Skillget

claude-routines

agent
Security Audit
Warn
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 6 GitHub stars
Code Pass
  • Code scan — Scanned 1 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This tool lets you manage Claude Code Routines as version-controlled Markdown files instead of configuring them through a web interface. It allows you to pull, validate, and deploy routines directly via your local repository.

Security Assessment
The automated code scan found no dangerous patterns, hardcoded secrets, or requests for excessive permissions. However, because the tool is written entirely in Shell and acts as an agent, it relies heavily on executing commands and interacting with the Claude API. By design, the deploy and pull features require making external network requests to Anthropic's servers to synchronize your routines. While the setup script (`install-hooks.sh`) modifies local repository configurations, it does not appear to perform any malicious actions. Overall risk is rated as Low, assuming it is used within a trusted network environment.

Quality Assessment
The project is active, with its last code push occurring today, and it is cleanly licensed under the permissive MIT license. However, it currently has very low community visibility, sitting at only 6 GitHub stars. This means it has not been broadly peer-reviewed or battle-tested by a large audience. Developers should expect the polish and reliability typical of an early-stage, personal project rather than a community-standard framework.

Verdict
Safe to use, but review the shell scripts before executing them given the project's low community visibility.
SUMMARY

Manage Claude Code Routines as code. Edit .md files, ask Claude to deploy them. No CLI to install, no tokens to manage.

README.md

claude-routines

Manage Claude Code Routines as code. Edit .md files. Ask Claude to deploy them. No CLI, no tokens.

License: MIT
Release: v0.2.0
Built for Claude Code

demo

[!TIP]
🎬 See every feature in action → docs/features.md
Ten GIFs walking through pull, deploy, validate, diff, {{include}} snippets, bulk operations, dry-run, list, run, and orphans.

Why

Anthropic's Routines (research preview, 2026-04-14) run a saved prompt on a schedule, on GitHub events, or via API. You manage them through the web UI, the desktop app, or /schedule in the CLI.

I wanted to migrate my openclaw onto it.

None of those let you keep routines as code. No fork-and-edit. No version history. No PR review. No bulk ops. No diff between local and cloud.

This repo fills that gap. Each routine = one .md file with YAML config + prompt body. Claude reads the file and calls the management API for you.

Use /schedule for quick conversational changes. Use this when you want to version, review, share, or bulk-edit.

A routine

---
name: "Daily PR Review"
cron: "0 9 * * 1-5"          # Mon–Fri, 9am UTC
env_id: env_01ABC...
allowed_tools: [Bash, Read, Edit, Grep, WebFetch]
sources:
  - url: https://github.com/your-org/your-repo
    allow_unrestricted_git_push: false
---

Review every PR opened in the last 24 hours. For each, leave inline comments...

A session

> deploy routines/daily-pr-review.md
✓ Created trig_01ABC... — Daily PR Review (Mon–Fri 9:00 UTC)

> list
trig_01ABC...   Daily PR Review                Mon–Fri 9:00 UTC   enabled
trig_01XYZ...   Alert Triage Responder         API trigger        enabled

> diff personal/oslo-apartment-hunter.md
✓ in sync

> run trig_01ABC...
✓ Started session — https://claude.ai/code/routines/trig_01ABC...

Quickstart

git clone https://github.com/hamzafer/claude-routines my-routines
cd my-routines
./scripts/install-hooks.sh        # enables the pre-commit safety hook
claude                            # opens Claude Code in this repo

Then ask Claude:

  • pull — import existing routines from claude.ai
  • validate — lint every routine against the schema
  • deploy <file> — push a single routine
  • deploy all — bulk deploy every routine
  • diff <file> — what's changed vs. the cloud
  • run <trigger_id> — fire now

Full reference: docs/reference.md. Migrating from the web UI: docs/migration-from-web.md.

Personal vs. shareable

  • routines/ — generic templates. Committed.
  • personal/ — your real routines. Gitignored. Pre-commit hook blocks accidental staging.

Same format, same operations. Both folders work.

Caveats

  • The management API is undocumented. Only /fire is in Anthropic's docs. The endpoints we call (/v1/code/triggers) are reverse-engineered. They may change.
  • Anthropic may ship official tooling. When they do, this deprecates gracefully.
  • Env vars are not real secrets. Visible to anyone with edit access on the environment.
  • update has a security gotcha. A partial job_config silently expands allowed_tools to a 19-tool default set. We prevent it via mandatory read-modify-write. Anyone calling the API directly should be aware. Details.
  • No DELETE via API. Use the web UI.
  • GitHub triggers / API tokens are web-UI-only. This framework preserves them on round-trip but can't create or modify them.

Full risk list: docs/superpowers/specs/2026-04-26-claude-routines/09-risks.md.

License

MIT. See LICENSE.

Contributing

PRs welcome. If you find undocumented API behavior we missed, add to docs/verification/2026-04-26-routines-api-experiments.md.

Reviews (0)

No results found