Knowledge Share
Technical articles, tutorials, and insights
LLM-Driven Automated Testing Series (14): Production Engineering — Why Testing Turns Out to Be the Cheapest Line Item in This Whole Series
A code-generation workflow I documented earlier burned thousands of dollars in tokens during debugging. The testing scenarios verified across this series — AutoRestTest testing a 15-operation API for about a dime, pytest-triage capping itself at 10 model calls per run by default — are off by orders of magnitude. This article puts the evidence from 13 prior articles side by side to answer the three questions the planning doc left for last: why testing-scenario cost is what it is, where the human-in-the-loop gate actually belongs, and which scenarios should explicitly avoid LLMs altogether.
LLM-Driven Automated Testing Series (13): Flaky Tests Aren't a Rerun-Count Problem, They're a Classification Problem
pytest-rerunfailures (477 stars) and flaky (397 stars) handle test instability with the same one-liner: if it fails, rerun it a few times. That doesn't answer what the planning doc actually asks — was this failure a real bug, environment jitter, or a badly designed test? The only project actually attempting that root-cause classification with an LLM is a 2-star, less-than-two-months-old repo called pytest-triage — and its most instructive feature turns out to be the four safety invariants it wraps around AI judgment, guaranteeing AI is never allowed to affect whether a test actually passes or fails.
LLM-Driven Automated Testing Series (12): Beyond Pixel Diffs — Who's Actually Making LLMs 'Understand' UI Changes
BackstopJS, at 7.2k stars, handles rendering noise with exactly four numeric config knobs — misMatchThreshold, requireSameDimensions, ignoreAntialiasing, usePreciseMatching — which is still just fighting font-rendering jitter with pixel-level tolerance. The project that actually wires a VLM/LLM into the visual-regression judgment loop is a 23-star repo called vlmkit, whose own internal benchmark notes state outright that one model hallucinates red into red. Putting these two side by side answers, concretely, what semantic visual regression actually solves — and what new problems it introduces.
LLM-Driven Automated Testing Series (11): Why API Testing Is Actually the Most Conservative Battleground for LLMs
Keploy has 18.5k stars and its README says 'Expand API Coverage using AI' — but grep the entire open-source Go codebase for LLM/AI keywords and you'll find zero implementation code. Schemathesis, at 3.6k stars, uses no LLM at all — it finds real server 500 errors purely through Hypothesis-based property-based fuzzing. The projects that actually call an LLM API in their core logic turn out to be two sub-100-star projects: AutoRestTest and api-automation-agent. This article pins down why: structured API input/output is naturally suited to traditional fuzzing, so what can LLMs actually do here that traditional tools can't — and why does this domain stay cautious instead of going all-in the way UI automation has?
LLM-Driven Automated Testing Series (10): Self-Healing Locators — Where Heuristic Scoring Hits the Ceiling of Semantic Understanding
Healenium is the most established open-source project in self-healing locators, and its README claims it 'leverages machine learning' — but reading the code directly reveals the core mechanism is a weighted scoring formula over DOM tree similarity, with hardcoded constants for weights and thresholds, nothing to do with machine learning. More notably, its 2026 addition of an AI-based XPath generation endpoint is gated behind a hard-coded exception: 'you must have a paid hlm-ai service.' This article first cracks open exactly how Healenium's scoring formula works, then examines how MarketSquare/robotframework-selfhealing-agents — a genuinely LLM-based self-healing project — designs its multi-agent architecture, and finally answers the question the planning doc raised: where exactly does the tradeoff between healing success rate and false-positive rate actually bite.
LLM-Driven Automated Testing Series (09): Mobile Automation (V) — When a Generic Agent Framework Sinks Down into a Testing Scenario
MetaGPT is a 'multi-agent software company' framework built for requirements analysis and code generation, but it ships a built-in Android testing demo. Reading the source directly overturns one premise from the planning doc: this demo does not reuse MetaGPT's usual Product Manager/Engineer/QA role division — it has exactly one custom Role, reusing instead the lower-level Team/Environment/Role/Action scheduling scaffolding. More importantly, the README states plainly that it 'referenced ideas and code from AppAgent' — this piece breaks down exactly what this generic framework inherited from AppAgent, what reusing generic scaffolding actually bought it, and what it cost.
LLM-Driven Automated Testing Series (08): Mobile Automation (IV) — AppAgent: Combining Parser Trees with Visual Features
AppAgent is Tencent's open-source CHI 2025 project. Unlike the pure-visual (ARTEMIS/Mobilerun) or pure-coordinate (Mobile-Agent-v3) approaches covered earlier, it uses the accessibility tree to draw numeric labels on screenshot elements, and the model only needs to pick a number — the 'parser tree' and 'vision' aren't two independent perception channels at all; structural information handles label placement while the vision model handles the decision. This piece breaks down how the autonomous-exploration phase builds documentation via a four-way reflection (BACK/INEFFECTIVE/CONTINUE/SUCCESS), how the grid-fallback mechanism degrades when the parser tree fails, and honestly points out a real coordinate-conversion bug living inside that grid-fallback path.
LLM-Driven Automated Testing Series (07): Mobile Automation Part 3 — Mobile-Agent-v3 and the Self-Trained GUI-Owl Approach
Mobile-Agent-v3 is Alibaba Tongyi Lab's multi-agent framework built on the self-trained GUI-Owl model — a completely different bet from the previous two articles' reliance on general-purpose VLMs (Gemini/GPT/Claude): train a dedicated GUI-grounding model that outputs 'absolute pixel coordinates' instead of guessing. This piece dissects the Manager/Executor/ActionReflector/Notetaker role split, the 'two consecutive failures before escalating to Manager' fault-tolerance threshold, and a real, verified HarmonyOS input bug.
LLM-Driven Automated Testing Series (06): Mobile Automation Part 2 — DroidRun/Mobilerun's Role-Level Model Split
DroidRun (now renamed Mobilerun) uses a unified device-driver abstraction to fit Android and iOS behind the same tool interface, but it takes a completely different path from ARTEMIS in the previous post: it doesn't pick 'Flash vs Pro' at the task level — it splits models 'by role,' with Manager, Executor, and FastAgent each getting their own model and temperature. This post digs into its multi-agent workflow, the Portal accessibility service, the App Card mechanism, and also states plainly that its '91.4%' benchmark badge has no locally verifiable documentation anywhere in the repo.
LLM-Driven Automated Testing Series (05): Mobile Automation Part 1 — Dissecting ARTEMIS's Dual-Mode Architecture
Google's open-source ARTEMIS lets AI assistants and test suites use real phones like a human. This post starts with what it actually is and what makes it distinctive, then digs into the Flash/Pro graph structure, the syntax-level 'verify vs assert' distinction, the two-layer pre-execution safety net, and the detail most write-ups skip — it routes element locating to a dedicated Google embodied-reasoning model, Gemini Robotics-ER, instead of letting a general-purpose VLM guess coordinates. This post tries to spell out exactly how these technical choices causally connect to its claimed AndroidWorld 99%+ score.
LLM-Driven Automated Testing Series (04): Web UI Automation — Midscene's Vision-Driven Scripting
Midscene, open-sourced by ByteDance, takes the pure-visual-locating path: no selectors at all — just hand a screenshot to a vision model and describe 'click the submit button' in natural language. This post breaks down its three model-role split (Default/Planning/Insight), aiAct's replanning loop, the distinct invalidation logic behind planning-cache and locate-cache, and where the robustness/cost trade-off actually lands compared to DOM/accessibility-tree-based approaches like Stagehand.
LLM-Driven Automated Testing Series (03): Unit Test Generation — TestGen-LLM and Qodo Cover
Tell an LLM to 'just write some tests' and you usually get code that doesn't compile, asserts something trivially true, or contributes nothing to coverage. Meta's TestGen-LLM paper doesn't try to make the model smarter — it bolts on a mechanical filter pipeline that discards anything that fails to prove its own worth. This post breaks down that filter design and how the open-source implementation, Qodo Cover, turned it into a runnable CLI tool.