A Claude Agent Skill that lets Claude act as an AI reviewer/agent inside Reloops — polling assigned review work, reading comments and version history, leaving feedback, uploading new versions, and moving assets through needs_review → in_review → approved — all through the public Reloops API.
SKILL.md is the instructions Claude reads; references/api_docs.md is the full API reference it loads on demand (endpoints, payload shapes, upload flow, etc.).
- Poll the assigned review queue and pick up work (
assigned-items/requested-review) - Read asset details, version history, and comment threads before acting
- Leave comments (including timecoded/range comments on video)
- Upload new asset versions (single or multipart) and stack them onto an existing asset
- Move assets between statuses, folders, and projects
- Create and manage workspaces, projects, folders, collections, and share links
- A Reloops account with access to the workspace(s) you want Claude to work in
- Org Owner or Admin role, needed to create API keys under Team Management
Reloops issues scoped API keys per AI agent from Team Management:
- In the Reloops app, open your organization's Team Management page.
- In the AI Agents & Developer Keys card, click Add AI Agent.
- Pick Claude as the provider (or a custom name), then create the agent.
- Copy the generated
reloops_live_...secret immediately — it's only shown once. If you navigate away or reload without copying it, you'll need to create a fresh key; existing keys can only be revoked from the agent registry, not re-revealed.
Treat this key like a password — anyone with it can act as that agent in your organization. Revoke it from the same page if it's ever exposed.
Claude loads skills from a directory containing a SKILL.md. Where you put it depends on how you use Claude:
Claude Code (all projects):
git clone https://github.com/Reloops-App/claude-skill.git ~/.claude/skills/reloopsClaude Code (single project only):
git clone https://github.com/Reloops-App/claude-skill.git .claude/skills/reloopsclaude.ai: zip this folder's contents (SKILL.md and references/) and upload it under Settings → Capabilities → Skills.
Restart or reload Claude Code / start a new chat afterward so it picks up the new skill.
The skill's API calls authenticate with Authorization: Bearer $RELOOPS_API_KEY. Make the key available in whichever way fits your setup:
-
Environment variable — export it before launching Claude Code:
export RELOOPS_API_KEY="reloops_live_..."
-
Ask Claude directly — if no env var is set, tell Claude the key at the start of your session (e.g. "Use this Reloops API key: reloops_live_...") and it will use it for subsequent calls.
Never commit the raw key to a repo or paste it into a public chat.
Ask Claude something like:
List my Reloops workspaces.
Claude should call GET /functions/v1/api-workspaces with your key and return your workspace list. If it 401s, double-check the key was copied correctly and hasn't been revoked.
To try the full review loop, ask Claude to check the assigned review queue and summarize what's waiting — it should walk through the polling workflow described in SKILL.md.
- API base:
https://api.reloops.app/functions/v1 - Full endpoint reference:
references/api_docs.md