Level 5a Claude Code plugin

Automation observability for your repositories.

Level 0: you watch every dashboard. Level 5: the repo's automations watch themselves, and you hear only what is broken. Which routines ran, which went silent, what each one cost.

>/plugin install level5 --marketplace humanagent/level5
Start free
Plugin forClaude CodeCodex soon

FreeOpen source (MIT)

> works with the tools you already use

01·claude code plugin

Plugin

It lives in Claude Code

Level 5 is a Claude Code plugin, open source, installed in one line. It adds four things to your sessions, and writes the one hook your routines need into the repo, since a routine runs on a fresh clone.

>/plugin install level5 --marketplace humanagent/level5

/level5:setup skill

Reads the repo, proposes monitors, a daily report and a pulse, and opens one PR. After the merge it creates the routines.

> /level5:setup ⏺ Opened PR #142 · 5 skills, 3 workflows ✓ Factory ready · 5 monitors watching acme/api

Above the prompt mod

What needs a look, where you already are. Silent when all is clear. /level5 says the rest.

Level 5 level5-report missed its last run /level5 Hide >

/level5:sweep workflow

Runs every monitor now. A finding becomes an issue only after two skeptics fail to refute it.

> /level5:sweep ⏺ 3 monitors · 4 findings · 3 refuted ✓ Filed #311 · heartbeats sent

/level5:mend workflow

The fixer. A test that fails first, the smallest fix, one PR per issue. It never merges.

> /level5:mend ⏺ #311 → PR #312 · 1 test, 4 lines ✓ Fails without the fix, passes with it

Also /level5:new-monitor, a monitor for any source from one line. Every file is in plugin/ on GitHub.

02·monitors

Status

The monitor of the monitors

A dead monitor and a healthy system both look like silence. Every routine leaves a heartbeat when it ends, and Level 5 checks them against your roster.

Start free

One page for the whole factory

Every skill with 30 days of history. It flags the ones that died and the ones that said ok after reading nothing.

acme/apiupdated 2 min ago
errorsdailyOperational
30 days ago100% oktoday
logsdailyOperational
30 days ago100% oktoday
costdailyOperational
30 days ago100% oktoday
llm-tracesdailyOperational
30 days ago96.7% oktoday
speeddailyChronically degraded
30 days ago36.7% oktoday

Automations, live

What is running right now, one click from the top of every page.

acme/apiActions3
Actions3 running · checked just now
  • /fix-issuesFix #4127: retry on 429Run regression test 6/93m 12s
  • ci.ymlPR #4131Unit tests 3/51m 14s
  • /pr-reviewReview #4129Read the diff 2/641s

Monitors

Each one reports when it ends. Two weeks at a glance.

errorsok
100%
logsok
100%
llm-tracesok
96.7%
speeddegraded
36.7%

Code your agents merged

Built for repos where agents ship more code than any team could review by hand.

2.27Mlines merged in 30 days+1.82M −446k
3,218PRs merged91%by agents14,602CI runs38 minmedian to merge
30 days agolines added and deleted per daytoday

Reports

One day of the factory in one email, every morning.

Daily factory report
Needs a look

2 things to look at

Sunday, 11 October · acme/api

Automations
41runs1 missed18.2Mtokens$38.40token cost
  • /growth-report did not run0 9 * * 1
  • speed: degraded4 of 7 days
  • /fix-issues3 PRs opened
Code
112PRs merged104 by agents+64k −17klines
example with invented data

Connect GitHub 01

Install the app on your account or your organization. Pick one repo, a few, or all of them, and change it later.

Your routines report 02

One command installs a hook: every Claude Code routine reports itself when it ends.

You read one page 03

Next run, tokens and state for every skill. It refreshes on its own.

03·tokens and costs

Costs

What the factory burns

Tokens for every routine run, next to what your cloud bills each day. A routine that burns 4× its usual tokens shows up the same day, not on the invoice.

$616last 30 days
Claude Code $233AWS $272Vercel $76Neon $36
Per routine30 days
RoutineRunsTokensCost
/fix-issuesdaily31212M$118
/monitordaily3096M$44
/pr-reviewdaily3071M$37
/growth-reportweekly49M$6

Price rule: /fix-issues over $6 a day, fired 3 h ago

example with invented data
04·in 3d

Factory

Your software factory, in 3D

Every component is a hall. Routines are the robot arms, agents are the workers, and every PR is a crate that rides the belt through the CI gates into main. Monitors are the light towers: the failing one flashes red.

112PRs merged today
ComponentsRoutinesCI gatesMonitorsAgents
Drag to turn it
example with invented data
05·read-only

Sources

Where it reads from

Read-only on your code. No cloud or billing credentials: your jobs send what they already know.

Integrations

The four platforms first. Anything else through one curl.

GitHubRepos, pull requests, issues, checks and every Actions workflow, live, through a read-only GitHub App.
awsAWSDaily spend by service from Cost Explorer, the cost explorer and the city map. Sent by your own cost job.
Fly.ioDaily spend next to everything else, and heartbeats from the machines and jobs you run there.
VercelDaily spend per project, and heartbeats from your deploys and crons.

Also: Claude Code routines, skills and tokens per session, OpenRouter for agents that run on its keys, Neon and the Anthropic API.

What it reads from each repo

It never writes to your repo, and every line that reads it is on GitHub.

.claude/skillsThe skills in your factory, and which ones run on a routine
.claude/factory.jsonYour monitor roster and Claude Code routines, with their cron
.github/workflowsScheduled Actions and the last run of every workflow
Issues, PRs and checksThe open queue, CI on every PR, and what your agents merged
06·start

Start in a minute

Connect GitHub and let Claude Code build the factory it will watch.

FreeOpen source (MIT)

  1. 1

    Connect GitHub

    Install the GitHub App on your account or organization and pick one repo, a few or all of them. It only reads.

    Start freeor sign in
  2. 2

    Add the plugin to Claude Code

    >/plugin install level5 --marketplace humanagent/level5

    Once per machine. It brings /level5:setup, what needs a look above the prompt, and /level5:sweep and /level5:mend for a sweep or a fix on demand.

  3. 3

    Let it build the factory

    >/level5:setup

    At the repo root. It gets the repo's key with your GitHub login, writes the heartbeat hook into .claude/, proposes monitors, a daily report and a pulse, and opens one PR with new level5-* skills and workflows. After the merge it creates the Claude Code routines and runs each once.