Loop Engineering
A pattern library and skill set for recurring agent loops: issue triage, dependency cleanup, PR babysitting
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- At higher autonomy levels it makes edits and opens PRs without line-by-line human review
- Regular unattended runs can noticeably increase token spend
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/loop-engineeringDetects the agents on your machine, checks the risk and pins the version.
Other ways to install
Run one by one in the Claude Code chat
/plugin marketplace add cobusgreyling/loop-engineering
/plugin install loop-engineering@loop-engineeringIn Claude Code, run /plugin marketplace add cobusgreyling/loop-engineering, then /plugin install loop-engineering. On any host you can run npx @cobusgreyling/loop init . --pattern daily-triage --tool claude, then npx @cobusgreyling/loop doctor . to verify.
Other ways from the author
/plugin marketplace add cobusgreyling/loop-engineering
/plugin install loop-engineeringThis is third-party code. Review the repository files before installing.
What it does
The project ships ready scenarios for agents that work around a repository on a recurring basis rather than once: daily issue and CI triage, PR babysitting, dependency cleanup, changelog drafting. Each pattern has a skill with instructions, a host-specific verifier subagent, and a STATE.md file that carries progress between runs. The loop CLI provides init to scaffold a pattern into a project, doctor to check the setup, and cost to estimate token spend. Work starts in report-only mode and only later moves to automatic edits after a week of observation.
Who it is for. For teams who want an agent to keep a repository healthy on a recurring basis rather than fixing one bug once.
Good fit when
- You need daily or weekly triage of issues, CI, and dependencies without constant human involvement
- You want an agent to babysit open PRs and react to changes
- You want a safe report-only start before trusting the agent with automatic edits
Not a fit when
- You need a one-off edit or refactor, not a recurring loop
- There is no way to run the agent on a schedule through GitHub Actions or similar
Example request
Set up the daily issue triage pattern in this repository and start it in report-only modeLimitations
The author explicitly warns that unattended loops make unattended mistakes and token costs can spike, so the package starts with a report-only week and requires setting up a verifier before moving to unattended edits.
How to disable. Remove the plugin via /plugin in Claude Code, or delete the pattern's files (STATE.md, LOOP.md, skills) from the repository where loop init placed them.
Security check
- At higher autonomy levels it makes edits and opens PRs without line-by-line human review
- Regular unattended runs can noticeably increase token spend
README in short
The README presents the project as a pattern library, not a rewrite-the-module button: a table lists run frequency, week-one output, and cost for patterns such as daily-triage, pr-babysitter, ci-sweeper, and dependency-sweeper. It separately covers a refactor path that turns todos into small PRs. The safety section honestly lists failure modes and anti-patterns, with community success and failure stories. MIT licensed.
FAQ
Does it only work with Claude Code?
No, the --tool flag can be set to grok, codex, or opencode, with examples for each in the examples folder.
Are edits applied right away?
No, by default the first week runs report-only; automatic edits are enabled later and gradually, through levels L1-L3.
Related
An MCP server built into the Netdata agent: metrics, logs, alerts and live process, service and container data for an AI assistant
GitHub's official MCP server: code, issues, pull requests, Actions and security alerts straight from the agent
Agent Skills for Google products
Agent Skills for Google products and technologies
Official Google skill collection for working with Google Cloud, BigQuery, GKE, ads and analytics from an agent
AWS's official MCP server suite: docs, IaC, containers, serverless, databases, cost and monitoring