frontier in one local form with recommended answers preselected and optional sandboxed HTML mockups for visual decisions, and repeats until every reachable branch is settled and the user confirms shared understanding. Manual-only; invoke with /batch-grill-me-html or ask to 'batch grill me in HTML'.	—	./plugins/batch-grill-me-html
worktrees	Bryan's git worktree workflow (his gw, gwct, gwb, gwl zsh aliases). Use whenever the task involves git worktrees in any way - creating a worktree, starting a new branch to work in parallel, checking out an existing or remote branch into a worktree, getting back to the main repo from a worktree, listing worktrees, or cleaning up merged worktrees and their branches. Trigger even on casual phrasing like 'new worktree for X', 'work on X in a worktree', 'spin up a branch for this', or 'clean up my worktrees'.	—	./plugins/worktrees
zoom-out	Manually ask the agent to zoom out one level of abstraction and map the relevant modules and callers using the project's domain glossary vocabulary.	—	./plugins/zoom-out
pre-batch-grill-me-html	Pre-generate an entire dependency-aware design interview and deliver it as one all-rounds HTML form. Walks the recommended-answer path to full depth up-front (research included), embeds the dependency graph so deviations lock their downstream questions live in the page, and loops rebuild passes until a clean pass closes on confirmed shared understanding.	—	./plugins/pre-batch-grill-me-html

ctrlshiftbryan-skills

Install to Claude Code

README

ctrlshiftbryan-skills

Plugins

Install

Claude Code (full plugin — commands + skill + agent)

skills.sh (skill only — natural-language trigger)

Layout

Adding new plugins

13 Plugins

0 Comments