Clio Coder
All tutorials

Verification5 min read

Check a change against your repository

Declare which checks matter, run them through verification, and inspect whether passing results still match the files you are reviewing.

Written for
v0.5.7
Works in
Terminal · Desktop alpha
Basis
Documented workflow with v0.5.7 captures
Desktop app with the Results panel showing a verify run that passed eight of eight tests beside the conversation.
Artifacts, Results: the latest verify run, its exit status, and the output that counts eight passing tests. Desktop alpha · v0.5.7
On this pageBefore you start Start with checks the project already owns Declare what particular changes require Inspect freshness after editing Read the result where you work

A useful coding result includes more than "done." You need to know what changed, which checks ran, and whether those results still apply to the files you are reviewing. Clio Coder v0.5.7 can use your repository's declared checks and a project quality policy to make those questions explicit.

Before you start

  • A Git repository with a test or build check it already declares, such as a package script or a CI step.
  • Clio Coder with a model that supports tool calling.
  • Time to read the resolved command before you approve it.

Start with checks the project already owns

The verifier derives checks from supported build systems, project scripts, and CI definitions. When it cannot resolve a requested check, it should not substitute an arbitrary command and call it a pass.

Ask Clio

Discover this project's verification checks. Run the relevant test check through verify. Report its exit status, failures, and anything you could not check.

  1. Terminal transcript showing Clio listing checks and running the test check with exit 0 and reporting seven passing tests.

    01 A turn that listed the declared checks and ran the test check through verify. Terminal · v0.5.7

  2. Terminal view overlay showing the declared test check and the project quality policy requiring the test check for .mjs changes.

    02 /view shows the declared test check and the project quality policy that requires it. Terminal · v0.5.7

A turn that listed the declared checks and ran one, and the recorded listing it read.

Read the resolved command and working directory before approving execution. verify is subject to command admission: naming a check does not grant permission to run it, and a denied command is not a validation result.

The tools guide covers discovery and execution. The temperature-calibration example follows a small numerical correction with recorded test results.

Declare what particular changes require

A quality policy selects required checks by changed path. At the Git workspace root:

.clio-coder/quality.yaml

version: 1
rules:
  - id: source
    paths: ["src/**", "tests/**"]
    inputs: ["src/**", "tests/**", "package.json", "package-lock.json"]
    checks: [test]
    allowLimitations: false
  1. Name the changes a rule covers

    paths selects the rule: here, any change under src/ or tests/.

  2. List every input the result depends on

    inputs feeds freshness. Include fixtures, configuration, helper files, and lockfiles; this example assumes the root package declares a test script and uses that lockfile.

  3. Name existing checks, not commands

    checks refers to declared checks. A policy cannot insert an arbitrary shell command or bypass approvals.

  4. Decide whether limitations are acceptable

    allowLimitations: false means a reported limitation does not satisfy the rule.

Clio does not write this policy for you. Review it as a repository contract: a command that passes a narrow subset of tests should not satisfy a policy that requires the full declared test check.

Inspect freshness after editing

When a passing check still counts

  1. Your project

    Declared check

    A package script, build target, or CI definition the project already owns.

  2. verify runs it

    Results

    Recorded pass

    Stored with source, check-declaration, and policy fingerprints.

  3. an input changes

    Assessment

    Stale

    The earlier pass no longer describes the files.

  4. verify runs again

    Assessment

    Current

    A result for the work that exists now.

A recorded result belongs to the inputs it ran against; an edit to those inputs makes it stale until the check runs again.

Clio compares source, check-declaration, and policy fingerprints. A passing check becomes stale after relevant inputs change, and a later failure supersedes an earlier pass of the same check.

This matters during an iterative fix. The agent might run a test, make another edit, and then summarize the earlier success. The current assessment should describe the work that exists now, not the best result observed along the way.

Required checks run through verify without argument or working-directory overrides. Declare subproject checks with the correct working directory in the supported catalog or a root script. The quality-policy guide explains check identity and freshness.

Read the result where you work

On the desktop alpha, open Artifacts, then Results, beside the conversation; Files shows paths and changes recorded by tools. In the terminal, /view opens artifacts and receipts.

Desktop app with the Results panel showing a verify run that passed eight of eight tests beside the conversation.
Artifacts, Results: the latest verify run, its exit status, and the output that counts eight passing tests. Desktop alpha · v0.5.7

What to look for

The actual check, its result, and any missing evidence. A receipt is an inspectable record of the run; it is not a certificate that every scientific or operational requirement was met. New receipts use a simpler schema, and older receipts remain readable and verifiable.

Clio Coder desktop trace of a verify turn: success status, tokens, runtime, model, target, start time, and a phase waterfall
The trace of one verify turn: its status, recorded facts, and execution sequence. Desktop alpha · v0.5.7

Boundaries

  • Snapshots need Git and a workspace at the repository root. They cover tracked and nonignored untracked source within the documented bounds.
  • They do not establish the state of external datasets, installed dependencies, running services, or other hidden inputs.
  • A valid policy selects high rigor by default and asks to continue for outstanding checks when recovery is possible; normal rigor makes the assessment advisory. Neither grants extra execution authority.
  • Numerical comparisons, integration tests, and human review may still be necessary.

By The Clio team · Clio Coder

Continue with the documentation