Contextual instructions that teach AI agents how to use Loop effectively in your project.

Agent Skill

The Loop agent skill (@dork-labs/loop) is a content-only npm package that teaches AI agents how to use Loop's REST API. It contains no executable code -- just structured instructions that agents read for context when working in your project.

The agent skill provides knowledge (instructions the agent reads), while the MCP Server provides tools (actions the agent can call). Install both for the best agent experience.

What's Inside

The package ships four artifacts:

File	Purpose
`SKILL.md`	Primary skill definition in agentskills.io format. Works with Claude Code, Codex, Cursor, Amp, Goose, and any platform that loads SKILL.md files.
`templates/AGENTS.md`	Ready-to-paste snippet for your repository's AGENTS.md file (Linux Foundation standard).
`templates/loop.mdc`	Cursor rule file for `.cursor/rules/`. Applied on request, not auto-applied.
`templates/openhands-loop.md`	OpenHands microagent with keyword triggers (`loop`, `loop api`, `ingest signal`, etc.).
`references/api.md`	Full API endpoint reference loaded on demand by agents needing technical details.

Installation

The easiest way to install the skill across any agent platform that supports openskills:

npx openskills install @dork-labs/loop

This downloads the package, copies SKILL.md to your agent's skill directory, and registers the skill for discovery.

Install the package as a dependency:

npm install @dork-labs/loop

Then reference the files in node_modules/@dork-labs/loop/ when configuring your agent.

What Agents Learn

When an agent reads the skill, it learns:

Authentication -- how to use Bearer token auth with $LOOP_API_KEY
The dispatch loop -- the core workflow of get task, do work, report completion, create discovered issues, repeat
Common API operations -- create issues, list issues, ingest signals, complete tasks, add comments
Issue types -- signal, hypothesis, plan, task, monitor
Status values -- triage, backlog, todo, in_progress, done, canceled
Error handling -- what 401, 404, and 422 responses mean and how to recover

The Dispatch Loop

The skill teaches agents the core Loop workflow:

Get next task  -->  Do the work  -->  Report completion  -->  Create discovered issues  -->  Repeat

Get next task -- fetch the highest-priority unblocked issue from Loop
Do the work -- execute what the issue describes
Report completion -- update the issue status to done and add outcome notes as a comment
Create discovered issues -- if bugs or new tasks are found during work, create them as new issues
Repeat -- get the next task and continue the loop

Platform-Specific Setup

Claude Code

Claude Code loads SKILL.md files automatically. After installing via openskills or npm, the skill is available in every session.

npx openskills install @dork-labs/loop

Set your API key so the agent can make calls:

export LOOP_API_KEY=loop_...

Cursor

Copy the Cursor rule file into your project:

mkdir -p .cursor/rules
cp node_modules/@dork-labs/loop/templates/loop.mdc .cursor/rules/loop.mdc

The rule is configured as on-request (not auto-applied), so the agent loads it when Loop-related context is needed.

OpenHands

Copy the microagent into your project:

mkdir -p .openhands/microagents
cp node_modules/@dork-labs/loop/templates/openhands-loop.md .openhands/microagents/loop.md

The microagent triggers automatically when the agent encounters keywords like loop, looped.me, loop api, loop issue, or ingest signal.

AGENTS.md

To add Loop context to any agent that reads AGENTS.md (Linux Foundation standard):

cat node_modules/@dork-labs/loop/templates/AGENTS.md >> AGENTS.md

This appends the Loop integration section to your existing AGENTS.md file.

Common Operations Reference

The skill teaches agents these key API calls:

Get the next issue to work on

curl -H "Authorization: Bearer $LOOP_API_KEY" \
  "https://app.looped.me/api/issues?status=open&limit=1"

Create an issue

curl -X POST -H "Authorization: Bearer $LOOP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"title":"Fix login redirect","type":"bug"}' \
  https://app.looped.me/api/issues

Ingest a signal

curl -X POST -H "Authorization: Bearer $LOOP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"source":"agent","title":"Error rate spike","data":{"count":47}}' \
  https://app.looped.me/api/signals

Complete an issue

curl -X PATCH -H "Authorization: Bearer $LOOP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"status":"done"}' \
  https://app.looped.me/api/issues/{id}

Add a progress comment

curl -X POST -H "Authorization: Bearer $LOOP_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"body":"Investigating root cause...","authorName":"agent"}' \
  https://app.looped.me/api/issues/{id}/comments

Next Steps

MCP Server -- Zero-code IDE integration that gives agents callable tools alongside the skill's knowledge.
Agent Integration -- Per-agent guides for Claude Code, Cursor, Windsurf, OpenHands, and custom setups.
API Reference -- Full endpoint documentation for all Loop REST API routes.
Concepts -- Understand how issues, signals, dispatch, and prompts work together.

Agent Skill

On this page