More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Waker AI

    1

    Local CLI supervisor that enforces on-disk evidence and test verification for autonomous coding agents.

    Free

    Security scanned
    Waker AI

    Waker AI

    Example session with this skill installed

    Use waker to supervise this repo: implement tier-aware execution E1–E4, add the
    planner adapter and swarm CLI, update the parity docs, and make
    python .agents/tests/test_v2_m3.py pass. Stop only when everything is green.
    Use OpenCode with opencode-go/deepseek-v4.1-flash.

    • Read your context and instructions
    • Compiled the waker ai

    waker planned the contract with 4 tasks, then started the supervisor detached:

    waker start --detach --repo . --tasks .waker/tasks.yaml
    --adapter opencode --model opencode-go/deepseek-v4.1-flash
    --test "python .agents/tests/test_v2_m3.py"

    waker · run 989bcfca · 4/4 — 100%
    ████████████████ ⏱ 2h47m 🔁 0/5 🧪 PASS 💰 $0.0850

    | ✅ | T1 | Tier-aware E1–E4 | .agents/v2/execution.py | pass |
    | ✅ | T2 | Planner adapter | .agents/v2/planner_adapter.py | pass |
    | ✅ | T3 | Swarm CLI + launchers | .agents/v2/swarm_cli.py | pass |
    | ✅ | T4 | Parity docs | docs/*.md | pass |

    Diagnosis: DONE · report: .waker/runs/989bcfca/report.md

    What happened

    • The first turn hit the 15-minute cap, so waker stopped it, checkpointed the
      work, and resumed the same session with a 30-minute budget.
    • Later the agent claimed "done" while the suite was still red; waker flagged
      FALSE_DONE and woke it with the real failure line.
    • Every task was accepted only after git diff ∩ scope and its verify
      command passed. Test files were hashed; no tampering.
    • Total: 2h47m including earlier attempts, $0.0850, exit 0.

    Next steps: open .waker/runs/989bcfca/report.md for the evidence ledger, or
    merge the branch.

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

    About this skill

    Manual supervision of autonomous coding agents is prone to human error and premature exits. This skill implements a strict watchdog protocol that forces AI agents to provide on-disk evidence for every task and pass automated verification before completion.

    What it does

    • Task Planning generates a structured contract in YAML format defining specific scopes and verification commands.
    • Background Execution launches agents in a detached process with defined constraints on cost and retries.
    • Automatic Verification cross-references git diffs against task scopes to ensure work is actually committed.
    • Blind Evaluation supports hidden test suites stored in an external vault for Kaggle-style or grading scenarios.
    • Status Reporting provides live Markdown tables of task progress, costs, and linting errors.

    How it works

    1. Environment Setup initiates with a system check to ensure the local CLI supervisor is installed and configured.
    2. Contract Generation takes a verbatim user prompt to create a specific task list with scope and verify fields.
    3. Execution Kickoff starts the background daemon using a specified adapter for models like Claude Code or Gemini.
    4. Monitoring tracks the run ID and pulls real-time reports from the daemon until all tasks satisfy the watchdog.

    Frameworks & tools

    This skill requires the waker-ai Python package and works with coding agents such as Claude Code, OpenCode, and Gemini. It utilizes git for diff analysis and standard CLI shells for test execution.

    Why this beats prompting it yourself

    Prompting an agent to "test your work" often results in hallucinated success or skipped edge cases. This skill offloads the verification to a local daemon that ignores the agent's claims and only validates based on exit codes and file changes.

    Use cases

    • Supervise long-running refactoring tasks while you are away from the keyboard.
    • Run "Kaggle-mode" coding challenges where the agent cannot see the hidden test suite.
    • Enforce strict git-based evidence for every ticket in a batch migration.
    • Limit LLM spend by setting hard cost caps on autonomous background runs.

    Known limitations

    Requires local installation of the waker-ai CLI. Supervision is limited to tasks that can be verified via file changes or CLI exit codes.

    Links & License

    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.

    Requires local pip installation of waker-ai. Only supports tasks verifiable via CLI or git diffs. Hidden tests must be manually added to the vault before execution.

    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 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 4 days ago

    • Free with an Agensi account

    Software
    GitPython 3.11+waker CLI (pip install waker-ai)pytest (for verify gates)

    Permissions

    Terminal / Shell
    Write Files
    Network Access
    Read Files

    File Scopes

    .waker/**
    **/*.py
    **/*.md
    **/*.json
    **/*.toml
    **/*.yaml
    **/*.yml

    waker is a local CLI supervisor. Terminal is required to run `pip install waker-ai` (with explicit user consent), `waker plan/start/status`, git and pytest. It reads the repository to compute `git diff ∩ task scope` and writes only `.waker/` (tasks, ledger, reports, checkpoints) plus git checkpoint commits. The host agent CLI (OpenCode/Claude/Gemini) makes model calls; waker itself stores no API keys and requires no service account. Hidden-test vaults (optional blind mode) live outside the repository under the OS user-data directory, not in the project tree.

    Listed4 days ago

    What's inside

    Frequently Asked Questions

    waker acts as an external watchdog that ignores an agent's claims of success, instead requiring verifiable proof through git diffs and successful CLI test executions before a task is marked complete.

    The skill is designed to work with popular coding agents like Claude Code, OpenCode, and Gemini, provided they are running in a local environment where the waker-ai CLI can monitor file changes.

    Yes, the 'Blind Evaluation' feature allows you to store test suites in an external vault, enabling the agent to attempt tasks without seeing the grading criteria, which is perfect for internal coding challenges.

    You will need the `waker-ai` Python package installed locally and access to git, as the skill uses git-based evidence to verify that code changes actually occurred.

    Absolutely. You can define hard cost caps and retry limits within the background execution settings to prevent autonomous agents from exceeding your budget during long-running tasks.

    Upon purchase, you receive the waker skill configuration, access to the local CLI supervisor integration, and documentation for setting up the watchdog protocol across your development environment.

    New in Agents & Orchestration

    Free