nemo-mbridge-resiliency

Resiliency features in Megatron Bridge including fault tolerance, straggler detection, in-process restart, preemption, and re-run state machine.

Quality

85%

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

82%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 highly actionable, lean reference skill with copy-paste-ready configuration for every resiliency feature and a solid verification section providing validation checkpoints. The main gaps are the catalog-style (rather than sequenced-workflow) structure with limited error-recovery feedback loops, and some inline parameter tables that could be split into reference files.

Suggestions

Add a brief validate-fix-retry feedback loop for the non-fault-tolerance features (e.g., straggler detection, in-process restart) in the Verification section, so each feature states what to do when the expected log signals do not appear.

Consider moving the larger parameter tables (e.g., InProcessRestartConfig's 12 rows) into a referenced reference file and keeping only the most-tuned knobs inline, to improve progressive disclosure and reduce SKILL.md token weight.

Add a short 'Choose a feature' decision pointer at the top of Enablement so users land on the right subsection for their scenario (Slurm fault tolerance vs. straggler detection vs. preemption) without scanning the whole catalog.

Dimension	Reasoning	Score
Conciseness	The body is dense and reference-style with no padding or explanation of concepts Claude already knows (no 'what is NCCL' or 'what is fault tolerance' prose); scored 4 rather than 5 because the volume of inline parameter tables and a few slightly verbose table-cell descriptions leave minor trim opportunities.	4 / 5
Actionability	Every feature ships copy-paste-ready code with real import paths ('from megatron.bridge.training.config import FaultToleranceConfig'), complete launch commands with all flags, required env exports, and a documented exit-code table; the examples cover all common cases.	5 / 5
Workflow Clarity	A dedicated Verification section provides explicit per-feature validation checkpoints with expected log signals ('Look for [FaultTolerance] / [RankMonitorServer] log lines', 'Simulated fault should trigger restart from checkpoint'); scored 4 not 5 because the body is a feature catalog rather than a single sequenced workflow, and explicit validate-fix-retry feedback loops are only present for fault tolerance.	4 / 5
Progressive Disclosure	Clear top-of-file overview references ('Stable docs: @docs/training/resiliency.md', 'Card: @skills/.../card.yaml'), well-organized sections (Enablement, Code Anchors, Pitfalls, Verification), and one-level-deep pointers to source files and example scripts; scored 4 not 5 because several full parameter tables are inlined rather than referenced out to separate reference files.	4 / 5
	Total	17 / 20 Passed

Description

87%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, specific description paired with a dedicated when_to_use field that together clearly answer both what the skill does and when to invoke it, using natural domain trigger terms with minimal conflict risk. The only minor weakness is the feature-noun framing rather than action verbs.

Dimension	Reasoning	Score
Specificity	The description names the domain ('Resiliency features in Megatron Bridge') and enumerates five concrete capabilities ('fault tolerance, straggler detection, in-process restart, preemption, and re-run state machine'), giving comprehensive coverage of the resiliency surface; scored 4 rather than 5 because the list is feature nouns rather than concrete action verbs.	4 / 5
Completeness	The description answers 'what' explicitly and the when_to_use field provides equivalent explicit trigger guidance ('Enabling resiliency features, or investigating a commit that caused training hangs, straggler detection failures, or broken restarts') with concrete trigger phrases, satisfying the 'what AND when' requirement; not below 5 because both are present and concrete.	5 / 5
Trigger Term Quality	The description plus the companion when_to_use field surface strong natural keywords a Megatron user would say ('fault tolerance', 'straggler detection', 'preemption', 'hang detection', 'automatic restart', 'in-process restart', 'nvidia-resiliency-ext'); scored 4 not 5 because a few natural variations/synonyms are still absent and there is no file-extension-style trigger.	4 / 5
Distinctiveness Conflict Risk	The scoped niche ('Megatron Bridge' resiliency) plus highly specific trigger terms ('nvidia-resiliency-ext', 'in-process restart', 'straggler detection') make this clearly distinguishable with minimal conflict risk against other skills.	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
referenced_paths_exist	Referenced path issues: 1 missing, 1 deeper-than-1-level	Warning

	Total	15 / 16 Passed

Repository: NVIDIA/skills
Path: skills/nemo-mbridge-resiliency/SKILL.md
Commit: 73a7813

Reviewed: 3 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.