DeltaReceipt — Real Checks for AI Code Changes

    1

    **Run real checks after the change.** Review and execute the repository's own lint, typecheck and test commands.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    About this skill

    What changed, and what do the checks actually show?

    DeltaReceipt is a free, MIT-licensed Agent Skill and local Python CLI that compares repository checks before and after AI-assisted code changes and records the evidence in a JSON Quality Receipt.

    How it works

    1. Capture the baseline before editing. Record the working-tree state, including relevant staged, unstaged and non-ignored untracked files, and run the repository's explicitly configured checks.
    2. Run real checks after the change. Review and execute the repository's own lint, typecheck and test commands. Record actual command outputs, exit codes and file hashes.
    3. Compare the evidence. Separate known historical findings from newly observed regressions within supported comparison constraints, then inspect the Quality Receipt. Check whether an existing receipt is still fresh before relying on it.

    What the Quality Receipt tells you

    • no_new_observed: no new issues observed within the supported checks and evidence constraints; this is not a guarantee that the code is defect-free.
    • regression: a newly observed regression was found in the supported comparison.
    • incomplete: the evidence does not support a safe comparison, including a missing genuine pre-edit baseline or stale or incomplete evidence.

    Supported comparisons: ESLint JSON diagnostics, TypeScript diagnostics and Node TAP test outcomes. Generic exit-only tests cannot certify zero new defects, and changed files can make attribution ambiguous.

    Requirements: Python 3.10+, Git, a Git repository with at least one commit, and already-installed project checking tools. The user or agent must review the planned shell commands before execution. No DeltaReceipt cloud account or telemetry is required by the core CLI.

    Limits: No automatic fixes or hooks, no proof that the AI authored a particular defect, no claim that every test is discovered, and no guarantee of correctness outside the configured checks. If a true pre-edit baseline is missing or the comparison is unsafe, the outcome is incomplete rather than fabricated green status. The underlying repository checks execute under the user's permissions and might themselves use the network.

    Verified host scope: The renamed $delta-receipt was successfully invoked in one isolated Codex CLI Windows task and returned a fresh no_new_observed Quality Receipt. Historical v0.1.0 Claude Code native skill invocation was also recorded using the predecessor's former name; the renamed /delta-receipt path was not verified because Claude Code's external API refused the connection before the task started. Cursor and Copilot installation locations are documented but native E2E invocation is unverified.

    DeltaReceipt is the successor to Agent Quality Skill v0.1.0. Existing .agent-quality.json, .agent-quality/ and agent-quality/v1 receipt conventions remain. Old v0.1.0 receipt compatibility is limited to the exact recognized implementation fingerprint. Inspired by the related dsh-lint-loop project; it runs independently of DSH.

    Version: v0.2.0

    Skill ID: delta-receipt

    Source repository: https://github.com/lemonxiny55/delta-receipt

    Release: https://github.com/lemonxiny55/delta-receipt/releases/tag/v0.2.0

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.

    Requires a genuine pre-edit baseline. Missing, stale, unsafe or incomplete evidence returns incomplete. Supported comparisons focus on ESLint JSON, TypeScript diagnostics and Node TAP; generic exit-only tests cannot certify zero new defects. Changed files can make attribution ambiguous. No automatic repairs, lifecycle blocking, zero-false-positive guarantee, or universal agent validation. Hashes are unsigned integrity checks. Codex v0.2.0 verified once; renamed Claude Code v0.2.0 invocation remains unverified.

    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

    1 use

    Times used to date

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 26 days ago

    • Free with an Agensi account

    Software
    Python 3.10+GitProject lint toolsProject typecheck toolsProject test tools

    Needs a Git repository with at least one commit. The lint, typecheck and test tools must already be installed.

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Runs Python, Git and explicitly reviewed/configured repository checks with the invoking user’s permissions. Reads selected repository files and Git metadata; writes baseline and receipt JSON under .agent-quality/ and approved configuration in .agent-quality.json. No automatic source edits, reset, stash, hooks or repairs. The core CLI needs no browser, API key, cloud account, telemetry or network. Repository checks may use network and inherit process environment; the optional disposable demo can run npm ci --ignore-scripts only with approval. Receipts can contain command output and local file paths.

    Listed26 days ago
    Updated22 days ago

    What's inside

    Frequently Asked Questions

    Yes. DeltaReceipt v0.2.0 is free ($0), MIT-licensed and open source. No subscription or paid license is required. Source: https://github.com/lemonxiny55/delta-receipt

    No. It runs explicitly when invoked, records checks and classifies supported evidence. It does not automatically repair code, install hooks or intercept edits. Users or coding agents handle repairs separately.

    One explicit native $delta-receipt invocation was verified in Codex CLI on Windows for v0.2.0. Claude Code has a native invocation record for the predecessor v0.1.0 under its former name; the renamed v0.2.0 /delta-receipt invocation remains unverified. Cursor and Copilot native E2E invocation is unverified.

    You can run current checks, but DeltaReceipt reports incomplete rather than claiming no new regressions. Supported outcomes are no_new_observed, regression and incomplete, limited to configured checks and the recorded evidence.

    New in Testing & Debugging

    Free