
ychampion/ystack
Plugin marketplace
ystack: Claude Code mods that turn agent house rules into enforced hooks (guards, delegation, long sessions, learning, cost, owner view, design).
- GitHub stars
- 5
- plugins
- 39
- last updated
- Nov 18, 2026
Add marketplace
/plugin marketplace add ychampion/ystackRun it inside Claude Code once, then install plugins from it with /plugin.
Plugins
1assumptions-ledgerLogs every subagent's 'Assumptions I made' and hands them to the main Claude to triage; only product or business calls reach the owner, via the decision inbox with context.
2auto-claudeceptionAfter a long debugging turn or a 'turns out' discovery, toasts and pre-fills /claudeception (Tab to run). Never runs it by itself.
3brief-linterAppends the missing standing requirements (real check, long output to a file, Assumptions I made, loop cap, decision inbox) to every subagent brief.
4code-threadsTraces code back to the Claude conversation that wrote it. Each commit Claude makes is linked, on your machine only, to the prompts that asked for it and the reasoning behind it; /why <file:line> and a tool the model can call bring that back. Nothing goes into git history.
5codex-ladderHolds Codex CLI runs to an effort ladder: max only with your one-shot grant, low only via --profile docs, escalate a tier instead of retrying, and briefs need DONE, verification and a stop condition.
6context-meterKeeps long sessions cheap. /handoff writes a handoff in one turn and swaps the conversation in place for it (no /clear; background agents keep reporting). Between turns it acts by itself: hands off after a push or merge on a big context or when a session is running on its own, and before the prompt cache lapses on an idle one it compacts if work is half done or hands off if not. Speaks once at the handoff mark and near usage limits; /ctx shows size, limits and cost.
7decision-inboxA non-blocking ask tool for owner decisions from any session or subagent, one inbox pane to answer them in a batch, answers delivered back to whoever asked.
8image-themeTurns a picture you choose into your Claude theme: the picture becomes the app background, softly veiled with a frosted column behind the chat, and the theme colours are taken from it. /image-theme opens a file picker.
9iron-rulesHouse rules on shell calls, each off until you turn it on in /config: bun instead of npm, your git commit identity, no AI attribution in commits and PRs. Always blocks nested Windows shell deletes, whose quoting breaks.
10keep-goingWhen Claude stops to offer an in-scope, reversible next step it could just take ("Want me to continue?"), sends it back to finish the work. Never for decisions, risky or outward actions, or finished work. /keep-going lists nudges, /keep-going off|on.
11lanesKeeps several Claude sessions in one repo from tripping over each other: tells a session when another holds the file it just edited, warns both sides early when two worktrees change the same file, and holds back git commands that hit state every worktree shares (stash, reset --hard, clean, gc, removing a worktree in use). /lanes lists who is where.
12living-themesAnimated backdrops behind the chat in the desktop app that react to what Claude is doing (thinking, running tools, subagents, a failure, a merge) and follow your local time of day and season; /live-theme picks one.
13loop-breakerNotices when the main session or a subagent goes in circles (the same failing command, the same error, undoing and redoing an edit, a test flipping) and breaks the loop: a note, then a second opinion from the same model, then a stop with a decision for the owner.
14memory-recallWhen a tool call fails, matches the error against memory notes and skills and points the model at the matching note (once per note per session).
15model-checkupThe first time you use a model you haven't used before, suggests /doctor prompt-audit, which checks your CLAUDE.md, skills and agents for wording that works against that model.
16model-routerGives every subagent the right model and effort from a routing table you control: reviews, security and hard bugs on the strong model at high effort, builds and fixes at medium, lookups and log triage on the small fast model. Replaces banned models and keeps explicit choices.
17night-shift/queue: feeds queued tasks to the main orchestrator one at a time, unattended, with soft budgets, rate-limit pause, compaction between tasks and a morning report.
18operator-contextGives Claude the facts you keep in a short profile file: who you are, what you own, what you authorize. Added once per session, framed as facts that never override safety rules; lines that try to switch safety off are dropped and shown. /operator shows it.
19output-dietShortens long build, install and test output before it enters the context: strips progress bars, collapses repeated lines, drops passing tests when something failed, and keeps every error with its context. File contents, diffs and searches pass through untouched, and the full output is always saved to a file Claude can read. /output-diet shows what it saved.
20owner-viewFollow what agents do without reading code: a plain-English caption under every tool row, /brief-me (done, left, risky, needs you) written from a cache-read fork and shown on its own when you come back from a break, a stage strip (Plan, Build, Review, Test, Ship) with what the change still needs, and the missing step offered as the prompt suggestion.
21package-bouncerChecks every package install Claude runs (npm, bun, pnpm, yarn, pip, uv, cargo, go, gem, composer) and stops made-up names, brand-new packages, look-alike names and versions with known security holes.
22pinRules the owner gives mid-session survive compaction word for word. /pin <rule> pins one for this session, /pin --project <rule> for every session in this folder, /pin lists them, /unpin <n> removes one. Suggests pins when a prompt states a hard rule; never pins on its own.
23prompt-dietShows what every turn sends to Claude, largest first (system prompt, CLAUDE.md, reminders, tool descriptions, skill listing), and tracks which tools and skills you actually use. Lets you load rarely used tools on demand and leave out noisy reminders for a session. Nothing becomes unreachable: deferred tools stay findable and every skill stays callable. /diet restore undoes it.
24push-guardStops the git and GitHub accidents that are hard to undo, in the main session and every subagent: keys in a commit or push, making a repo public, force-pushing to a protected branch, and deleting remote branches or repos. It says why and how to go on safely.
25review-panel/review-panel: picks the specialist reviewers from the diff and has the orchestrator dispatch them on Opus in parallel, then verify findings. On demand only.
26self-learning/learn <topic>: researches a technology from its official docs, checks facts across sources, and writes a reusable skill (install, examples, pitfalls). /update-skill refreshes one; /list-skills shows them.
27session-board/sessions opens a board of every Claude Code session on this machine from the last day: folder, running or not, context size with a handoff mark, usage today, idle time, subagents and its last line. Costliest first, with a resume command to copy and a button to ask a running session to hand off.
28session-recoveryAfter a restart or crash, the first Claude session you open offers to reopen every session that was running, each as `claude --resume` in its own Windows Terminal tab (tmux on Linux). /recover, /recover auto on.
29slop-guardChecks every UI edit for AI-design tells (glows, gradient text, generic fonts, fake proof, hype copy and 12 more) and tells Claude the line and the reason. Four unmistakable tells hold the turn at Stop until fixed or explained. /slop-guard lists the session's findings.
30spec-sharpenerOn a large, open build request, has the main model ask 3-4 pointed questions before building (no side model, ~80 tokens). /sharpen on|off.
31transcript-compressCompresses old Claude transcripts in place to free disk space, losslessly: Windows and macOS use the file system's own compression, so /resume still works. Off by default: Claude suggests it at a natural break when a lot is waiting, or turn on Always on.
32twinsTells you when two copies of something drift apart, and never edits anything itself. List pairs of folders in twins.json; when a session edits one side, a band says the other copy is behind. /twins shows the changed files and a line count; only /twins sync <name> asks Claude to port the change, and /twins done or ignore clears it.
33user-voiceTies building to what real users said. On a feature request it finds matching quotes in your notes folder (simple keyword search, no model) and hands them to Claude; on a rough request Claude proposes 2-3 concrete options first and lets you pick in one question. A quiet band says when no user note mentions it. /user-voice shows the matches.
34vaultLets Claude use API keys without seeing them: /vault add NAME saves a key in the system's secure storage, Claude writes {{vault:NAME}} in commands, and the value is hidden again in everything Claude reads.
35verifyVerify gate for the main loop and every subagent: no success claim without a real check. Optional live runner of only the tests related to each edit (opt-in: it runs the project's own test scripts).
36weekly-digest/digest [days] writes a plain-English digest of what shipped across the repos your sessions touched: merged PRs and pushed commits, work in flight, decisions waiting on you, and time and usage per repo. Saved as markdown and shown in a pane; never sent anywhere unless you turn on polish.
37ystackInstalls every ystack mod and plugin for Claude Code in one go. Uninstall any single one afterwards if you don't want it.
38ystack-setup/ystack-setup lists recommended Claude Code settings (auto-compact windows, two subagent types, agent teams off, subagent model), shows which are already in place, and applies only the ones you tick when you press Apply, with backups and Undo. /ystack-setup check prints a text report. It also keeps ystack current: at each start or /reload-plugins it pulls a newer ystack, refreshes the skills and guards of every agent that has them, and updates the ystack mods (/ystack-setup update runs it now).
39ystack-stats/ystack-stats shows what ystack saves you, from your own logs: your last days of API calls re-priced under ystack's rules, and real usage per active hour before and after you installed it. '/ystack-stats share' prints a paste-ready summary with numbers only.
Details
- Repository
- github.com/ychampion/ystack
- Categories
- community
- Listed since
- Nov 18, 2026