remote-compute-modal

Run GPU jobs on the user's own Modal account via host.compute.create('byoc:modal', ...). Covers the create→submit→wait_for_notification flow, the compute_provider kernel for env setup, image/volume resolution, and the two approval cards. Load once you've decided to dispatch to Modal.

Quality

79%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./configs/microservice/bff-service/configs/agent-skills/claude-science/remote-compute-modal/SKILL.md

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.

Highly actionable content with excellent workflow sequencing, validation checkpoints, and error-recovery loops. Its weaknesses are length/rhetorical framing that could be tightened and progressive disclosure that inlines reference material and points to a referenced file (env-setup.md) that is not present in the bundle.

Suggestions

Move the GPU tier reference table, the Network egress deep-dive, and "Common env gotchas" into reference files under references/ and link to them, so SKILL.md stays a lean overview.

Provide the referenced env-setup.md (or remove/redirect the dangling references) so signaled references resolve to real bundle files.

Trim rhetorical section openers ("Two timeouts, one timeline", "Two surfaces, one provider") to denser leads to recover token budget.

Dimension	Reasoning	Score
Conciseness	The body is information-dense with operational detail Claude would not already know (watchdog behavior, egress fence semantics, kwarg quirks) and avoids explaining basic concepts, but it runs very long with rhetorical framing ("Two timeouts, one timeline", "Two surfaces, one provider") and architecture rationale that could be trimmed, fitting anchor 4 rather than the fully lean anchor 5.	4 / 5
Actionability	Abundant copy-paste-ready, executable code (host.compute.create, submit_job, build_env, compute_provider_config) with specific kwarg names, concrete commands, and worked examples covering the common quick-inline and full-job cases — matching anchor 5.	5 / 5
Workflow Clarity	A clear "## Workflow" sequence (read ledger → build env if absent → submit → wait_for_notification → harvest) with explicit validation checkpoints (compute_provider_config() before expensive ops, image id fails closed via Image.from_id, the "Before you finish the task" teardown checklist) and detailed error-recovery feedback loops in "When the job fails", matching anchor 5.	5 / 5
Progressive Disclosure	Sections are present and env-setup.md is signaled one-level-deep, but that referenced file is absent from the bundle (no references/ directory) and substantial reference material (latency table, GPU tier table, the deep Network egress section, Common env gotchas) is inlined into a ~790-line monolith, fitting anchor 3.	3 / 5
	Total	17 / 20 Passed

Description

73%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 specific, well-scoped description that clearly conveys both what the skill does and when to load it, with strong distinctiveness. Its main weakness is mixed voice (third-person "the user's" alongside second-person "you've decided"), which triggers the rubric's specificity penalty.

Suggestions

Rewrite in consistent third-person voice (e.g., "Loads once the agent has decided to dispatch to Modal") to avoid the second-person specificity penalty.

Make the "when" clause reference user-mentionable triggers (e.g., "Use when the user asks to run GPU jobs on Modal") to push completeness toward 5.

Add a synonym or two users naturally say (e.g., "Modal sandbox", "cloud GPU") to broaden trigger-term coverage.

Dimension	Reasoning	Score
Specificity	Names several concrete actions ("create→submit→wait_for_notification flow", "compute_provider kernel for env setup", "image/volume resolution", "two approval cards"), which sits at anchor 4, but the description uses second-person voice ("Load once you've decided"), so the rubric's voice penalty reduces it by one to 3.	3 / 5
Completeness	Explicitly answers "what" (run GPU jobs via host.compute.create('byoc:modal') covering the flow, kernel, image/volume resolution, approval cards) and "when" ("Load once you've decided to dispatch to Modal"), but the when references the agent's own decision rather than user-mentionable cues, so it is not the fully explicit trigger-phrases of anchor 5.	4 / 5
Trigger Term Quality	Good keyword coverage with natural terms a user would say ("GPU jobs", "Modal account", "Modal") plus the specific trigger "byoc:modal"; a few common variations are absent, fitting anchor 4 rather than 5.	4 / 5
Distinctiveness Conflict Risk	A clear niche scoped to the user's own Modal account via the specific 'byoc:modal' provider, explicitly distinguished from remote-compute-SSH, with minimal conflict risk — matching anchor 5.	5 / 5
	Total	16 / 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
skill_md_line_count	SKILL.md is long (790 lines); consider splitting into references/ and linking	Warning

	Total	15 / 16 Passed

Repository: UnicomAI/wanwu
Path: configs/microservice/bff-service/configs/agent-skills/claude-science/remote-compute-modal/SKILL.md
Commit: 99df6e4

Reviewed: 9 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.