machine-guard

claude-mods

Claude Code mods (function-hook plugins) I find helpful. Each folder is one plugin.

Requires a Claude Code build with function-hook plugins (2.1.289 or newer). machine-guard reads macOS tools (sysctl, memory_pressure, ioreg).

git clone https://github.com/joeldg/claude-mods ~/Projects/claude-mods

dev-servers

A pane of your project's running dev servers, so you don't have to ask Claude to restart them.

• /servers opens the Servers pane:
• running servers whose working folder is in this repo: name, port, pid and uptime, with Restart, Stop and Log
• known start commands that aren't running, with Start: package.json dev/start/serve/preview scripts (run with your lockfile's package manager), .claude/launch.json and Procfile
• a count of other listeners on the Mac
• Only button presses start or stop anything:
• Start runs the command detached, logging to ~/.claude/dev-servers/<project>/.
• Stop sends SIGTERM. If the server ignores it, pressing again within 10s force-stops it.
• Restart stops the server, waits for the port to free up, then starts it.
• Before any signal, it checks the pid still runs the same command.
• When a command fails with "address already in use", Claude gets a note (and you a toast) naming the holder, e.g. Port 4000 is held by node (pid 123, up 2h, in /Users/me/other).
• Status line: servers: :4000 :5173.
• Makes no model calls: lsof and ps every 15s.

downloads-drop

Puts files you just downloaded into your prompt with one click.

• Watches ~/Downloads (top level). When a new file arrives (PDF, Markdown, images, 3MF/STL/OBJ, zip, video…), a band appears above the prompt: New in Downloads: paper.pdf, model-b.3mf · 2m ago [Attach] [Dismiss].
• Attach puts @"/Users/you/Downloads/paper.pdf" mentions in your prompt. Dismiss hides those files.
• Waits until a file has finished downloading (skips partial downloads and files still growing), and ignores hidden and zero-byte files.
• /downloads lists the 10 newest files, numbered. /downloads attach 1 3 (or 2-4) adds those, and /downloads clear dismisses everything new.
• Stacks with other mods' bands (repo-brief, standing-orders, secret-guard) instead of hiding them.
• Makes no model calls.

Settings: folder (~/Downloads), extensions, pollSeconds (5), maxAgeMinutes (120).

effort-router

Sets effort per message, so you don't have to switch it by hand.

• Git chores ("merged", "#219 merged", "commit and push", "push it", "open a PR", "close the issue") run at low effort and come back faster.
• Deep asks (audit, review, plan, design, investigate, root cause, "why does…", "figure out") run at max.
• Everything else, including approvals like "yes", "go ahead" and "continue" and anything that starts new work ("merged 219, go ahead with #214"), keeps your session's own effort.
• Prompt cache: changing effort makes the whole conversation get cached again. So it never switches mid-turn, raises effort at once, and over a large, warm cache lowers it only after 2 routine turns in a row. In small contexts, or once the cache has lapsed, it switches right away.
• Model guard: set avoidModel (a regex such as fable) to send those requests to fallbackModel instead, subagents included.
• /route shows the last decision and the session's counts. /route off and /route on toggle it; /route deep and /route routine force the next turn.
• Status line while a turn is routed: effort: low (routine).
• Makes no model calls.

Settings: routineEffort (low), deepEffort (max), routinePattern, deepPattern, avoidModel, fallbackModel (opus), stickyTurns (2), freeSwitchTokens (30000), cacheTtlMinutes (60).

job-watch

A Jobs pane for long-running work: training runs, downloads, extractions.

• Picks up background Bash tasks and detached nohup … > log & launches by itself.
• /watch <log> [label] adds any other log file.
• Shows progress, ETA and the last log line, and flags a job as stalled when its log goes quiet.
• Shows free space on / and /Volumes/* (the NAS).
• Toasts when a job finishes or stalls. The status line shows jobs: 2 running · 1 stalled.
• /jobs opens the pane, /unwatch <label|done|all> removes jobs.
• Makes no model calls: it reads logs with tail, checks processes with ps, and runs df.

Settings (in /config): stall minutes (10), refresh seconds (10), how long finished jobs stay (120 min), auto-open (on), which disks to show.

machine-guard

Memory, swap and GPU on the status line. It refuses heavy local jobs when the Mac can't take them.

• Status line: RAM tight 12% free · swap 7.9/8G · top python 31G · GPU 87%.
• It refuses heavy jobs (training, inference, rendering, extraction, Blender, Docker, ffmpeg) when:
• macOS reports critical memory pressure, or
• the Mac is reserved with /busy.

When memory is only tight, the job runs and Claude gets a note to start one heavy job at a time.

• /busy 3h training a vision model reserves the Mac in every Claude session. /busy off lifts it. The reservation lives in ~/.claude/machine-guard.json, so a training script can write it too: ``bash echo '{"reason":"overnight training","until":'$(( ($(date +%s) + 8*3600) * 1000 ))'}' > ~/.claude/machine-guard.json ``
• /guard shows what it sees. /guard pause 15m lets heavy jobs through in this session; /guard on resumes the guard.
• Remote runs (modal run, ssh), tests (pytest) and installs are never treated as heavy.
• Add your own heavy commands with the "Also heavy" setting (a regex), e.g. overnight_|nightly_run\.sh.

mod-monitor

Watches how the other mods behave in real use, without changing them. It is listed first in CLAUDE_CODE_PLUGIN_DIRS, so the other mods' hooks run beneath it.

• Failures: any mod hook that throws, times out or rejects, read from the hook chain's results (next.trace), with the mod's name, the event and how long it ran. Slow hooks (over 1.5 s) are recorded too. The first failure of each mod in a session raises a toast.
• What each mod did: its toasts ("#219 merged → …", "Blocked: …"), status-line changes, the mod commands you used (never their arguments), and failed subprocesses (a burst of 5 in 10 minutes raises a toast). Git checks run outside a repository are logged as expected, not as failures. It also records model calls (the only usage the mods cost: /second-opinion, /recall ask) and file writes (folders only, never contents).
• /mods: a pane with one row per mod: ✓ active, ⚠ failing, ✗ not seen this session, · seen but idle. Each row shows today's counts and last activity, with Details for its recent events. It also says which mods it can't see, if any of them run above it.
• /mods report [24h|7d|30d]: a per-mod report across all sessions, also written to ~/.claude/mods/monitor/report-latest.md for a scheduled review or Claude to read.
• /mods failures [7d]: failures and failed subprocesses only.
• Logs: ~/.claude/mods/monitor/<date>/<session>.jsonl, flushed every minute and at session end, with secrets masked and old days removed after 30 days.
• Makes no model calls and adds no measurable latency.
• Error lines mods log themselves ($.ui.log with wording like "failed" or "could not"): shown in Details and in /mods failures. Three in an hour mark the mod ⚠ and raise one toast. That is how effort-router's per-request hook, which runs inside the response stream where no monitor should sit, reports a failure. It also always sends the request on unchanged.
• Transcript-row hooks (secret-guard masks /secrets records there) are watched for failures and slow runs, but not counted per run.

Settings: alerts (on), slowMs (1500), watchRender (on), watchCommands (on; off stops "mod-monitor" appearing beside other mods' command output), watchAppend (on), retentionDays (30), flushSeconds (60).

modal-meter

Keeps an eye on Modal so idle GPU containers don't burn credits.

• Status line while containers run: Modal: 1 running (2 containers). Deployed apps with no containers cost nothing, so they stay off it.
• A toast when an app has had containers up longer than alertMinutes (30), repeated at most every 30 minutes.
• /modal opens a pane of apps with state, containers and uptime. Stop asks for Confirm, then runs modal app stop. Nothing is stopped any other way.
• Shows today's spend and alerts on a budgetToday where the Modal CLI supports billing report (1.3.3+, Team/Enterprise workspaces). Otherwise /modal says why spend isn't shown.
• Finds the CLI as modal or python3 -m modal. It checks PATH first rather than running a command that can only fail, and stays silent when Modal isn't set up.
• Makes no model calls: only the Modal CLI, every 60s.

pr-autopilot

Does the "merged #219, clean up branches and start #214" round trip for you, and surfaces CI failures with their logs.

• Watches your open PRs in the session's repo: it adopts them at session start, and picks up every gh pr create Claude runs. It polls gh pr view every 60s.
• Status line: PRs: #219 ✓ · #220 CI… · #221 ✗. Toasts when CI fails (with the failing check names) or passes.
• When a PR merges, it cleans up with plain local git, then toasts the outcome and suggests carrying on (Tab to accept):
• git fetch --prune, switch to the default branch (only from the PR's own branch) and git pull --ff-only.
• Never with uncommitted changes, never --force, never other branches.
• Deletes the local branch only if it points at exactly the commit GitHub merged, so nothing local is lost. It also leaves a branch checked out in another worktree alone.
• A merge seen mid-turn is cleaned up when the turn ends, so git never races Claude.
• When you mention failing CI ("#258 is failing", "CI failed, fix it"), your message goes to Claude with gh pr checks and the tail of the failed log attached, so you don't paste it.
• /prs lists watched PRs. /prs watch <n|url> and /prs forget <n|all> add and remove them.
• Makes no model calls: only gh and git, at about one GitHub API call per open PR per minute.

Settings:

• pollSeconds (60)
• attachCiLogs (on)
• logLines (120)
• deleteRemoteBranch (off): deletes the branch on GitHub too, only while it still points at the merged commit. GitHub's own "Automatically delete head branches" setting does the same job.

It never closes issues; put "Closes #N" in PR bodies for that.

recall

Search everything you've done with coding agents, from Claude or from /recall. It replaces the broken agent-memory plugin.

• What it searches: Claude Code sessions, Codex sessions, subagent and workflow runs, Claude's memory files, standing orders, second-opinion reviews, and your /remember notes. Routine (scheduled) runs are left out unless you add routines:include to a query.
• What it keeps: prompts, answers, compaction summaries, session titles, commands, files touched, commits, PRs, issues, URLs, tasks and decisions (what you approved or ruled out). Read-only look-ups like grep and cat are kept but ranked low.
• History survives cleanup: extracts stay searchable after Claude Code deletes old transcripts.
• Claude searches it itself with four read-only tools, search, expand, recap and list, which run without permission prompts. It checks them when you say "like last time" or "what did we decide", and before asking you something you already settled.
• Commands:
• /recall <query> opens a pane of hits grouped by session. Open shows the conversation around a hit, Attach sends it with your next message, and Copy resume command copies claude --resume <id>.
• /recall last [n] recaps your last session in this repo: last asks, last answer, PRs, commits, open tasks and decisions. Send to Claude attaches it.
• /recall timeline [7d|30d|90d] [all]
• /recall decisions|commands|files|prs|commits|issues|urls|tasks|notes [query]
• /recall ask <question> answers from your history with Haiku 4.5, citing sessions. It costs a little usage and sends the matching excerpts to the model.
• /recall stats, /recall reindex, /recall forget session <id>|project <name>|before <date> (asks you to confirm), /recall help.
• /remember <fact>, /remember list, /remember forget <ref>.
• Bands:
• Once per session: Last session here (2d ago): "…" · PR #99 · 3 open tasks [Recap].
• When a prompt mentions #214, ABC-12, a file name or a quoted phrase seen in past sessions, a band offers what happened then. Nothing is sent unless you click.
• Query syntax: words must all match. OR gives alternatives, "quotes" an exact phrase, and -word excludes. Filters: project:name, kind:decision, since:7d, until:2026-09-30, source:codex, routines:include.
• Privacy:
• The index lives at ~/.claude/recall/index.db, readable only by you, and never goes in a repo.
• Secrets are masked before anything is stored: known token shapes, labelled values ("password: …"), the values of secret-named exports in ~/.zshrc, ~/.zprofile, ~/.bashrc and ~/.bash_profile, and any literal strings you list in ~/.claude/recall/redact.txt (one per line). Editing that list re-masks the existing index on the next update.
• Cost: no model calls except /recall ask. The first index takes about 2 minutes in the background, with progress on the status line. After that it updates incrementally (about 1s) at session start and every 10 minutes.
• Requires macOS's /usr/bin/python3 (Command Line Tools), whose SQLite has FTS5. Nothing else to install.

Settings: dbPath, python, sources, includeSubagents (on), includeRoutines (off), updateMinutes (10), relatedBand (on), lastSessionBand (on), maxResults (8), askModel (claude-haiku-4-5-20251001).

repo-brief

Catches Claude up on the repo when a session starts, so you don't have to ask "check the recent commits/PRs and issues".

• Gathers in the background at session start:
• branch, ahead/behind and uncommitted files
• the last 8 commits
• open PRs with CI ✓/✗/…
• issues labelled owner, todo, P0 or blocked
• stale branches (merged, or upstream gone)
• A one-line band above the prompt, e.g. main ↑1 · 3 changed · PRs #123 ✗ #124 ✓ · 2 owner issues · 2 stale branches · last commit 2h ago. Hide dismisses it.
• Claude gets the same summary once, in its first message, so the prompt cache stays warm. It refreshes after compaction.
• /brief re-gathers now and prints the full summary.
• Makes no model calls: only git and gh. The band refreshes after a turn at most every 2 minutes.

Settings: focus labels, refresh minutes, and whether to brief Claude.

routine-watch

Keeps scheduled routines (daily digests, newsletters) from silently stalling while you're away.

• Knows a session is a routine from its scheduled-task prompt, and does nothing in your other sessions.
• When a routine stops to wait for your OK on a permission prompt or an AskUserQuestion, you get a Mac notification and a toast, and the status line shows routine: daily-report · waiting on you 3m.
• When a turn ends in an error, or the routine finishes, you get a notification: Routine daily-report finished after 23m · waited on you 2 times.
• Phone push (optional): notifyCommand runs a command on the same events, e.g. curl -s -d {message} ntfy.sh/your-topic. {title} and {message} are filled in as single arguments, never through a shell.
• allowWebReads (off by default): lets routines use WebFetch and WebSearch without asking. It only replaces a prompt; your deny rules still apply, and nothing else is ever auto-allowed.
• /routine shows the routine's name, how long it has run, its waits, and the settings.
• Makes no model calls.

second-opinion

A Fable review in the background, without switching your session's model. Each run is one Fable call against your usage.

• /second-opinion: reviews recent work. On a feature branch that's the branch against the default branch; otherwise the last 12 commits, plus the diff and git status, capped at 60k characters.
• Other forms:
• /second-opinion commits 5
• /second-opinion diff (uncommitted changes)
• /second-opinion file docs/ADR-007.md
• /second-opinion <question>: adds a question for Fable to answer first.
• The command returns at once, and the status line shows second opinion: reviewing…. When the review is ready you get a toast, and a pane opens with it, ranked: wrong assumptions, bugs and risks, what's missing, what to do next.
• Send to Claude attaches the review to your next prompt (once) and drafts "What do you agree with, and what would you act on?".
• Reviews are saved in ~/.claude/second-opinions/<project>/. /second-opinion list lists them, and /second-opinion show [n] reopens one.

Settings: model (claude-fable-5-1), effort (high), maxContextChars (60000).

standing-orders

Keeps your "always / never / don't / from now on" instructions alive across compaction.

• When you write an instruction like "never open bambu with full spectrum files", a band asks: Keep as a standing order? [Project] [This session] [No]. Nothing is saved without a click.
• Project orders live in ~/.claude/standing-orders/<repo>.json and apply to every session in that repo. Session orders and your active /goal last for the session.
• Claude gets them at the start of every conversation and again after each compaction or /clear, so the prompt cache isn't disturbed. A newly saved order also rides along once with your next message.
• /orders lists them. /orders add [project|session] <text>, /orders forget <n>, /orders clear session|project, and /orders export (a Markdown block for CLAUDE.md).
• Makes no model calls.

secret-guard

Stops keys and passwords from going into a prompt, and so into your transcripts, and turns them into env vars instead.

• Catches known token shapes: AWS, GitHub, Anthropic, OpenAI, Slack, Google, Hugging Face, GitLab, npm, Stripe, private keys and bearer tokens.
• Also catches labelled values ("password: …", "api key = …", "the wifi password is …") and the two-line "Access Key ID / Secret Access Key" paste.
• Leaves alone $NAME references, placeholders, plain URLs, paths, git SHAs and ordinary prose about passwords.
• On a hit, the prompt isn't sent and goes back in the box. A band shows the secret masked (…vxrm) with a suggested name such as OPENDATALAB_SECRET_ACCESS_KEY, which you can edit:
• Save as env var appends export NAME='…' to ~/.zshrc (reusing an existing identical export) and replaces the secret in your prompt with $NAME.
• Send anyway lets exactly that text through once.
• Edit dismisses the band.
• The value is never shown in toasts, status, state or the transcript, and /secrets test <text> output is masked too.
• /secrets test <text> shows what would be caught. /secrets off and /secrets on toggle it for the session.
• Makes no model calls.

Settings: enabled (on), extraPatterns (a regex), zshrcPath (~/.zshrc).

slicer-handoff

Makes Claude's open commands hand 3D files to the right slicer.

• Full-spectrum files go to Snapmaker Orca. Bambu Studio and OrcaSlicer can't open them. A file counts as full-spectrum when:
• its name or folder matches full.?spectrum|snapmaker-only|-fs\.3mf$|-u1[-.], or
• its 3MF names a Full Spectrum filament profile.

An open -a BambuStudio … for one becomes open -b com.snapmaker.snapmaker-orca …, with the rest of the command untouched. You get a toast, and Claude gets a note so it doesn't try again.

• Earlier windows close first. Before opening a file, it asks the running slicer to quit (a normal quit, never forced), so windows don't pile up. If one won't close, for example because it's waiting on a save prompt, it stops trying and tells Claude to leave it alone.
• /slice <file> [bambu|snapmaker|orca] opens a file yourself, with the same rules.
• Recognizes open -a <app>, open -a /Applications/X.app and open -b <bundle id>, including variables set earlier in the command (S=… && open -a BambuStudio "$S/x.3mf") and files copied in the same command.
• Makes no model calls.

Settings:

• closePrevious (on): turn it off if you keep your own slicer window open, since the quit request reaches your windows too.
• fullSpectrumPattern (the regex above)
• checkContents (on)

The quit request goes out when Claude issues the command, before any permission prompt for it.

Loading

• One session from a terminal: pass --plugin-dir once per mod, e.g. claude --plugin-dir ~/Projects/claude-mods/job-watch --plugin-dir ~/Projects/claude-mods/pr-autopilot
• Every session, including the desktop app: add to ~/.claude/settings.json. Put mod-monitor first so it sees the others; CLAUDE_CODE_PLUGIN_DIR_WATCH makes desktop sessions pick up edits and show mod failures: ``json { "env": { "CLAUDE_CODE_PLUGIN_DIR_WATCH": "1", "CLAUDE_CODE_PLUGIN_DIRS": "~/Projects/claude-mods/mod-monitor:~/Projects/claude-mods/job-watch:~/Projects/claude-mods/machine-guard:~/Projects/claude-mods/repo-brief:~/Projects/claude-mods/slicer-handoff:~/Projects/claude-mods/pr-autopilot:~/Projects/claude-mods/routine-watch:~/Projects/claude-mods/modal-meter:~/Projects/claude-mods/second-opinion:~/Projects/claude-mods/downloads-drop:~/Projects/claude-mods/dev-servers:~/Projects/claude-mods/standing-orders:~/Projects/claude-mods/effort-router:~/Projects/claude-mods/secret-guard:~/Projects/claude-mods/recall" } } ``

Checking

Run with Claude Code 2.1.289 or newer; older CLIs ignore per-test settings, so a few tests fall back to defaults.

claude plugin validate job-watch && claude plugin test job-watch
claude plugin validate machine-guard && claude plugin test machine-guard
claude plugin validate repo-brief && claude plugin test repo-brief
claude plugin validate slicer-handoff && claude plugin test slicer-handoff
claude plugin validate pr-autopilot && claude plugin test pr-autopilot
claude plugin validate routine-watch && claude plugin test routine-watch
claude plugin validate modal-meter && claude plugin test modal-meter
claude plugin validate second-opinion && claude plugin test second-opinion
claude plugin validate downloads-drop && claude plugin test downloads-drop
claude plugin validate dev-servers && claude plugin test dev-servers
claude plugin validate standing-orders && claude plugin test standing-orders
claude plugin validate effort-router && claude plugin test effort-router
claude plugin validate secret-guard && claude plugin test secret-guard
claude plugin validate recall && claude plugin test recall
claude plugin validate mod-monitor && claude plugin test mod-monitor
(cd recall/engine && /usr/bin/python3 -m unittest)