CodewrightConsequence-aware terminal workflows

Terminal workflows that

Disclose risk.
Require confirmation.
Define verification.
Include recovery.

Codewright turns terminal work into reviewed procedures for developers and coding agents. You see what changes, how success is proven, and how to recover before you run anything.

  • Consequence-aware

    Every step exposes impact, risk, and required safeguards.

  • Reviewed and reproducible

    Structured recipes replace plausible-looking snippets.

  • Recovery is part of the plan

    Undo and verification are documented before action.

One command to install

npx @codewright/install
See every install path

How it works

Know what happens next.

FLOW 1:1
Exploded Codewright workflow assemblyFour separated procedure plates show inspection, action, verification, and recovery connected into one consequence-aware workflow.INPUT / FUZZY PROBLEMOUTPUT / VERIFIED FIX1INSPECTREAD-ONLY FACTSNO CHANGES MADEPART CW-012ACTEXPLICIT ACTIONCONFIRMATION REQUIREDPART CW-023VERIFYEVIDENCE COLLECTEDSUCCESS IS OBSERVABLEPART CW-034UNDORECOVERY PATHDEFINED BEFORE ACTIONPART CW-04
01InspectUnderstand the current state before choosing a fix.02ActRender the exact platform command with its safeguards.03VerifyRun the recipe's own check and compare the outcome.04UndoReturn to the prior state when verification fails.
current stage pending recovery available
CW consequence-aware terminal workStatus: building in the openIllustrative flow, real product safeguards

A fuzzy problem becomes a verified fix.

Workflows are more than a searchable list of commands. They run safe diagnostic checks first, branch on real output, and route you to the narrowest applicable recipe.

  1. 01
    Problem

    A port is already in use

    Start with the symptom. Codewright does not pretend the first matching command is automatically the right one.

    Intent captured
  2. 02
    Inspect

    Identify the listening process

    Run read-only checks first and branch on what the machine actually reports.

    Cause isolated
  3. 03
    Act

    Choose the narrowest safe fix

    See risk, platform, and confirmation requirements before a state-changing command is rendered.

    Action disclosed
  4. 04
    Verify

    Confirm the port is free

    The workflow defines its own observable success condition instead of stopping after execution.

    Outcome verified
Browse workflows Browse recipes

One source / five working surfaces

The safety model travels with the command.

Recipes, workflows, API responses, the CLI, MCP tools, and the agent skill speak the same language for risk, verification, and recovery.

01Recipe libraryReviewed terminal proceduresBrowse recipes →02Diagnostic workflowsRead-only triage before a fixBrowse workflows →03Companion CLIInspect, confirm, execute, verifyRead CLI docs →04MCP serverThe same safeguards inside an agent hostRead MCP docs →05Agent skillA shared safety language for humans and agentsRead skill docs →

Issue your first work order

Install once. Stop guessing.

Detect your supported agent harness and register Codewright's MCP server and agent skill from one command.

npx @codewright/install
See install paths