ForLoop Agents & Skills

A collection of specialized AI agents and reusable skills that deeply integrate with the ForLoop platform — an AI-powered platform that runs on Loop Engineering: a continuous cycle of Plan → Code → Review → Deploy → Learn. Each phase tightens with every iteration as AI agents and humans work together to ship faster. These agents and skills give your AI coding assistant direct access to ForLoop sprints, stories, files, AI developer agents, and team coordination tools.

OpenCode users: For the best experience, install the ForLoop opencode plugin first — it provides native tools that are faster and more reliable than CLI commands. The plugin installer also sets up these agents and skills automatically.

What is This?

Agents are AI personas with defined roles, temperature, and permissions — they act as specialized team members like a sprint planner or story evaluator. Skills are markdown playbooks agents load on-demand to handle specific tasks correctly, from sprint planning to file management.

Together they enable your AI coding assistant to:

• Plan sprints — define goals, estimate stories, assign work across your team
• Create and track stories — generate structured stories from natural language, with status tracking
• Trigger AI development — dispatch ForLoop's AI developer agents to implement stories automatically
• Manage files — upload, download, and sync project files with S3-backed sprint storage
• Coordinate teams — manage organizations, users, permissions, and shared knowledge bases
• Everything in one platform — planning, coding, testing, and deployment all happen within ForLoop's cloud environment

These capabilities come from ForLoop's platform, exposed through either the plugin's native tools (@forLoopPlanner) or the standalone CLI (@ForLoopPlannerCLI). The agents decide which to use and load the corresponding skill automatically.

Getting a ForLoop API Token

All ForLoop operations authenticate with an API token. To get one:

1. Go to forloop.cc/profile?tab=api-tokens
2. Click Create New Token
3. Select scopes: sprint:read, sprint:write, story:read, story:write, agent:query, profile:read
4. Copy the token (it starts with floop_)

Set the token by telling your agent "set my API token" and pasting it when prompted. Tokens are stored securely at ~/.config/forloop/tokens.json with restricted permissions.

Quick Install

curl -fsSL https://raw.githubusercontent.com/forloop-cc/forloop-agents-skills/main/install.sh | bash

The installer clones this repo and symlinks everything into the right discovery paths.

Flag	Target	Agents in	Skills in
(default)	opencode	~/.config/opencode/agents/	~/.config/opencode/skills/
--claude	Claude Code	—	~/.claude/skills/
--codex	Codex	—	~/.agents/skills/
--all	All three	opencode only	all three platforms

Agents use opencode-specific frontmatter (permissions, mode, temperature). Skills are platform-agnostic markdown and work everywhere.

Manual Install

git clone https://github.com/forloop-cc/forloop-agents-skills.git ~/.config/forloop/agents-skills

# opencode
ln -sf ~/.config/forloop/agents-skills/agents/*.md ~/.config/opencode/agents/
ln -sfn ~/.config/forloop/agents-skills/skills/*/ ~/.config/opencode/skills/

# Claude Code
ln -sfn ~/.config/forloop/agents-skills/skills/*/ ~/.claude/skills/

# Codex
ln -sfn ~/.config/forloop/agents-skills/skills/*/ ~/.agents/skills/

Agents

ForLoop Planner (@forLoopPlanner)

Planning-only sprint specialist. Uses ForLoop plugin tools to gather context, confirm requirements, produce plan documents, and create actionable stories. Requires the ForLoop opencode plugin.

• Type: Primary agent (switchable with TAB)
• Temperature: 0.3 — focused and deterministic
• Best for: Sprint setup, story creation, task breakdown, sprint reviews, team coordination
• Dependency: ForLoop opencode plugin

Try saying: "Set up sprint 15 for our API redesign, starting next Monday, two weeks"

ForLoop Planner CLI (@ForLoopPlannerCLI)

A standalone agent that uses the forloop CLI binary to interact with the ForLoop platform. No opencode plugin required — just install the CLI, authenticate, and it works on any AI coding platform with bash access. Ideal for users who prefer CLI workflows or don't have the plugin installed.

• Type: Primary agent (switchable with TAB)
• Temperature: 0.3 — focused and deterministic
• Best for: Standalone CLI workflows, environments without the opencode plugin, CI/CD automation
• Dependency: @forloop-cc/forloop-cli (npm install -g @forloop-cc/forloop-cli), then forloop auth login --api-key floop_xxxxx

Try saying: "Create sprint 15 for our API redesign, starting June 15, two weeks"

Story Evaluator (@forLoopStoryEvaluator)

Subagent that evaluates task complexity and produces ready-to-create story payloads for the Planner to execute.

• Type: Subagent (invoked by Planner or via @ mention)
• Temperature: 0.2 — highly deterministic for accuracy
• Best for: Point estimation, story breakdown, complexity analysis, structuring tasks
• Permissions: Full tool access with external directory access to ~/.forloop/

Try saying: "@forLoopStoryEvaluator break down our authentication feature and estimate the stories"

Skills

Skills are organized by category. Agents load them automatically when they detect a matching trigger.

Planning (10 skills)

The core planning workflow — from sprint setup through execution.

Skill	Description
sprint-planning	Create actionable sprints with well-defined goals, capacity planning, and appropriately sized stories. Integrates with ~/.forloop/sprint-{id}/ for persistent context.
story-creation	Well-structured user stories following INVEST principles with Gherkin Given/When/Then acceptance criteria. Planning-only — does not implement features.
story-points	Systematic estimation using complexity, effort, uncertainty, and risk. Fibonacci scale: 0, 1, 2, 3, 5, 8, 10. Stories above 5 points should be split.
task-tracking	Generate actionable task lists from plan documents. Estimates points, applies templates, creates stories in ForLoop, and maintains synchronized task files.
template-based-tasks	Transform natural language into structured ForLoop stories using Basic Task and Basic Note templates. Ensures consistent metadata for AI agents and collaboration.
plan-documentation	Create and maintain sprint plan documents with user confirmation. Stored in ~/.forloop/sprint-{id}/plan/, synced to S3 for team access.
agent-auto-assignment	Automatically match stories to the right AI agent based on story type and intent. Classifies work as planning, development, or deployment.
forloop-context	Session startup — loads knowledge, plans, and tasks from ~/.forloop/ for continuity. Resolves sprint ID from manifest, env var, or git branch.
tech-stack-default	Standardized ForLoop tech stack defaults (React 18 + Vite, Lambda Node.js 20, DynamoDB, Terraform). Assumed automatically during planning — never asks for confirmation.
knowledge-management	Capture project learnings and technical decisions automatically. Syncs to S3 for team-wide access.

Administration (4 skills)

User, file, and core platform management.

Skill	Description
forloop-skill	Core ForLoop integration using plugin tools — token setup, sprint CRUD, story CRUD, AI agent queries. Used by @forLoopPlanner.
forloop-cli	Core ForLoop integration using the forloop CLI binary — authentication, sprint CRUD, story CRUD, file ops, sync, and developer triggers via bash. Used by @ForLoopPlannerCLI.
file-management	Complete file lifecycle in ForLoop sprints: presigned upload, folder organization, list, download, and delete. S3-backed with immediate upload pattern.
user-management	User profiles, storage quotas, organization creation/update/deletion, and member administration.

Quality (1 skill)

Skill	Description
verification-before-completion	No completion claims without fresh verification evidence. Applies to all planning operations — file uploads, story creation, S3 sync, sprint updates.

Sync (1 skill)

Skill	Description
aivy-documents-sync	Keep ~/.forloop/ documents synced with Sprint S3. Runs at session start and after file changes.

Requirements

ForLoop Plugin

The plugin-based planner (@forLoopPlanner) and several skills require the ForLoop opencode plugin:

• forloop-skill, sprint-planning, story-creation, story-points
• task-tracking, template-based-tasks, plan-documentation
• file-management, user-management, agent-auto-assignment
• aivy-documents-sync, forloop-context

ForLoop CLI

The CLI-based planner (@ForLoopPlannerCLI) uses the forloop-cli skill and requires the forloop CLI binary:

npm install -g @forloop-cc/forloop-cli
forloop auth login --api-key floop_xxxxx

Get your API token at forloop.cc/profile?tab=api-tokens.

Standalone (no plugin or CLI needed)

These skills work independently on any AI coding platform:

• knowledge-management — capture learnings to local files
• verification-before-completion — evidence-first completion check
• tech-stack-default — reference stack (adaptable to any platform)

Storage

Skills that persist data expect a ~/.forloop/sprint-{id}/ directory structure:

~/.forloop/
└── sprint-{id}/
    ├── plan/      ← plan-documentation, sprint-planning
    ├── task/      ← task-tracking
    ├── knowledge/ ← knowledge-management
    └── files/     ← file-management, aivy-documents-sync

How It Works

These agents and skills are directory-based configurations — they're not plugins or packages. AI platforms discover them by scanning specific directories:

Platform	Agents discovered from	Skills discovered from
opencode	.opencode/agents/*.md	.opencode/skills/<name>/SKILL.md
Claude Code	—	.claude/skills/<name>/SKILL.md
Codex	—	.agents/skills/<name>/SKILL.md

The installer creates symlinks from those discovery paths to this repo. When an AI platform starts, it scans these directories and makes agents available for selection and skills available for on-demand loading.

Agents reference skills in their system prompts, and the platform's native skill tool lets agents load a skill's full content when a task matches its triggers.

Uninstall

# Remove symlinks
rm -rf ~/.config/opencode/agents/forLoopPlanner.md \
       ~/.config/opencode/agents/forLoopStoryEvaluator.md \
       ~/.config/opencode/skills/forloop-skill \
       ~/.config/opencode/skills/sprint-planning \
       ~/.claude/skills/forloop-skill \
       ~/.agents/skills/forloop-skill
# (repeat for all symlinked skills)

# Delete the cloned repo
rm -rf ~/.config/forloop/agents-skills

Updating

cd ~/.config/forloop/agents-skills && git pull origin main

Symlinks survive updates — they point into the repo directory. Just pull and restart your AI platform.

Contributing

Skills are markdown with YAML frontmatter. Each SKILL.md needs:

---
name: my-skill
description: Clear description of when to use this skill
version: 1.0.0
category: planning | administration | quality
---

Agents are markdown with opencode frontmatter:

---
description: What the agent does
mode: primary | subagent
model: optional-model-override
temperature: 0.0-1.0
---

Add new skills by creating skills/<name>/SKILL.md. Add new agents by creating agents/<name>.md. See existing files for examples.

License

MIT — see LICENSE