- Home
- Skills
- Agents & Orchestration
- longrun universal
Works with the AI Agents you already use
longrun universal
A harness-neutral playbook for multi-session, unattended agentic development with durable state and verification gates.
Free
longrun universal
Example session with this skill installed
Start a longrun for the new authentication module. I have a 50 dollar budget and want to use a feature branch. Use the universal playbook.
- Read your context and instructions
- Compiled the longrun universal
- Generated the document
I have initialized the LONGRUN development environment. A new feature branch feat/authentication-module has been created, and the durable state files are now tracking the run intent and initial task backlog.
LONGRUN Initialization Report
| Component | Status | Details |
|---|---|---|
| Playbook | Active | references/LONGRUN-universal.md |
| Branch | Created | feat/authentication-module |
| State Files | Initialized | TASKS.md, FEATURES.json, LOG.md |
| Budget Gate | Set | $50.00 USD |
| Run Intent | Logged | Implement core auth module with JWT support and bcrypt hashing. |
Initial Task Backlog
- Define User schema and migration scripts.
- Implement password hashing utility with bcrypt.
- Create JWT generation and validation service.
- Build login endpoint and integration tests.
Next steps
- Review the
FEATURES.jsonfile to ensure the acceptance criteria match your requirements. - The agent will now begin the first task in the queue.
- To run this unattended, execute the driver script:
./scripts/longrun-driver.sh -n 10 -- [your-agent-command]
longrun-universal.pdf
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Long-running agentic tasks often fail when context windows fill up, sessions time out, or the agent loses track of the original intent during unattended runs. This skill implements the LONGRUN playbook to enable multi-session, unsupervised development that survives context loss and hardware reboots.
What it does
- State persistence manages project progress through
TASKS.md,FEATURES.json, andLOG.mdfiles that serve as the ground truth across sessions. - Evidence-based verification enforces a strict loop where no task is marked done without verifiable command output or test results.
- Automated driver integration supports headless execution via a driver script to run back-to-back sessions until the backlog is cleared.
- Reversibility management ensures all work happens on feature branches with frequent commits to prevent destructive state loss.
- Stop-rule enforcement monitors for loop exhaustion or blocking issues, halting the agent to wait for human intervention rather than burning tokens.
How it works
- Start by initializing the run intent, creating a feature branch, and scaffolding the state files from provided templates.
- Resume at the start of any new session by reading the on-disk state and reconciling the current git head with
TASKS.md. - Execute the work loop by picking the next task, applying changes, and capturing evidence of success.
- End by writing a session retrospective and updating the durable state files before the process exits.
Frameworks & tools
This tool is harness-agnostic, requiring only a shell, git, and file system access. It includes a longrun-driver.sh script for Unix-like environments (macOS/Linux) to manage headless agent cycles.
Why this beats prompting it yourself
Standard prompts forget the "why" and "how" after a few thousand tokens of terminal output. This skill forces the agent to treat its own instructions as a persistent protocol, ensuring it doesn't skip tests or hallucinate progress when you aren't watching.
Use cases
- Executing a large backlog of refactoring tasks overnight.
- Managing complex feature migrations that span multiple days of development.
- Maintaining project continuity when switching between different LLM providers or agent interfaces.
- Running autonomous agents in headless environments with strict verification gates.
Known limitations
Does not support concurrent background jobs in headless mode. Requires an agent harness that allows automated git commits and file writes to maintain the state loop.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
No support for concurrent background jobs in headless mode. Requires agent permission to commit to git and write files without interactive prompts to function unattended.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 11 days ago
- Free with an Agensi account
- Software
Git