ada-agent-delivery-audit

Audits multi-phase agent deliveries by re-running gates and verifying claims against evidence.

Updated Jul 23, 2026
One-click install
npx skills add https://github.com/wubing7755/Ada --skill ada-agent-delivery-audit-wubing7755
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ada-agent-delivery-audit
Source: https://github.com/wubing7755/Ada/tree/main/skills/software-development/ada-agent-delivery-audit
Command: npx skills add https://github.com/wubing7755/Ada --skill ada-agent-delivery-audit-wubing7755

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When a long-running or delegated agent claims it completed a large multi-phase plan, its delivery report is only a claim, not a fact. This Skill independently verifies the delivery by re-collecting evidence instead of trusting statements like "migrated", "tests pass", or "done". ## Core Features & Use Cases - Forensic Commit Review: Maps git commits to plan phases, flags scope creep, unrecorded decisions, and new commits made during the audit. - Irreversible-Gate Checks: Verifies that destructive actions (deletions, migrations, overwrites) requiring user confirmation were not executed prematurely. - Data and Migration Verification: Queries databases directly (e.g., SELECT COUNT per table) and inspects timestamp signatures to detect silent failures like relative connection strings creating empty databases. - Gate Re-Runs and Exact Command Reproduction: Re-runs full test suites, format checks, and security scans using the exact CI/script commands rather than invented equivalents. - Use Case: An agent reports it migrated all content to a database and all tests pass. This Skill queries the database directly, discovers the new tables are empty due to a relative connection string, re-runs the test suite itself, and produces a severity-graded issue report with file:line evidence. ## Quick Start Ask the agent to audit the delivery of the agent that just claimed to finish the multi-phase plan, verifying its claims against git history, the database, and a fresh test run.

Frequently Asked Questions about ada-agent-delivery-audit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify an agent's claim that it completed a multi-phase plan?

Start with a forensic snapshot using git log, git status, and per-commit git show --stat to map commits to plan phases. Then check irreversible gates, verify data claims with direct database queries, and re-run all test and format gates yourself rather than trusting the report.

How to verify a database migration actually ran?

Query the database directly with SELECT COUNT per table instead of trusting a "migrated" claim. Check timestamp signatures on rows, since imported data written by a tool shows current timestamps while old seed data retains source dates, revealing whether the tool ever ran on that database.

Why did a migration report success but the database is empty?

A common cause is a relative connection string like Data Source=app.db, which resolves against the current working directory and silently creates a new empty database. Require explicit absolute connection strings, per-table progress logs, idempotent upserts, and a final row-count self-check.

Should I trust an agent's report that all tests pass?

No, treat test counts in a delivery report as claims and re-run the full test suite, format checks, and security scans yourself. Reproduce the exact CI or release script commands verbatim, since different arguments can produce different results and false positives.

When should I not use this delivery audit process?

Do not use it for routine PR reviews, pre-commit self-review, batch plan execution, or general code quality analysis, which belong to dedicated review and delivery skills. It is specifically for independent verification of large multi-phase deliveries before merge or release.