Pre-Commit

Run all checks, fix any issues found, then present a summary table.

Checks

Rust checks

Run the following checks if there are changes in the tapo/, tapo-py/, or tapo-mcp/ directories. Fix all issues found. Run independent checks (cargo check, cargo clippy, cargo fmt, cargo test) in parallel. Unless otherwise specified, run checks at the workspace level (no -p flags) with --all-features to ensure cross-crate issues are caught. Link-free checks (check, clippy) take --workspace so they cover tapo-py too; cargo test must NOT — it links test binaries and examples, which fails with tapo-py's pyo3/extension-module, so it relies on default-members excluding tapo-py.

• cargo check --workspace --all-features
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• cargo fmt --all
• cargo test --all-features
• cargo clean --doc && RUSTDOCFLAGS="-D warnings" cargo doc --workspace --no-deps --all-features
• No unwrap() in non-test code without a // safe: comment
• No unsafe in non-test code without a // SAFETY: comment
• No unnecessary clones
• No deeply nested use (max one level of {} nesting)
• If a protocol puts a session token or other secret into a request URL in a new place, update redact_session_token in tapo/src/error.rs and its tests, because reqwest prints the full URL in its errors
• Module layout, for every module added or touched by the change:
• A module with submodules is declared by module_name.rs next to the module_name/ folder, never by module_name/mod.rs
• module_name.rs holds only its mod declarations and use/pub use re-exports (plus //! docs and attributes such as #[cfg(...)] on them). Move any types, functions, impls, constants, or tests into a submodule file

Python checks

Run the following checks if there are changes in the tapo/ or tapo-py/ directories. Fix all issues found.

• Update .pyi stubs if Python-exposed Rust types changed
• Verify new #[pyclass] types are imported and registered in tapo-py/src/lib.rs
• Verify Python examples in tapo-py/examples/ are updated to match corresponding Rust examples in tapo/examples/
• cd tapo-py and activate the virtual environment
• uv run mypy . — fix all type errors
• uv run black . — fix all formatting issues

MCP checks

Run the following checks if there are changes in the tapo-mcp/ directory. Fix all issues found.

• Verify that all #[derive(JsonSchema)] types have schemars annotations — including tool input params, response types, enums, and their fields/variants. Check for descriptions and range constraints where applicable.
• Verify that tapo-mcp/README.md reflects any MCP API changes (tools, resources, capabilities, env vars, auth)
• Verify that the OpenClaw skill reflects any MCP API changes. The skill spans three files — check each:
• tapo-mcp/openclaw-skill/SKILL.md — frontmatter (description, version, requires), Setup, Tools section with example npx mcporter call invocations
• tapo-mcp/openclaw-skill/references/setup.md — verification table (tool, description, parameters)
• tapo-mcp/openclaw-skill/references/tapo-mcp-setup.md — Tools table, Resources table, Configuration env vars, Authentication, Deployment (kept in sync with tapo-mcp/README.md)
• The device family words in tapo-mcp/src/server.rs with_instructions(...) and the OpenClaw description: are checked by /verify-device-support, which the Rust and Python checks below run when a model or category changes

Non-exhaustive enum checks

Run these checks if a change adds a variant to a #[non_exhaustive] enum in tapo/src (DeviceType, DiscoveryResult, ChildDeviceHubResult, ChildDeviceHubIrResult, RecordingType, …). Fix all issues found.

#[non_exhaustive] lets end users of tapo skip new variants with a _ => arm. tapo-py, tapo-mcp and the examples in tapo/examples don't get that choice: they must support every variant tapo has. Their matches end in a _ => unreachable!() arm only to compile, so the compiler no longer flags a missing variant there. The grep below lists those enums and their unreachable!() sites. Other _ => arms (the capability matches in tapo-mcp) don't show up in it and are covered by the MCP capabilities check:

grep -rn -A2 '#\[non_exhaustive\]' tapo/src/api/discovery tapo/src/responses | grep 'pub enum'
grep -rn 'unreachable!()' tapo-py/src tapo-mcp/src tapo/examples

For each new variant:

• No `unreachable!()` hits: every match that ends in _ => unreachable!() names the new variant. A variant without a capability is still named, in the arm that rejects it (for example, a device without on/off support in the WrongDeviceType arm of control_device)
• Python mirrors: enums that mirror the Rust one get the variant too, with all their per-variant methods: PyDiscoveryResult (tapo-py/src/api/discovery/discovery_result.rs), PyRecordingType and the hub child list macros. Add it to the .pyi stubs, including those for enums exposed directly, such as DeviceType
• MCP capabilities: tapo-mcp/src/requests/get_devices.rs and check_device.rs and the tools under tapo-mcp/src/tools/ give the new variant every capability it supports, and list its children if it has any

Documentation checks

Run the following checks if there are changes in the tapo/ or tapo-py/ directories. Fix all issues found.

• Run /verify-device-support when a device model or category was added or removed, a handler's public method list changed, or a method's #[cfg(feature = "debug")] gating changed. It checks SUPPORTED_DEVICES.md, the "Tested with" sentences, the builders, the handler and device info docs, the Python stubs, the examples and tapo-mcp against the code. Fix every finding it reports
• Verify that TROUBLESHOOTING.md still matches the library when the change touches protocols, authentication, discovery, error types or messages, or logging. Read the whole file, not only the sections the change seems to concern:
• Correct: every error message, log line, URL path, port, protocol name, and example file it quotes is what the code produces now. Grep each quoted string against tapo/src/ and tapo-py/ and fix the ones that drifted
• Relevant: every section describes a failure a user of the current library can still hit. Rewrite or remove sections, symptoms, and solutions about behavior the change removed. Describe only current behavior, with no notes about what older versions did
• Complete: if the change adds or rewords a failure that users must resolve on their side (a new error variant, a message that tells them to change a setting, a new protocol or firmware limitation), add it to the matching section or write a new one in the existing Symptoms / Cause / Solutions layout. Skip this when the error message already states the cause and the fix, so the guide would only repeat it: an entry has to add something the message cannot carry, such as several steps to try in order, a cause that is not obvious, or symptoms that only show in the logs

Decompiled app checks

Run these checks on every change, whatever directories it touches. Fix all issues found.

The decompiled Tapo Android app is a private reverse-engineering aid that lives outside the repo, and its names are obfuscated per app release. Nothing committed may point at it.

• Scan the files the change touches, including new untracked ones, for references to the decompiled sources, and for app class names written the Java way (Type.CONSTANT, Type.method()) or right after "app's":

  { git diff --name-only --diff-filter=d HEAD -- . ':!.claude/skills/pre-commit/SKILL.md'; git ls-files -o --exclude-standard -- . ':!.claude/skills/pre-commit/SKILL.md'; } | xargs grep -nEi '\.(java|kt|smali)\b|decompil|base\.apk|apkmirror|jadx|com[./]tplink'
  { git diff --name-only --diff-filter=d HEAD -- . ':!.claude/skills/pre-commit/SKILL.md'; git ls-files -o --exclude-standard -- . ':!.claude/skills/pre-commit/SKILL.md'; } | xargs grep -nE '`[A-Z][A-Za-z0-9]*\.([A-Z_]+|[a-z][A-Za-z0-9]*\(\))`|app(.s)?[^`]{0,30}`[A-Z][a-z]+[A-Z][A-Za-z]*`'

The second pattern works one line at a time, so it misses a class name that wraps onto the line after "app". Then read every comment in the touched files that mentions the app, not only the lines the diff changed, for what the patterns miss: two/three-character package or class names (ab1, xy2, zz9/c1), app-internal class and method names (KeyMixer.d(), StreamGatewayImpl), and absolute paths into the decompiled tree.

• Rewrite every hit to state what the app does, not where that was read. Keep wire-level names (JSON methods and fields, HTTP headers, request and response shapes as they travel); drop file, class, and line-number citations.
• Not OK: `//! The Tapo app (ab1/c.java, xy2/b.java) derives the keys from the Key-Exchange header`
• OK: `//! The Tapo app derives the keys from the Key-Exchange header`

Integration test checks

Run these checks if anything under tapo/tests/ was added or changed, including new untracked files. Fix all issues found.

The simulator stands in for a real device, so a test only proves something if the simulator behaves like the device and not like the library. Everything it knows comes from a fixture or from what real devices were seen to do.

#### Fixtures

Read every added or changed fixture in full.

• Only device answers: every key is the discovery response, a method result, a login answer keyed by sub_method, or an error response the device gave. Trimming the per-login fields the simulator computes is fine; adding a flag, setting, or hint that no device sends is not
• Folder and name: a camera or camera hub lives in fixtures/smartcam/, everything else in fixtures/smart/. The file name follows the convention in tapo/tests/simulator/fixture.rs and names the protocol the device takes. The simulator reads both, so a wrong folder or protocol changes how it behaves
• No personal data: only the device's behavior may survive the capture. Check every value that could identify a person, home, network, account, or single device:
• Ids, tokens, and keys (device_id, owner, hw_id, fw_id, oem_id, dev_id, …) are obvious placeholders: hex counting patterns such as 0123456789ABCDEF… or FEDCBA98…, all zeros, or 8000…0110 ending in the model number
• MAC addresses are AA-BB-CC-DD-EE-FF or end in a short model suffix such as -35, in every spelling the device uses (AABBCCDDEEFF in the discover login answer)
• IP addresses are 192.168.1.100
• Names are made up, like "Simulated Plug" or "Simulated Camera". Decode nickname, ssid, and every other base64 string to check; camera names such as device_name are plain text
• latitude and longitude are 0, or the London placeholder 515074 / -1278 with region Europe/London
• No email addresses, usernames, passwords, cloud tokens, user ids, room names, schedules, or other free text

Leave values that only describe the device: model, versions, component lists, signal strength, settings, and usage or energy counters.

#### Simulator

• The simulator must not use the library it tests. The only allowed hits of this grep are in assertions.rs, which checks what the library returned:

  grep -rnE '\btapo::|use tapo\b' tapo/tests/simulator/

• The simulator must not copy the library's rules either. Read every changed function that decides device behavior (protocol, password transform, port, error) and check that it follows the fixture's data, folder, or name, or what a real device did, not the branches in tapo/src/. A comment that justifies the simulator with what "the library" does is a sign of this; rewrite it to say what the device does

Code Review

After fixing all issues found in the checks, review the code changes for correctness, readability, and maintainability and propose improvements. Summarize the findings according to severity.