Invoke when the user asks to review, check, audit, or look over Qt6 QML code -- or suggest before committing. Runs deterministic linting (47+ rules) then six parallel deep- analysis agents covering bindings, layout, loaders, delegates, states, and performance. Optionally invokes system qmllint for t

Qt QML Code Review Read only review of Qt6 QML code: a deterministic Python linter (with qmllint folded in when available) first, then agent driven deep analysis across seven rule domains, then one consolidated report. Suggest it before the user commits QML code. Scope detection Diff scope "this commit", "these changes", "the diff", "what I changed", "staged changes", "before I commit". The range is HEAD (working tree plus index), cached for staged changes only, or HEAD~1..HEAD for "this commit". Do not load the diff itself: the linter's diff mode (Phase 1) lists the changed files with their changed line counts and reports only findings on changed lines. Codebase scope "review the codebase", "audit src/", or an explicit path with no commit language. Glob .qml under the path. Token budget rules These rules exist to keep the review cheap. Follow them. 1. Do not load the reference files into the main context. Agents read references/agent rules.md themselves; you pass its absolute path and the section IDs. Only an inline run (Phase 2) reads it in the main context. Lint messages are self describing; the checklist is for humans. 2. Pass lint results in summary form (grouped by rule, paths relative to a root printed once), never the raw per line listing or qmllint's JSON. 3. Point agents at content, never paste it. Every token you write into a prompt is output on the orchestrating model, the most expensive tokens in the review. Prompts carry paths, commands and section IDs; the agents load the rules, the files and the diff. In diff scope an agent gets the git diff U50 <range <files command instead of the hunks, and reads other files only to trace a specific symbol. 4. Scale the agent count to the scope (see Phase 2). Launch no agents for a small scope and split a large one into shards of about equal line count. 5. Use the dedicated review agent where the tool has one. If the qt development skills:qml review agent definition shipped with this plugin is available, launch the agents with it (read only, tools limited to Read, Grep, Glob and read only Bash); otherwise use the tool's general purpose agent. Execution order Never skip a phase. Phase 1: Linting (Python linter + qmllint) Python 3.7+, no dependencies. If Python is unavailable, warn and continue (in diff scope, take the file list from git diff stat <range ). The linter is authoritative; do not second guess its output. Rule families: IMP, ORD, BND, LAY, LDR, DEL, STA, IMG, PRF, STY, SIG, ERR, JS (48+ checks). Exit code 1 simply means findings exist; 2 means a usage or git error. qmllint also runs qmllint and merges its warnings as QL:<id rows; add qmllint args=" I <dir " for a build directory with QML modules. The linter looks in $QT HOST PATH/bin , on PATH, in the host bin directory reported by qtpaths or qmake , and in the newest Qt 6 kit of a Qt online installer under ~/Qt , C:\Qt or /opt/Qt ( <version /<kit /bin , kits such as gcc 64 , macos , msvc2022 64 , mingw 64 ). If it reports not available but Qt is installed elsewhere (a custom installer path, or CMAKE PREFIX PATH in the project's CMakeCache.txt ), rerun once with qmllint=<that kit /bin ; it accepts the binary or its bin directory. Do not search further. Copy the printed qmllint: status into the report header. qmllint is authoritative for type level issues, the Python linter for style, ordering and performance patterns. Phase 2: Deep analysis (inline, or one agent per shard) The seven sections of references/agent rules.md are the analysis. Agents read it themselves; read it yourself only when running inline: Section Mission A1 Bindings & Properties A2 Layout & Anchoring A3 Component Loading & Lifecycle A4 ListView & Delegates A5 States, Transitions & Migration A6 Performance & Code Quality A7 Logic & Wiring Choose the fan out from the scope size. Count the lines of the files in scope (the changed lines in diff scope), taken from the lint summary or wc l . A block is about 1,500 lines; the line count alone decides the split, never the number of files. Scope Agents Up to 1,500 lines 0 (inline) More than 1,500 lines one per shard, ceil(lines / 1500) shards Inline. A small scope does not repay the cost of agents, which each rebuild the context and roughly double the review's token cost for the same findings. Do the analysis yourself: read references/agent rules.md and the files (or the diff) with the one Bash call shown under the agent prompt below, and apply A1 A7 in a single pass under the same rules, confidence thresholds and output format. Do not re report lint rows, and do not run inline above the limit because the code looks simple. Report Agents: 0 (inline) . Shards. Cut the files, in directory order, into n = ceil(lines / 1500) shards of about lines / n lines each; how many files a shard holds does not matter. Keep a directory in one shard where it fits, so that components and the files that instantiate them land together; split a larger directory by file. A file larger than a shard gets a shard to itself. Each shard agent applies all seven sections to its shard only and may Grep the whole scope to trace an id or a model role. Run up to six shard agents at a time and the rest in a second wave. Cross shard checks (ids, model roles) are then best effort; say so in the report header. Launch the agents in parallel, named after their shard (e.g. "Review shard 2/6: qml/controls"). Keep each prompt short; it contains only: 1. The absolute path of references/agent rules.md and the section IDs to apply (A1 A7) 2. The root directory from the lint summary, the files in scope relative to it and, in diff scope, the git diff U50 <range <files command for them 3. The lint summary rows (including QL: rows) for that agent's own files only (a shard agent never needs the other shards' rows), with the instruction "do not re report these lines" 4. The instruction to load everything with one Bash call before analysing, and to return findings in the format at the top of the rules file. Each tool call re sends the agent's whole context, so one batched read instead of one Read per file is the single largest saving on large scopes. Both forms print per file line numbers for the file:line locations ( cat n on several files numbers them as one run, and git diff has no line numbers): Agents are read only and must never edit files. In tools without agents, run each mission as whatever prompt chain or subprocess the tool supports; the contract is the same. Phase 3: Consolidation Merge lint and agent findings. Deduplicate on file+line+issue; when duplicates disagree on confidence, keep the block with the highest score and its trace, never average scores. Keep at most 10 investigation targets, and renumber the agents' D /I blocks consecutively. Findings below confidence 60 are suppressed. Write the report in the format below. Output format Use exactly this structure. Lint and qmllint findings are grouped by rule with one mitigation each; only deep analysis findings get a full block, copied verbatim from the agents. References references/agent rules.md agent missions A1 A7 and the agent output format (read by the agents, not the orchestrator) references/qt qml review checklist.md rationale for every linter rule (human reference, not loaded during a review) references/lint scripts/qt qml lint.py single pass linter, summary for grouped output, diff=<range for changed lines only, qmllint to merge qmllint, json for tooling Copyright (C) 2026 The Qt Company.

qt-qml-review