Agents I have designed and shipped for QE workflows — one public, the rest running inside organisation. The internal agents are described at a high level only; source and prompts stay with the client.
Standalone LLM-based code review tool. Paste a GitHub or GitLab URL and the agent scans the diff for style, naming, POM, and framework-convention violations. Pluggable LLM backends (Claude, Gemini, OpenAI, Ollama), structured JSON output, GitHub Actions wiring.
Detects, ranks, and explains flaky tests from CI run history. Scores each test's flakiness, ranks tests by wasted CI time, and uses an LLM to suggest a likely root cause.
Reads business requirements from Confluence, generates compliance-format acceptance criteria and pushes them into the matching Jira ticket. Cuts BA-to-QE handoff time.
Analyses Jira acceptance criteria, generates Gherkin feature files, and detects existing step definitions to prevent duplication. Reduced feature authoring cycle from 3 days to 18 hours.
Inspects DOM for new UI flows. Reuses existing Page Object components and locators when possible, otherwise generates framework-conformant new ones and auto-maps feature files to glue and step definitions.
Analyses commit diffs for impacted Salesforce modules
(@opp, @lead, @pardot), triggers a targeted subset for
fast PR feedback, then runs full @regression as second-pass safety.
Hardened version of the public Code Review Agent, scoped to internal banking microservice repos. Runs against GitLab merge-request diffs, enforces POM and framework standards, posts inline comments. Adopted across 4 active QE repos.
Across the agents above: regression reduced from 5 days to 4 hours, 25% drop in production defects, 60% improvement in MTTD for critical issues, and SDET onboarding time cut from 3 weeks to 10 days.