CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

skypilot-multi-cloud-orchestration

Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers.

Quality

74%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/cloud-compute/skypilot/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

61%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, highly actionable reference: nearly all guidance is executable code or commands with good tables and real, one-level-deep bundle references. Weaknesses are moderate redundancy (Key features vs. When to use, duplicated spot/any_of sections) and missing validation checkpoints in the batch and destructive workflows, which caps workflow clarity at 3.

Suggestions

Add explicit verification steps to the risky workflows: after the sweep loop, run 'sky jobs queue' to confirm all 10 jobs launched successfully; and note checking 'sky status' before 'sky down' to avoid terminating clusters running others' work.

Cut the 'Key features' bullet list (it duplicates 'When to use SkyPilot') and the trivial 'Hello World' section, and merge the duplicate spot-recovery and any_of examples.

Move the Sky Serve autoscaling policy detail, storage-mode table, and GPU catalog into references/advanced-usage.md, leaving the body a leaner overview with signaled links.

DimensionReasoningScore

Conciseness

The bulk is efficient — code blocks, command tables, and almost no explanation of concepts Claude already knows — but there is genuine tightening opportunity: the 'Key features' bullet list largely restates 'When to use SkyPilot', the 'Hello World' section is trivial, spot-recovery guidance appears in both 'Spot instances' and 'Managed jobs', and the any_of pattern is demonstrated twice. Not a 4 because these redundancies are more than minor trimming; not a 2 because nothing is concept-explainer padding and most sections earn their tokens.

3 / 5

Actionability

Overwhelmingly concrete and copy-paste ready: install commands with 'sky check' verification, a full task YAML schema, CLI command tables, torchrun distributed-launch snippet with SkyPilot env vars, service YAML, and two complete workflow examples. Not a 5 due to minor gaps — the 'Head-node-only execution' block wraps YAML ('run: |') inside a bash fence, and 'sky serve status my-service' is presented as 'Get endpoint', which is not what that command returns; these are small enough to keep it above the 3 anchor.

4 / 5

Workflow Clarity

Quick start has a real validation checkpoint (install, then 'sky check' before launching), but the operationally risky workflows lack verification steps: the hyperparameter sweep fires 10 managed jobs with no status/validation check, and destructive commands ('sky down', autostop with down: true) have no pre-flight confirmation guidance. The rubric explicitly caps batch/destructive workflows without validation at 3, and this content does not meet that bar.

3 / 5

Progressive Disclosure

Structure is good: clear section headers, and a 'References' section pointing to two real bundle files (references/advanced-usage.md, references/troubleshooting.md) that are one level deep, clearly signaled with scope descriptions, and themselves contain no nested references. Not a 5 because substantial advanced material that belongs in advanced-usage.md is inlined in the body (Sky Serve autoscaling policy details, storage-mode table, GPU catalog, debugging section) rather than split out; not a 3 because what is inlined is well-organized and the references that do exist are properly signaled.

4 / 5

Total

14

/

20

Passed

Description

78%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 description with an explicit 'Use when...' trigger clause covering three concrete scenarios and good natural keyword coverage. Its main deductions are the second-person voice (a one-point specificity penalty per the rubric) and missing common synonyms like provider names and 'distributed training'.

Suggestions

Rewrite the trigger clause in third person to match the rubric's voice rule, e.g. 'Use when the user needs to run training or batch jobs across multiple clouds...'.

Add commonly-said synonyms such as AWS/GCP/Azure and 'distributed training' to broaden natural trigger coverage.

DimensionReasoningScore

Specificity

The description lists several concrete actions — 'run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers' — which would otherwise land at 4, but it uses second person ('Use when you need to'), which the rubric penalizes by one point on this dimension. It is not a 4 because of that explicit voice penalty, and not a 2 because the actions named are concrete and domain-specific, not generic.

3 / 5

Completeness

It explicitly answers both questions: what ('Multi-cloud orchestration for ML workloads with automatic cost optimization') and when ('Use when you need to run training or batch jobs across multiple clouds, leverage spot instances with auto-recovery, or optimize GPU costs across providers'). The 'when' clause contains three concrete trigger conditions, matching the 5 anchor exactly; it is not a 4 because the triggers are already explicit and specific rather than merely adequate.

5 / 5

Trigger Term Quality

Good natural keyword coverage: 'training', 'batch jobs', 'multiple clouds', 'spot instances', 'GPU costs', 'ML workloads' — phrases a user with this need would plausibly say. It falls short of 5 because it omits common synonyms users actually say, such as specific provider names (AWS, GCP, Azure), 'distributed training', and 'serverless'/'on-demand GPUs'.

4 / 5

Distinctiveness Conflict Risk

The niche is clear — multi-cloud ML job orchestration with spot auto-recovery and GPU cost optimization — and unlikely to trigger for unrelated skills. Minor overlap risk remains with closely related orchestration skills (Ray, Kubernetes, Modal-style serverless GPU), which the skill body itself acknowledges as alternatives; it is not 5 because those adjacent-tool boundaries are not drawn in the description itself.

4 / 5

Total

16

/

20

Passed

Validation

81%

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

Validation — 13 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

skill_md_line_count

SKILL.md is long (519 lines); consider splitting into references/ and linking

Warning

metadata_version

'metadata.version' is missing

Warning

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

13

/

16

Passed

Repository

synthetic-sciences/openscience

Path

backend/cli/skills/cloud-compute/skypilot/SKILL.md

Reviewed

Table of Contents

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.