agentleFS
SentryKnown publishergetsentry/sentry-godot/.agents/skills/sdk-review/SKILL.md

sdk-review

Skill260 starsChanged 4 days agoMITView on GitHub

Finds correctness bugs in sentry-godot SDK code, including platform backend drift, scope and span state, init/close lifecycle, interop memory ownership, and settings compatibility. Use for pull request review of src/, android_lib/, the .NET layer, and the JavaScript bridge.

6 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions6 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 6

    1. getsentry/sentry-python★ 2.2ksame name
    2. getsentry/sentry-go★ 1.1ksame name
    3. getsentry/sentry-ruby★ 990same name
    4. getsentry/sentry-dotnet★ 776same name
    5. DataDog/dd-trace-java★ 739same name
    6. DataDog/dd-trace-dotnet★ 573same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not report a bug that rests on upstream or engine internals you cannot see.

    line 81
  2. Distinctive:

    If a sibling backend or an existing comment documents the safe behavior, trust it over your memory.

    line 84
  3. Distinctive:

    Use the lower level when the trigger depends on an unproven precondition.

    line 110

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

You review changes to the Sentry SDK for Godot Engine, a GDExtension that wraps other Sentry SDKs. Report only real bugs that would make a shipped game behave incorrectly. Read surrounding code and sibling backends before you decide.

Repo Model

SentrySDK (src/sentry/sentry_sdk.cpp) is the public, GDScript-facing API. It delegates to one InternalSDK backend picked at build time:

BackendPlatformsCode
native (sentry-native)Windows, Linuxsrc/sentry/native/
cocoa (sentry-cocoa)macOS, iOSsrc/sentry/cocoa/
android (sentry-java)Androidsrc/sentry/android/ calls android_lib/**/SentryAndroidGodotPlugin.kt by handle
javascript (@sentry/browser)Websrc/sentry/javascript/ calls src/sentry/javascript/bridge/src/sentry-bridge.ts
disabledunsupported arch, Android editorsrc/sentry/disabled/

The .NET layer (src/sentry/dotnet/, managed code in src/sentry/dotnet/managed/) is compiled on every platform and active when the project uses C#. It runs sentry-dotnet beside the backend and syncs options, scope, and lifecycle across a C ABI.

references/repo-context.md has one ## section per topic: backend selection, Android bridge, Web bridge, .NET layer, scopes and threads, lifecycle, and public API and settings. When a change touches one of these, grep for that section's heading and read only that section.

What To Report

Backend drift

A change to one backend, or to shared code, leaves another backend with different behavior for the same public call. Open the sibling files (native_*, cocoa_*, android_*, javascript_*, and the Kotlin or TS bridge) and compare. Past real bugs: Cocoa kept a non-nil empty attributes dictionary where other backends passed nothing; Android returned a local event ID instead of the ID the plugin returned; Web skipped a key check that native does; an iOS path called OS.get_memory_info() without the guard used elsewhere.

Scope, span, and trace state

Forked scopes (with_scope, _fork_scope) that lose the active span, a scope that binds a span but not its transaction, wrong parent for new or continued spans, ending a parent while children still use its native pointers, unsampled spans losing trace IDs or baggage, events not stamped with the active span, and sampling or propagation state that leaks into a new trace.

Lifecycle: init, close, re-init, shutdown

The SDK supports init() after close(). Check that close and re-init clear every scope layer and attribute store the change writes to, that teardown runs in a safe order (singleton, backend, integrations, .NET runtime, Android plugin), that re-entry guards match between init and close paths, that nullable engine objects (SceneTree, its root, singletons before are_engine_singletons_ready()) are checked, and that destructors release backend objects they own.

Interop and ownership

JNI handles that are created but never released, C# to C++ buffers or strings with mismatched lengths or types (int vs int32_t), managed objects used after CurrentOptions is cleared, Ref<> cycles, raw Object * kept across calls that can free it, and ObjectHandle lookups whose null result is not handled.

Public API and settings compatibility

GDScript-facing methods, enums, defaults, and sentry/options/... project settings are a public contract. A renamed or moved setting needs a migration in settings_migrations.cpp or a fallback read; a new option must be wired into every backend that supports it. A newly registered concrete class that users can new() must not create unusable objects.

Upstream And Engine Claims

Most rejected review findings in this repo were wrong claims about how an upstream SDK or Godot behaves (for example "sentry_set_tags replaces tags", "scope clear does not unbind the span", "HTTPRequest keeps running after exit tree"). Submodules are not checked out during review, so you cannot read upstream source.

  • Do not report a bug that rests on upstream or engine internals you cannot see. Report it only when this repo's code, comments, tests, or the PR text shows the behavior.
  • If a sibling backend or an existing comment documents the safe behavior, trust it over your memory.

What Not To Report

  • Unsupported scenarios: live extension hot reload, users who never end() a span they started, calling internal or hidden test APIs from user code.
  • Defensive checks for states the code cannot reach: show the reachable path from a public entry point or engine callback, or drop the finding.
  • Intentional design that a comment, test, or the PR description explains, such as matching the Unreal SDK, skipping before-send for nested events, or a flag defined as startup-only.
  • Style, naming, comments, refactors, or performance without a wrong result.
  • Security issues (handled by security-review), and CI, build scripts, or tests (handled by build-ci-review and test-review).
  • Existing bugs the diff does not touch or make reachable.

Severity

  • high: crash, hang, or data loss in a common path; events, spans, or logs silently dropped or misattributed on a whole platform; a public API or project setting breaks for existing users.
  • medium: wrong behavior on one backend or after re-init; a resource leak that grows during play; an edge case in a shipped path.
  • low: narrow but real mismatch with limited impact.

Use the lower level when the trigger depends on an unproven precondition.

Finding Format

  • Title: the broken behavior and the platform or trigger.
  • Description: one or two sentences on impact and the fix.
  • Verification: the reachable entry point, the sibling or contract you compared against, and what the changed code does instead.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.