triaging-issues

# PyTorch Issue Triage Skill This skill helps triage GitHub issues by routing issues, applying labels, and leaving first-line responses. ## Contents - [MCP Tools Available](#mcp-tools-available) - [Labels You Must NEVER Add](#labels-you-must-never-add) - [Issue Triage Steps](#issue-triage-for-each-issue) - Step 0: Already Routed — SKIP - Step 1: Question vs Bug/Feature - Step 1.5: Needs Reproduction — External Files - Step 2: Transfer - Step 2.5: PT2 Issues — Special Handling - Step 3: Redirect to Secondary Oncall - Step 4: Label the Issue - Step 5: Escalate — High Priority (human review), then release triage - Step 6: bot-triaged (automatic) - Step 7: Mark Triaged - [V1 Constraints](#v1-constraints) **Labels reference:** See [labels.json](labels.json) for the full catalog of labels suitable for triage. **ONLY apply labels that exist in this file.** Do not invent or guess label names. This file excludes CI triggers, test configs, release notes, deprecated labels, and labels requiring human decision. **PT2 triage guide:** See [pt2-triage-rubric.md](pt2-triage-rubric.md) for detailed labeling guidance when triaging PT2/torch.compile issues. **Response templates:** See [templates.json](templates.json) for standard response messages. --- ## MCP Tools Available Use these GitHub MCP tools for triage: | Tool | Purpose | |------|---------| | `mcp__github__get_issue` | Get issue details and existing labels | | `mcp__github__get_issue_comments` | Get existing issue comments | | `mcp__github__update_issue` | Apply labels or close issues | | `mcp__github__add_issue_comment` | Add comment (only for redirecting questions) | | `mcp__github__search_issues` | Find similar issues for context | --- ## Labels You Must NEVER Add | Prefix/Category | Reason | |-----------------|--------| | Labels not in `labels.json` | Only apply labels that exist in the allowlist | | `ciflow/*` | CI job triggers for PRs only | | `test-config/*` | Test suite selectors for PRs only | | `release notes: *` | Auto-assigned for release notes | | `ci-*`, `ci:*` | CI infrastructure controls | | `sev*` | Severity labels require human decision | | `merge blocking` | Requires human decision | | `actionable`, `needs design`, `needs reproduction`, `needs research` | Reserved for human reviewers after they have reviewed the issue | | Any label containing "deprecated" | Obsolete | | `oncall: releng` | Not a triage redirect target. Use `module: ci` instead | **If blocked:** When a label is blocked by the hook, add ONLY `triage review` and stop. A human will handle it. These rules are enforced by a PreToolUse hook that validates all labels against `labels.json`. ### Never Override Human Labels If a human has already applied labels (especially `ci: sev`, severity labels, or priority labels), do NOT remove or replace them. Your job is to supplement, not override. --- ## Issue Triage (for each issue) ### 0) Already Routed — SKIP **If an issue already has ANY `oncall:` label, SKIP IT entirely.** Do not: - Add any labels - Add `triaged` - Leave comments - Do any triage work That issue belongs to the sub-oncall team. They own their queue. ### 1) Question vs Bug/Feature - If it is a question (not a bug report or feature request): close and use the `redirect_to_forum` template from `templates.json`. - If unclear whether it is a bug/feature vs a question: request additional information using the `request_more_info` template and stop. ### 1.5) External Files Check if the issue body contains links to external files that users would need to download to reproduce. **Patterns to detect:** - File attachments: `.zip`, `.pt`, `.pth`, `.pkl`, `.safetensors`, `.onnx`, `.bin` files - External storage: Google Drive, Dropbox, OneDrive, Mega, WeTransfer links - Model hubs: Hugging Face Hub links to model files **Action:** 1. **Edit the issue body** to remove/redact the download links - Replace with: `[Link removed - external file downloads are not permitted for security reasons]` 2. Use the `request_self_contained_reproduction` template from `templates.json` 3. Do NOT add `triaged` — wait for the user to provide a reproducible example ### 1.55) Missing Reproduction — Other Cases Request a self-contained reproduction and stop when: - The user reports a hardware-specific issue (e.g., specific GPU model) without a self-contained repro script - The user references a specific model/checkpoint/dataset that is not publicly runnable in a few lines - The issue describes version-upgrade breakage but only provides a high-level description without a minimal script - The repro depends on a specific training setup, distributed environment, or non-trivial infrastructure ### 1.6) Edge Cases & Numerical Accuracy If the issue involves extremal values or numerical precision differences: **Patterns to detect:** - Values near `torch.finfo(dtype).max` or `torch.finfo(dtype).min` - NaN/Inf appearing in outputs from valid (but extreme) inputs - Differences between CPU and GPU results - Precision differences between dtypes (e.g., fp32 vs fp16) - Fuzzer-generated edge cases **IMPORTANT — avoid keyword-triggered mislabeling:** Label based on the **root cause**, not keywords that appear in the error or title. A keyword tells you what failed, not why. - An `undefined symbol: ncclAlltoAll` error at `import torch` is a **packaging** issue (`module: binaries`), not a distributed training bug — the user never ran distributed code. - A `nan` in a parameter name or tolerance check is not `module: NaNs and Infs` unless the bug is actually about NaN propagation. - A stack trace mentioning `autograd` does not mean `module: autograd` — check whether the bug is in autograd itself or just on the call path. - A test failure with tolerance thresholds is `module: tests`, not `module: numerical-stability`. Ask: "Where would the fix need to be made?" That determines the label. **Action:** 1. Add `module: edge cases` label 2. If from a fuzzer, also add `topic: fuzzer` 3. Use the `numerical_accuracy` template from `templates.json` to link to the docs 4. If the issue is clearly expected behavior per the docs, close it with the template comment ### 2) Transfer (domain library or ExecuTorch) If the issue belongs in another repo (vision/text/audio/RL/ExecuTorch/etc.), transfer the issue and **STOP**. ### 2.5) PT2 Issues — Special Handling **PT2 is NOT a redirect.** `oncall: pt2` is not like the other oncall labels in Step 3. PT2 issues continue through Steps 4–7 for full triage — add `oncall: pt2`, then proceed to label with `module:` labels, mark `triaged`, etc. **Every `oncall: pt2` issue MUST have at least one `module:` label.** The PT2 oncall queue is too broad without a module label — the team needs to know which component is affected (e.g., `module: dynamo`, `module: inductor`, `module: helion`, `module: dynamic shapes`). If you cannot determine the specific module, use `module: compile ux` as a fallback, but always try to be specific first. See [pt2-triage-rubric.md](pt2-triage-rubric.md) for detailed guidance. ### 3) Redirect to Secondary Oncall **CRITICAL:** When redirecting issues to a **non-PT2** oncall queue, apply exactly one `oncall: ...` label and **STOP**. Do NOT: - Add any `module:` labels - Mark it `triaged` - Do any further triage work The sub-oncall team will handle their own triage. Your job is only to route it to them. #### Oncall Redirect Labels | Label | When to use | |-------|-------------| | `oncall: jit` | TorchScript issues | | `oncall: distributed` | Distributed training (DDP, FSDP, RPC, c10d, DTensor, DeviceMesh, symmetric memory, context parallel, pipelining). **Special handling:** after applying this label, invoke the distributed triage sub-skill (`/distributed-triage` on this issue) for second-level triage — it will route to a sub-oncall, add module labels, and mark triaged. | | `oncall: export` | torch.export issues | | `oncall: quantization` | Quantization issues | | `oncall: mobile` | Mobile (iOS/Android), excludes ExecuTorch | | `oncall: profiler` | Profiler issues (CPU, GPU, Kineto) | | `oncall: visualization` | TensorBoard integration | **Common routing mistakes to avoid:** - **MPS ≠ Mobile.** MPS (Metal Performance Shaders) is the macOS/Apple Silicon GPU backend. Do NOT route MPS issues to `oncall: mobile`. MPS issues stay in the general queue with `module: mps`. - **DTensor → `oncall: distributed`.** DTensor issues should always be routed to `oncall: distributed`, even if they don't mention DDP/FSDP. - **ONNX → `module: onnx`.** There is no `oncall: onnx`. Use `module: onnx` and keep in the general queue. - **CI/releng → `module: ci`.** Do not use `oncall: releng`. Use `module: ci` for CI infrastructure issues. - **torch.compile + distributed.** When `torch.compile` mishandles a distributed op (e.g., `dist.all_reduce`), the issue typically needs BOTH `oncall: pt2` and `oncall: distributed` since the fix may span both codebases. **Note:** `oncall: cpu inductor` is a sub-queue of PT2. For general triage, just use `oncall: pt2`. ### 4) Label the issue (if NOT transferred/redirected) Only if the issue stays in the general queue: - Add 1+ `module: ...` labels based on the affected area - Prefer specific labels over general ones when both exist. Check `labels.json` descriptions for guidance on when a specific label supersedes a general one (e.g., `module: sdpa` instead of `module: nn` for SDPA issues, `module: flex attention` instead of `module: nn` for flex attention). - `feature` — wholly new functionality that does not exist today in any form - `enhancement` — improvement to something that already works (e.g., adding a native backend kernel for an op that already runs via fallback/composite, performance optimization, better error messages). If the enhancement is about performance, also add `module: performance`. - `function request` — a new function or new arguments/modes for an existing function - If the issue says the operation "currently works" or "falls back to" a slower path, that is `enhancement`, not `feature` **Commonly missed labels — always check for these:** | Condition | Label | |-----------|-------| | Segfault, illegal memory access, SIGSEGV | `module: crash` | | Performance issue: regression, slowdown, or optimization request | `module: performance` | | Issue on Windows | `module: windows` | | Previously working feature now broken | `module: regression` | | Broken docs/links that previously worked | `module: docs` + `module: regression` (NOT `enhancement`) | | Issue about a test failing (not the underlying functionality) | `module: tests` | | Backward pass / gradient computation bug | `module: autograd` (in addition to the op's module label) | | `torch.linalg` ops or linear algebra ops (solve, svd, eig, inv, etc.) | `module: linear algebra` | | `has workaround` | Only add when the workaround is **non-trivial and non-obvious**. If the issue is "X doesn't work for non-contiguous tensors," calling `.contiguous()` is the tautological inverse of the bug, not a workaround. A real workaround is something like installing a specific package version, adding a synchronization point, inserting `gc.collect()`, or using a different API that isn't obviously implied by the bug description. | **Label based on the actual bug, not keywords.** Read the issue to understand what is actually broken. A bug about broadcasting that happens to mention "nan" in a parameter name is a frontend bug, not a NaN/Inf bug. ### 5) Escalate — High Priority (human review), then release triage Two independent decisions, in this order. Work through 5a first, then 5b for **every** issue — 5b is not limited to issues you escalated in 5a, and an issue can end up with both labels, one, or neither. #### 5a) High Priority — REQUIRES HUMAN REVIEW **CRITICAL:** If you believe an issue is high priority, you MUST: 1. Add `triage review` label and do not add `triaged` Do NOT directly add `high priority` without human confirmation. High priority criteria: - Crash / segfault / illegal memory access - Silent correctness issue (wrong results without error) - Regression from a prior version - Internal assert failure - Many users affected - Core component or popular model impact #### 5b) release triage — Affects an Upcoming Release Add `release triage` when an issue would affect a release if it went unfixed. The label only surfaces the issue for whoever owns the release; it is not a cherry-pick request and does not decide anything. **You are told which version is current — never guess it.** Your prompt carries a `RELEASE CONTEXT` block giving the most recent released minor version. If the block says `unknown`, skip the two version-dependent criteria below and judge the rest on their own merits. Add it when **any** of these hold: - **Regression against the most recent released minor version.** The issue reports something that worked on that minor (or later) and is broken now. Pair it with `module: regression`. If the last-working version is older than that minor, it is not release-relevant. - **Critical correctness or stability:** silent correctness (wrong results, no error), backwards-compatibility break, crash / segfault, deadlock or hang, or a large memory leak. - **Critical fix to a feature introduced in the most recent minor release** — new surface that shipped broken. Go by what the issue says ("new in 2.x", "added in 2.x", "since upgrading to 2.x") checked against the version in `RELEASE CONTEXT`; do not try to recall which features shipped in which release. - **Binary / packaging:** anything affecting wheels, Docker images, install, or the release build itself. Pair it with `module: binaries`. - **Would ship broken in the next release.** A defect on main, a nightly, an RC, or the release branch that reaches users if nobody fixes it — including anything surfaced by release validation or downstream canaries. This holds even when it is not a regression against anything, e.g. a bug in code that has never shipped. Apply it generously. A false positive costs the release manager one glance; a miss costs a broken release. When unsure, add it. `release triage` is independent of the `triage review` decision in 5a — an issue can carry both. Do not add it to feature requests, enhancements, or documentation-only issues, and do not add it for a regression against a version older than the last released minor. ### 6) bot-triaged (automatic) The `bot-triaged` label is automatically applied by a post-hook after any issue mutation. You do not need to add it manually. ### 7) Mark triaged If not transferred/redirected and not flagged for review, add `triaged`. --- ## V1 Constraints **DO NOT:** - Close bug reports or feature requests automatically - Close issues unless they are clear usage questions per Step 1 - Assign issues to users - Add `high priority` directly without human confirmation - Add module labels when redirecting to oncall - Add comments to bug reports or feature requests, except a single info request when classification is unclear **DO:** - Close clear usage questions and point to discuss.pytorch.org (per step 1) - Be conservative - when in doubt, add `triage review` for human attention - Add `release triage` whenever an issue would affect an upcoming release (step 5b); err toward adding it - Apply type labels (`feature`, `enhancement`, `function request`) when confident - Add `triaged` label when classification is complete **Note:** `bot-triaged` is automatically applied by a post-hook after any issue mutation.