warp-debug-gradients

Use to diagnose and fix incorrect gradients in differentiable Warp programs. Anything trained, optimized, calibrated, or fit through Warp kernels depends on wp.Tape gradients, so treat any misbehavior of such a workflow as a gradient problem until proven otherwise — use this when training diverges or NaNs, won't train at all, stalls or plateaus above the expected loss, converges to a wrong or biased answer, is worse than a reference implementation, works at small scale but fails at production scale, or fails a QA/validation recheck. Also for explicit symptoms — exploding, NaN/inf, zero, or subtly wrong gradients, suspected wp.Tape/backward issues, gradcheck failures — but users usually describe only the surface symptom ("the sim explodes", "the fit gets dragged toward outliers") without mentioning gradients: make that leap. Not for forward-only Warp work, build/install problems, or autograd issues in other frameworks without Warp.

Quality

87%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

85%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured diagnostic workflow with strong validation gates, a signature-driven decision table, and clean progressive disclosure into four genuine reference files. The body leans on references for the executable harness, which is appropriate but leaves a minor actionability gap.

Dimension	Reasoning	Score
Conciseness	Efficient and assumes Claude's competence — it gives Warp-specific caveats (tape behaviors, version gating) Claude would not already know rather than explaining gradients/tapes in general. A few passages (the examples walkthrough, some step prose) could be tightened.	4 / 5
Actionability	Concrete executable signals are present (`wp.config.verify_autograd_array_access = True`, `wp.autograd.gradcheck`, `wp.autograd.gradcheck_tape`, `wp.__version__`) with a structured signature table, but the full copy-paste FD harness is delegated to references/verification.md rather than inlined.	4 / 5
Workflow Clarity	An explicit 8-step sequence with hard validation checkpoints — "measure before hypothesizing", before/after FD comparison required ("A gradient fix without a before/after FD comparison is not a fix"), re-verify the shipped file, and close the loop on the original symptom.	5 / 5
Progressive Disclosure	Clear overview in SKILL.md with one-level-deep references to four real files (quick-checks.md, verification.md, custom-gradients.md, case-studies.md), each cited inline at the relevant step and listed in a dedicated "Reference files" section.	5 / 5
	Total	18 / 20 Passed

Description

90%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, third-person description with comprehensive natural trigger phrasing, clear what/when coverage, and sharp scope boundaries. The only soft spot is action specificity — it lists two verbs (diagnose, fix) rather than a richer action inventory.

Dimension	Reasoning	Score
Specificity	Names the domain ("differentiable Warp programs") and two concrete actions ("diagnose and fix incorrect gradients"), but the action set itself is minimal — it enumerates many symptoms rather than additional distinct capabilities.	3 / 5
Completeness	Explicitly answers both what ("diagnose and fix incorrect gradients in differentiable Warp programs") and when (a long list of trigger symptoms plus explicit "Not for..." exclusions) with concrete trigger phrases.	5 / 5
Trigger Term Quality	Comprehensive natural-language triggers users actually say — "training diverges or NaNs", "the sim explodes", "the fit gets dragged toward outliers", "gradcheck failures" — including surface-symptom synonyms, not just jargon.	5 / 5
Distinctiveness Conflict Risk	Narrow Warp-specific niche with explicit exclusions ("Not for forward-only Warp work, build/install problems, or autograd issues in other frameworks without Warp"), giving minimal overlap risk.	5 / 5
	Total	18 / 20 Passed

Validation

93%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_field	'metadata' should map string keys to string values	Warning

	Total	15 / 16 Passed

Repository: NVIDIA/warp
Path: skills/warp-debug-gradients/SKILL.md
Commit: 238de60

Reviewed: 5 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.