Guardrails
335 mods. Block, rewrite or audit tool calls before they run.
- delegation-drift 0.1.0 Counts main-thread work since the last subagent dispatch, nudges the model to delegate at every tenth call, and refuses foreground poll loops. 29
- rollback-notice 0.1.0 When a gate denies a Bash call, tells the model the whole call was rolled back and must be re-run in full; refuses git commit -F from shared /tmp. 29
- tunnel-telemetry 0.1.0 MIDDLEWARE TUNNEL stage 4: Telemetry: records the call and its result for the ledger 29
- force-push-guard 0.1.0 Refuses git push --force before it runs 20
- lemo-guard 0.2.0 限时:Claude 单轮工作超时自动中止 · 等你确认的时间不计 · 默认关闭,在「安全」页开启 / Time limit: stops Claude when a turn runs over time · time waiting for you does not count · off by default, switch on in Safety 19
- lemo-todo 0.2.0 笔记:在面板记下笔记 · 压缩后自动摘要(Haiku)默认关闭,在「安全」页开启 / Notes: keep notes in the panel · Auto summary after compaction (Haiku) is off by default, switch on in Safety 19
- lemo-journal 0.2.0 记录:面板显示每轮统计 · 日志和自动分类默认关闭,在「安全」页开启 / Journal: per-turn stats in the panel · Log and Auto tags are off by default, switch on in Safety 19
- lemo-lot 0.2.0 抽签:给 Claude 增加抽签工具 draw_lot,抽到的签显示为签文卡片 · 默认关闭,在「安全」页开启 / Lots: adds a draw_lot tool for Claude and shows drawn lots as fortune cards · off by default, switch on in Safety 19
- lemo-assistant 0.2.0 助手:只读子 agent(Haiku),写三行周报 · 点按钮才派;「Claude 可派助手」默认关闭,在「安全」页开启 / Assistant: a read-only subagent (Haiku) that writes a three-line report · sent only when you press its button; "Claude can send the assistant" is off by default, switch on in Safety 19
- zsh-safe 0.1.0 Lets Bash commands written for bash run under zsh: unmatched globs, leading =, and a missing timeout 17
- collision-guard 1.0.0 Asks before Claude edits a file another open chat changed in the last 30 minutes: Proceed, Move to a worktree, or Cancel (/guard) 16
- hello-mod 0.1.0 Starter mod: logs each tool call and blocks Bash commands that contain rm -rf. 9
- dice-tool 0.1.0 Example: registers a tool the model can call and answers it 9
- risky-mod 0.0.1 Synthetic fixture: every red flag on purpose. Never install. 9
- toolmarks 0.1.0 KOZMOS Toolmarks: badges on tool rows. Each finished call gets a family-colored glyph and its duration at the row's end, Bash calls their exit status, failed calls a red mark, and a folded group its total time and failures. /toolmarks toggles them. 9
- ballast 0.1.0 KOZMOS Ballast: a context coach. A toast when the context window passes warnAt, then a band with an anchor and a water-level gauge at actAt: "⚓ ctx 89% — compact before the next big task" with Compact now (queued to the end of a running turn) and Not now. Optional auto-compaction between turns, a... 9
- self-command 0.1.0 For mod development: lets the model run a slash command in its own session, such as /reload-plugins after a plugin update, and read the output as the next prompt. 8
- ask-autopick 0.3.1 Picks the recommended option of a question that waited unanswered for a set time, so a session you left does not stop. Off until you turn it on. 8
- gemini-advisor 0.3.2 Gives the model a Gemini advisor tool it calls by itself: Gemini reads the whole conversation and the model's message and answers with a second opinion. Off until /gemini-advisor on; gemini-core holds the key, tier, model and thinking level. 8
- council 0.1.4 Asks a council of models about a hard problem, when the model calls its tool or you run /council <question>: Claude models, and Gemini models when gemini-core has a key, answer in parallel, and the model of the session writes one verdict from their answers. 8
- flaky-memory 0.4.1 Remembers which tests failed on which code, and tells the model when a failing test has both passed and failed on the same code, so it runs the test again instead of changing code. 8
- shot-inline 0.3.1 Draws each PNG or JPG the model saves or reads under its tool row: pixels in kitty and Ghostty, quadrant block cells in every other terminal. 8
- edit-loop 0.5.2 Adds a note to the fifth edit of one file in one turn, so the model re-reads the code path and states the root cause instead of trying again. 8
- ua-fallback 0.2.2 After a curl or wget an automated-client filter answered 403 or 429, gives the model the browser User-Agent to retry with, and the two cases where it must not. 8
- storage-guard 0.1.4 After an edit that keeps browser data in localStorage or sessionStorage, names each line, so the model stores the data in a cookie instead. 8
- tool-coach 0.1.1 Stops the model from repeating a tool call that just failed with the same input, until a file or command has changed something. 8
- probe-runner 0.1.0 For mod development: runs a live check of plugins in a fresh Claude Code session in tmux, types the steps one by one, collects the pane and the transcript, then deletes the temp directory, its transcripts and the inline store files. 8
- commitlint-preflight 0.1.0 Denies gh pr create and gh pr edit when the PR title fails the repository's commitlint config, before gh runs 8
- html-shelf 0.7.1 终端里 Claude 的回复更顺手:文件路径单击打开(HTML、文档 md/pdf/txt/json/csv/Office、图片、音视频、文件夹;脚本只在文件夹里选中不运行);代码块画成卡片,一键复制,单条 shell 命令可 Insert 到输入框;文字里 `反引号` 写的命令也补一张卡片;悬停回复右上角 Copy all;Write/Edit 工具行带 Open;/open [N|list] 打开最近的文件(Windows / macOS / Linux) 7
- screen-guard 0.1.0 Masks sensitive names, orgs and secrets in the transcript for screen sharing; /mask toggles, click a mask to reveal 6