How do I install the Dual Loop skill?

Run npx skillmds@latest add richfrem/dual-loop in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Dual Loop safe to use?

This skill has not completed SkillMD's automated safety review yet. Capability flags: docs only. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Dual Loop?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Dual Loop free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Dual Loop?

richfrem (@richfrem) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Dual Loop by richfrem — Agent Skill

Name: Dual Loop
Rating: 4.2 (5 reviews)
Author: richfrem

Dual Loop

(Industry standard: Sequential Agent / Agent as a Tool) Primary Use Case: Delegating a well-defined task to a worker agent, verifying its execution, and repeating if necessary. Inner/outer agent delegation pattern via strategy packets, with verification and correction loops.

richfrem aaa8bd9 9 files · 6.2 KB Updated Sep 10, 2026

File contents

Dual-Loop (Inner/Outer Agent Delegation)

The Dual-Loop architecture splits work between a strategic Outer Loop (Supervisor/Director) and an execution-focused Inner Loop (tactical coding sub-agent). The Outer Loop organizes tasks, delegates via Strategy Packets, verifies output, and generates Correction Packets.

1. Setup & Interactive Orientation (Outer Loop)

Decomposition: Break work down into atomic, testable Work Packages (WPs).
Interactive CLI & Model Selection: Ask user for preferred CLI (agy, claude, copilot, codex, llama) and model. Consult references/cheapest_models.json.
Workspace Isolation: Receive isolated git worktree or directory branch from orchestrator.

2. Strategy Packet Generation & Dispatch

Author Strategy Packet (handoffs/task_packet_NNN.md):
- Scope, target files, acceptance criteria, and explicit "NO GIT" rule for the Inner Loop.

Dispatch sub-agent with standard input redirected (< /dev/null to prevent SIGTTIN halts):

python scripts/run_agent.py handoffs/task_packet_001.md <target_file> handoffs/result.md \
  "Execute strategy packet exactly." --cli <cli> --model "<model>" < /dev/null

3. Supervised Verification Gates

Upon completion signal, the Outer Loop verifies output:

Delta Audit: Run git diff to verify only authorized paths were modified; zero stub placeholders (TODO, TBD).
Automated Test Run: Run unit and integration tests mechanically.
Outcome Resolution:
- PASS: Mark complete, update task tracker, proceed to closure.
- FAIL: Issue a Correction Packet categorized by severity:
  - 🔴 CRITICAL: Fails compile/tests. Return immediate error diagnostics.
  - 🟡 MODERATE: Breaks architecture/conventions. Cite specific rule.
  - 🟢 MINOR: Style only. Supervisor fixes directly.

4. Retrospective & Closure

Conduct post-run self-assessment survey (references/post_run_survey.md).
If friction reoccurs 3+ times, flag for triple-loop escalation.
Merge verified branch and emit handoff summary.

Operational References

Architecture Diagram: references/diagrams/dual_loop_architecture.mmd
Acceptance Criteria: references/acceptance-criteria.md
Fallback Protocol: references/fallback-tree.md

---name: dual-loopdescription: (Industry standard: Sequential Agent / Agent as a Tool) Primary Use Case: Delegating a well-defined task to a worker agent, verifying its execution, and repeating if necessary. Inner/outer agent delegation pattern via strategy packets, with verification and correction loops.---# Dual-Loop (Inner/Outer Agent Delegation)The **Dual-Loop** architecture splits work between a strategic **Outer Loop** (Supervisor/Director) and an execution-focused **Inner Loop** (tactical coding sub-agent). The Outer Loop organizes tasks, delegates via Strategy Packets, verifies output, and generates Correction Packets.---## 1. Setup & Interactive Orientation (Outer Loop)1. **Decomposition**: Break work down into atomic, testable Work Packages (WPs).2. **Interactive CLI & Model Selection**: Ask user for preferred CLI (`agy`, `claude`, `copilot`, `codex`, `llama`) and model. Consult `references/cheapest_models.json`.3. **Workspace Isolation**: Receive isolated git worktree or directory branch from orchestrator.---## 2. Strategy Packet Generation & Dispatch1. Author Strategy Packet (`handoffs/task_packet_NNN.md`): - Scope, target files, acceptance criteria, and explicit "NO GIT" rule for the Inner Loop.2. Dispatch sub-agent with standard input redirected (`< /dev/null` to prevent `SIGTTIN` halts): ```bash python scripts/run_agent.py handoffs/task_packet_001.md <target_file> handoffs/result.md \ "Execute strategy packet exactly." --cli <cli> --model "<model>" < /dev/null ```---## 3. Supervised Verification GatesUpon completion signal, the Outer Loop verifies output:1. **Delta Audit**: Run `git diff` to verify only authorized paths were modified; zero stub placeholders (`TODO`, `TBD`).2. **Automated Test Run**: Run unit and integration tests mechanically.3. **Outcome Resolution**: - **PASS**: Mark complete, update task tracker, proceed to closure. - **FAIL**: Issue a Correction Packet categorized by severity: - 🔴 **CRITICAL**: Fails compile/tests. Return immediate error diagnostics. - 🟡 **MODERATE**: Breaks architecture/conventions. Cite specific rule. - 🟢 **MINOR**: Style only. Supervisor fixes directly.---## 4. Retrospective & Closure1. Conduct post-run self-assessment survey (`references/post_run_survey.md`).2. If friction reoccurs 3+ times, flag for triple-loop escalation.3. Merge verified branch and emit handoff summary.---## Operational References* **Architecture Diagram**: [`references/diagrams/dual_loop_architecture.mmd`](references/diagrams/dual_loop_architecture.mmd)* **Acceptance Criteria**: [`references/acceptance-criteria.md`](references/acceptance-criteria.md)* **Fallback Protocol**: [`references/fallback-tree.md`](references/fallback-tree.md)

Frequently asked questions