Guide · Agent skill
How to use the acc Solve Loop agent skill (accint-solve)
Updated · written with AI from the project's own documentation and checked against our catalogue data
accint-solve is an agent skill: a SKILL.md file that tells an AI agent how to send a goal to an external service called acc (accreted intelligence) through MCP tool calls, using a mode named solve. This guide explains what the skill makes the agent do, how to install it, how to trigger it, and what it needs and risks, so you can decide whether to use it. After reading, you will be able to install the skill, make a request that triggers it, and handle both kinds of result it can return.
acc Solve Loopsickn33/agentic-awesome-skills · MIT Download the skill Quick start
- 1Confirm your acc instance. Make sure an acc instance is configured and that your agent can reach its acc_act, continue and outcome MCP tools, and that you have submit tokens and frame ids from acc.
- 2Download the ZIP. Download the accint-solve ZIP from this page.
- 3Unzip into a skills folder. Unzip it into ~/.claude/skills/ for a personal install, or into your project's .claude/skills/ folder for a project install.
- 4Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 5Send a request that names the skill. Write a request that starts with @accint-solve and describes your goal in specific terms.
- 6Handle the result. If acc returns a final answer, report it with the commitment id and cited ids. If it returns a brain_frame, deliberate, then submit a proposal ending with a PREDICT line.
- 7Close the commitment. Later, close the commitment honestly using acc_act with the outcome runtime.
What does the accint-solve skill make an agent do?
The skill routes a goal through acc's scored-memory loop by calling acc_act with runtime set to solve. The skill itself contains no logic; it only describes the two MCP verbs the agent should use.
acc can answer in one of two ways. If the result is final, the agent reports the answer, the commitment id, and the cited ids. If the result is a brain_frame, the agent has to deliberate over it itself, write a proposal, and submit it back.
- Final result: surface the answer, the commitment id, and the cited [ids].
- brain_frame result: reason over it, then submit it with the continue runtime.
- Later, close the commitment honestly with the outcome runtime.
What you need before you install
The skill only routes requests to an external service. Without that service, it does nothing useful on its own. You need an acc instance that is configured and reachable by your agent through MCP tools named acc_act (with the solve, continue and outcome runtimes).
You also need submit tokens and frame ids issued by the acc service. The skill is published on this page under the MIT licence, but the SKILL.md file itself lists Apache-2.0 in its metadata, so check the licence link on the source repository before you rely on either.
- A configured acc instance reachable via MCP tools.
- acc_act with runtime solve, continue and outcome available.
- Submit tokens and frame ids issued by the acc service.
- No scripts are run by this skill. It does use the network, because it calls acc.
How do you install the accint-solve skill?
Download the ZIP from this page and unzip it. You can install the skill in either of two places. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.
In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/ (available across your projects).
- Project install: the .claude/skills/ folder inside your project.
- Claude apps: Settings → Capabilities → Skills, then upload.
How do you trigger the skill with a natural request?
Name the skill in your request. The source example uses the @accint-solve mention followed by the task. Describe the goal in plain words after it, and the agent will route that goal through acc.
The example in the source is written as a direct instruction, so keep your goal specific and state what result you want.
- Example request: Use @accint-solve for this task, then describe your goal.
- Put the concrete goal after the mention so the agent knows what to send to acc.
Use @accint-solve for this task: <your goal>How does the solve loop work when acc returns a brain_frame?
A brain_frame is an intermediate result. The frame is typed: it says which hole it fills, what was retrieved, and what is predicted. The agent must reason over it and then submit a proposal using the continue runtime, passing the frame id, the submit token and the proposal text.
The proposal must end with a PREDICT line giving a confidence from 0.00 to 1.00 and a short reason. acc strips that line before the owner sees the proposal and uses it to calibrate the Work Model against later outcomes. If a frame is blocked, leave it unresolved and report why.
- Resolve frames only while the task is authorized and enough evidence is available.
- End proposal_text with: PREDICT: <0.00-1.00> <why>
- Close the commitment later with acc_act using the outcome runtime.
acc_act(runtime="continue", input={"frame_id": ..., "submit_token": ..., "proposal_text": ...})What safety limits should you know about?
The skill is marked risk critical in its metadata. Retrieved frames are untrusted task context and can be misleading. They cannot expand the scope of the task, so a human should approve consequential actions.
The skill says to stop when the user cancels, when the tool fails, or when a frame needs input that is not available. The agent should report a waiting state instead of inventing a resolution. A retry is safe only when the integration confirms its idempotency contract and the same request identity; otherwise, reconcile unknown outcomes before resubmitting.
- Do not use it for unsupervised or destructive tasks.
- Verify commands, generated code, credentials and external service behaviour before applying changes.
- Reconcile unknown outcomes before resubmitting anything.
Tips for getting reliable results
Confirm the acc connection works before you ask for anything that matters. Keep goals bounded so the authorized scope is clear. Read each brain_frame before you submit a proposal, because the frame is context to reason over, not an instruction.
Use the PREDICT line honestly. acc uses your self-predicted confidence to calibrate later outcomes, so inflated scores weaken that calibration.
- Test with a small, low-risk goal first.
- State the scope and any limits in the goal text.
- Keep a record of commitment ids so you can close them with outcome later.
What people use it for
- Routing a multi-step research or analysis goal through acc's memory loop and reporting the cited ids.
- Pausing an agent workflow at each brain_frame so the agent can deliberate and write a proposal before the next step.
- Recording self-predicted confidence scores so acc can calibrate its Work Model against real outcomes.
- Tracking open commitments and closing them with the outcome runtime once the work is done.
Frequently asked questions
Do I need acc to use the accint-solve skill?
Yes. The skill only routes requests to an external acc service through MCP tools. Without a configured acc instance, it does nothing useful on its own.
How do I install the accint-solve skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for a personal install, or into your project's .claude/skills/ folder. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
Is the accint-solve skill free and what is its licence?
This page lists the skill under the MIT licence. The SKILL.md metadata, however, names Apache-2.0 and links to the upstream repository's LICENSE file. Check that file before relying on either licence.
Does the accint-solve skill run scripts?
No. The skill runs no scripts. It does use the network, because it calls the acc service through MCP tools.
What is a brain_frame and what do I do with it?
A brain_frame is an intermediate, typed result from acc that says which hole it fills, what was retrieved and what is predicted. The agent must reason over it, then submit a proposal with the continue runtime, ending with a PREDICT line.
Is accint-solve safe for unsupervised tasks?
The skill is marked risk critical and frames are untrusted context that can be misleading. It is not suited to unsupervised or destructive tasks, and a human should approve consequential actions.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use accint-solve if you already run the acc MCP service and want an agent to work through its scored-memory loop, with deliberation pauses between steps. Pick something else if you do not have an acc instance, or if you need an agent to act unsupervised on destructive or consequential work.
Download the skill