More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Checkpointed Execution

    1

    Execute complex, multi-step tasks with mandatory review checkpoints and evidence-based verification.

    Free

    5 usesSecurity scanned
    Checkpointed Execution

    Checkpointed Execution

    Example session with this skill installed

    /checkpointed-execution mode=standard. Let's migrate the local sqlite users to the production postgres instance. Stop after each step.

    • Read your context and instructions
    • Compiled the checkpointed execution

    I will manage the migration of users from the local SQLite database to the production PostgreSQL instance using standard checkpointed execution.

    Migration Plan: Local SQLite to Production PostgreSQL

    StepOwnerActionVerificationStatus
    0[ME]Preflight: Check connectivity to both databasesConfirm read access and schema matchPENDING
    1[ME]Export: Dump local users to users_dump.jsonCheck file size and row countPENDING
    2[YOU]Sanitize: Review dump for PII or test dataUser confirmation of data cleanlinessPENDING
    3[ME]Schema: Apply migrations to PostgreSQLQuery information_schema for table existencePENDING
    4[ME]Import: Load users_dump.json into PostgreSQLCount rows in target vs sourcePENDING

    Assumptions: Production credentials in .env have INSERT permissions.
    Out of Scope: Migration of related orders or posts tables.

    Step 0: Preflight

    • Action: Attempted connection to local.db (SQLite) and prod_db (PostgreSQL).
    • Result: Successfully connected to both. SQLite contains 1,240 records in users. PostgreSQL users table does not yet exist.
    • Verification: SELECT count(*) FROM users on SQLite returned 1240.

    Next steps

    1. Review the plan above for any missing constraints.
    2. Reply go to start Step 1: Exporting local data.

    ⏸ Checkpoint 0/4 · standard — next: export the source records. Reply go to continue.

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

    About this skill

    Avoid expensive rollback costs and configuration drift in complex deployments. This skill enforces a structured, step-by-step execution framework designed for stateful operations like migrations, cluster builds, and API integrations. It replaces the "black box" execution style with a transparent series of reviewed checkpoints where every action is verified by evidence rather than status codes.

    What it does

    • Multi-step planning creates a detailed execution map with owner labels, verification steps, and risk assessments.
    • Mandatory checkpointing pauses execution after every discrete step to allow for user review and redirection.
    • Evidence-based verification queries the actual state of the system to prove success instead of relying on return codes.
    • Stateful runbooks maintain a persistent record of every command run, surprise encountered, and result observed.
    • Variable modes adjust the stopping frequency from lite (phase boundaries) to strict (before every write).

    How it works

    1. Initialize: Start by invoking /checkpointed-execution with a chosen mode or by describing a complex multi-step task.
    2. Plan: Review the generated plan which identifies assumptions, scope, and specific verification methods for each step.
    3. Execute & Verify: The agent performs one step, verifies the outcome through read-only checks, and presents a step report.
    4. Confirm: Provide a "go" or "next" command to proceed, or provide corrections that are immediately applied to the plan and future steps.

    Frameworks & tools

    This skill is language-agnostic and works across CLI tools, cloud providers (AWS, GCP, Azure), database migrations, and CI/CD pipelines. It is particularly effective for Terraform, Kubernetes, and complex Python or Node.js integration scripts.

    Why this beats prompting it yourself

    Standard agents often "run ahead," completing five steps before realizing a typo in step one broke the entire state. This skill forces rigorous stopping points and formal evidence reporting that generic prompts ignore when they prioritize autonomous completion.

    Use cases

    • Infrastructure builds: Setting up VPCs, clusters, or databases where mistakes are costly to undo.
    • Data migrations: Transferring records between systems while verifying integrity at each batch.
    • Production changes: Implementing updates on live systems where every write requires a second pair of eyes.
    • Onboarding/Learning: Guiding a developer through a new environment where they perform the manual steps.

    Known limitations

    Not intended for single-file edits, trivial bug fixes, or read-only investigations. Requires a user to be present for confirmation unless running in headless mode for report generation only.

    Best for multi-step stateful work. Not for trivial bug fixes or single-turn edits. May require user feedback to proceed past checkpoints.

    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

    5 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 29 days ago

    • Free with an Agensi account

    Permissions

    Read Files
    Listed29 days ago

    What's inside

    Frequently Asked Questions

    Standard agents often hallucinate success or overlook errors until the entire task fails; this skill enforces mandatory verification steps that require the agent to prove a step succeeded before moving to the next.

    Yes, you can toggle between 'strict' mode, which pauses after every individual command, and 'lite' mode, which only stops at major phase boundaries or high-risk actions.

    The skill is tool-agnostic and integrates with any environment the host agent can access, including Terraform, Kubernetes, AWS CLI, and various database management tools.

    You receive the full Checkpointed Execution logic framework, a set of pre-configured templates for common migration and build scenarios, and documentation for integrating it into your existing AI workflows.

    While the skill is designed for interactive human-in-the-loop oversight, it can be configured for 'headless' mode where it generates a detailed, evidence-backed audit log for post-execution review.

    New in Agents & Orchestration