Skip to content

Guide · Agent skill

How to use the accint-frames agent skill for acc queues

Updated · written with AI from the project's own documentation and checked against our catalogue data

accint-frames is an agent skill that teaches an AI agent to work through a queue of pending decision points, called frames, held by a separate system named acc (accreted intelligence). Those frames are created when headless automated runs pause to ask for deliberation. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and what to check before letting it submit anything.

Accint Frames Queue Drainersickn33/agentic-awesome-skills · MIT

Quick start

  1. 1Set up acc. Confirm you have a configured acc instance with its CLI and MCP tools, and that you hold the submit tokens it issues.
  2. 2Install the skill. Download the ZIP from this page, unzip it, and place the accint-frames folder in ~/.claude/skills/ or a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills.
  3. 3List the queue. Run acc frames to see the open and waiting frames without changing anything.
  4. 4Make a scoped request. Ask the agent to use @accint-frames and name the authorized queue subset so it stays within scope.
  5. 5Review and submit. Check each proposal and its PREDICT line, then let the agent submit through acc_act, and confirm the outcome before any retry.

What the accint-frames skill makes the agent do

An agent skill is a folder with a SKILL.md file that gives an AI agent extra instructions for one kind of task. This skill covers one task: clearing acc's deliberation queue, which is made of open or waiting brain_frames that headless runs have checkpointed.

The skill describes itself as routing logic over two MCP verbs and says no decision logic lives in the file itself. The real behaviour sits in the acc system. The agent lists the open frames, reads each frame's typed hole and retrieved context, writes a proposed answer, and submits it back through acc's tool interface.

What you need before installing

The skill only routes calls to acc, so it does nothing useful on its own. You need a configured acc (accreted intelligence) instance with its CLI, which provides the acc frames command, and its MCP tools, which provide acc_act.

You also need credentials or submit tokens supplied by your acc instance, because each resolution is submitted with a submit_token. The skill makes network calls and runs no scripts, according to the listing.

  • An acc instance with the CLI and MCP tools available
  • Submit tokens or credentials issued by that acc instance
  • An agent environment that can load skills

Install the accint-frames skill

Download the ZIP from the skill's page on this directory and unzip it. Place the accint-frames folder in one of these locations, depending on how you want it available:

  • Personal: ~/.claude/skills/
  • Project only: the .claude/skills/ folder inside your project
  • Claude apps: upload the ZIP under Settings → Capabilities → Skills

Trigger the skill with a natural request

Name the skill in your request so the agent loads it. The source example uses the @ mention form and then describes the task in plain words. Describe the goal, not the internal steps, and say which queue subset is authorized.

The skill expects you to resolve the configured acc instance and the authorized queue subset first, so state both in the request.

  • Use @accint-frames to drain the open acc frames for the billing project only.
  • Use @accint-frames to list the waiting frames and resolve only the ones that have enough context.
Use @accint-frames for this task: Drain acc's deliberation queue — open/waiting brain_frames checkpointed by headless runs — via acc_act(runtime="continue").

How a queue run works

The skill follows a fixed sequence. It lists the queue with the read-only acc frames command, then handles each open or waiting frame one at a time. Each submission goes through acc_act with the frame_id, submit_token and proposal_text fields.

Each proposal ends 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 its Work Model against later outcomes. Frames the agent cannot resolve stay in the waiting state.

  • Read each frame's typed hole and retrieved context before answering
  • Surface each resolution's commitment id and cited [ids]
  • Check the installed runtime's duplicate-submit contract before any retry
acc frames

Safety notes before you run it on real work

Submitted resolutions can influence later automated decisions in acc, and the skill file has no approval gate for each submission. Do not run it unattended on live systems. Review what the agent proposes before the submissions are sent, or limit the run to a small, authorized subset.

The skill treats retrieved frames as untrusted task context that cannot widen its scope. It stops when you cancel, when the tool fails, or when a frame needs input it does not have. Retries are only safe when the integration confirms its idempotency contract. If an outcome is unclear, reconcile it before resubmitting, so you do not duplicate a decision.

  • Requires network access to reach your acc instance
  • Runs no scripts, according to the listing
  • Verify credentials and external service behaviour in a test environment first

Tips for getting reliable results

Start with a narrow queue subset and a single frame so you can see the proposal format and the PREDICT line. Keep the goal in your request specific, because the skill will not go beyond the scope you authorize.

Keep a record of each commitment id and the cited ids the agent reports. These let you trace a resolution back to the frame and the context it used.

  • Run acc frames yourself first to see what is waiting
  • Ask the agent to leave unresolved frames waiting rather than guessing
  • Check the license file in the repository, because the listing and the file header name different licenses

What people use it for

  • Clearing a backlog of waiting frames created by a team's headless agent runs
  • Reviewing proposed answers for a limited set of frames before they feed later automated decisions
  • Tracing which context and cited ids supported each resolution using the commitment ids the agent reports

Frequently asked questions

How do I install the accint-frames agent skill?

Download the ZIP from this page and unzip it. Put the accint-frames folder in ~/.claude/skills/ for personal use or in a project's .claude/skills/ folder, or upload the ZIP under Settings → Capabilities → Skills in the Claude apps.

What do I need to use accint-frames?

You need a configured acc (accreted intelligence) instance with its CLI, which provides acc frames, and its MCP tools, which provide acc_act. You also need credentials or submit tokens from that instance, since the skill cannot do anything useful without acc.

Is accint-frames free to use?

The directory listing says MIT, but the SKILL.md header names Apache-2.0 and links to a LICENSE file. Read that LICENSE file in the source repository to confirm the terms before you use or redistribute the skill.

Does accint-frames run scripts or use the network?

According to the listing, it runs no scripts and does use the network, because it sends calls to your acc instance. Its submissions can influence later automated decisions, so review them before use.

What should I do if a submission response is unclear?

Do not assume it succeeded. The skill says to check the installed runtime's duplicate-submit contract first and to reconcile any unknown outcome before resubmitting.

How is this different from other agent skills for queues?

This skill is narrow. It handles only acc's brain_frames deliberation queue through acc_act and holds no decision logic itself, so it suits teams already running acc and is not a general task manager.

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-frames if you already run an acc instance and want an agent to work through the frames your headless runs leave waiting, with a human reviewing submissions. Pick something else if you do not have acc, or if you want unattended processing on live systems, since the skill has no per-submission approval gate.

Download the skill