claude-kit: Durable Task Execution

    1

    Keep long-running coding work anchored to a durable task file: preserve the goal, execute one increment, verify results, and leave a clear next step for resuming after context compaction.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    claude-kit: Durable Task Execution

    Example session with this skill installed

    Use the durable task workflow to implement docs/specs/2026-10-29-dark-mode.md. Keep a task file that another session can resume.

    The example shows the initial task file before implementation; no tests or completion are claimed.

    • Read your context and instructions
    • Compiled the claude-kit

    Illustrative deliverable

    Example initial task file, not a recorded execution or benchmark.

    Task: Dark-mode toggle

    Spec: docs/specs/2026-10-29-dark-mode.md
    Status: in_progress
    Created: 2026-10-29

    Goal

    Add a theme toggle whose saved choice survives reload. Keep the accepted scope and evidence in a file so a later session can continue without relying on conversation history.

    Acceptance criteria

    • Settings exposes a keyboard-operable Light/Dark control with a visible label and selected state.
    • Changing the control updates the page without a reload.
    • Reload restores the saved theme.
    • Storage failure does not prevent switching themes for the current session.
    • Navigation and settings remain readable in both themes.

    Constraints

    No account sync, scheduled theme changes, or operating-system theme detection. Preserve unrelated work.

    Plan

    • Inspect current theme and settings code.
    • Add the toggle and persistence, then verify the behavior.
    • Check keyboard interaction and readability; record evidence.

    Next step

    Read the linked spec and inspect the existing settings and theme code.

    Progress log (append-only)

    • 2026-10-29 — Example task initialized from the approved spec; implementation and verification are pending.

    Resume

    In a fresh session, ask the agent to resume .claude/tasks/dark-mode.md using this workflow. It reads the file and linked spec, then continues from Next step.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Keep the task outside the conversation

    This workflow turns an approved spec or task description into a durable .claude/tasks/<slug>.md file. The agent records the goal, acceptance criteria, constraints, plan, next step, and append-only progress log so a later session can resume from the file.

    The execution loop

    1. Re-read the task file from disk.
    2. Complete one planned increment.
    3. Verify the result with relevant checks.
    4. Record evidence and update the next step.
    5. Make a small commit when working in a Git repository.

    The skill tells the agent to re-read the task and linked spec after compaction, stop after three consecutive failures on the same problem, and record adjacent issues without silently expanding scope. Completion requires evidence for every acceptance criterion.

    This package contains the original task/SKILL.md from Bradley Brewington’s claude-kit. Pair it with the separate Spec-First Interviews skill, or supply your own approved spec. It is a workflow instruction file; it does not run as a background service or guarantee that an agent will execute every instruction.

    Agensi supplies text instructions through MCP. The original /task syntax is shorthand for the locally installed Claude Code skill; this listing does not install a slash command, shell profile, or global configuration.

    Source: https://github.com/BradleyBrewington/claude-kit

    Requires local filesystem and terminal access plus the project’s own build/test tools. The agent must follow the workflow and update the task file; there is no background service or runtime enforcement. After three consecutive failures on the same problem, the instructions call for user input.

    How to use

    1. 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. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 19 days ago

    Requires

    Terminal access
    Your project's test and build tools

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Reads the spec and project files; writes .claude/tasks/*.md and files required by the approved task; runs project checks and makes Git commits. Browser or network access depends on the task and must follow the host agent’s existing permissions. No external service is inherently required.

    Listed19 days ago

    What's inside

    Frequently Asked Questions

    It keeps the goal, acceptance criteria, constraints, next step, and progress log in a Markdown file. The instructions tell the agent to re-read it for each execution increment and after context compaction. This supports continuity but does not guarantee model compliance.

    The original workflow was written for Claude Code. An agent using it through Agensi MCP needs local filesystem and terminal tools for implementation, project checks, and Git. Compatibility has not been independently tested across every agent or IDE.

    Task files are stored in .claude/tasks/<slug>.md within your project. They can be tracked in Git, provided their content is appropriate for that repository.

    After three consecutive failed attempts at the same problem, the workflow instructs the agent to mark the task blocked, record the attempts and reasons, and ask the user for help. This is an instruction, not a runtime-enforced retry limit.

    Ask your agent to use the durable task workflow with a spec path, existing task-file path, or description. It creates or resumes the task file. Agensi delivers instructions through MCP and does not install the original local /task slash command.

    New in Workflow Automation