experiment-queue

SSH job queue for multi-seed/multi-config ML experiments with OOM-aware retry, stale-screen cleanup, and wave-transition race prevention. Use when user says "batch experiments", "队列实验", "run grid", "multi-seed sweep", "auto-chain experiments", or when /run-experiment is insufficient for 10+ jobs that need orchestration.

Quality

88%

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

Quality

Content

81%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.

The body is highly actionable with executable code, a clearly sequenced workflow, and explicit validation/recovery loops for a batch orchestration skill. Its main weakness is conciseness: narrative rationale and legacy-resolver commentary add tokens without adding instructional value, and a few inlined algorithm sections could live in reference files.

Suggestions

Trim or fold the 'Why This Exists' and 'Rationale / Source' sections — the session-audit narrative and post-mortem attribution ('Codex GPT-5.5 xhigh') do not guide execution and can be dropped or moved to a one-line note.

Condense the Step 3a legacy resolver-chain prose; the four-layer fallback is already captured in the bash snippet, so the surrounding paragraph largely restates it.

Consider moving the detailed algorithm internals (OOM Handling, Stale Screen Detection, Resume-on-restart) into a short reference file linked from the Workflow section to reduce the inline body length.

Dimension	Reasoning	Score
Conciseness	Core instructional content is concrete and efficient, but several sections pad the token budget: the 'Why This Exists' and 'Rationale / Source' narratives (including the post-mortem attribution 'Codex GPT-5.5 xhigh') and the lengthy legacy resolver-chain commentary in Step 3a restate or contextualize rather than instruct.	3 / 5
Actionability	Fully executable guidance throughout: copy-paste bash for path binding, scp, nohup launch, and jq-based monitoring; concrete YAML manifests; a real OOM regex; and specific state-machine/stale-screen detection steps covering the common orchestration cases.	5 / 5
Workflow Clarity	Steps 1-5 are clearly sequenced with explicit validation checkpoints (Step 2 pre-flight precondition checks with 'show user which jobs are blocked and why') and error-recovery feedback loops (bounded OOM retry, the instruction to delete a bad pre-2026-08 state file and re-run).	5 / 5
Progressive Disclosure	Good structure with clear headers, one-level-deep signaled references to scripts/queue_manager.py and scripts/build_manifest.py (verified present), and well-placed See Also pointers. Not a 5 because the ~430-line body inlines several substantial algorithmic sections (Grid Spec, Wave Chaining, OOM, Stale Screen, Resume) and references an external shared-references/ path outside the bundle.	4 / 5
	Total	17 / 20 Passed

Description

95%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.

The description is specific, trigger-rich, and clearly distinguishes itself from /run-experiment, explicitly covering both what the skill does and when to use it. Only minor: the capability list leans on reliability features rather than a fully comprehensive enumeration of all functions.

Dimension	Reasoning	Score
Specificity	Names the domain and lists several concrete capabilities ('OOM-aware retry, stale-screen cleanup, and wave-transition race prevention'), fitting anchor 4 rather than 5 because the actions cluster around orchestration reliability rather than comprehensively enumerating every capability.	4 / 5
Completeness	Explicitly answers both what ('SSH job queue ... with OOM-aware retry, stale-screen cleanup, and wave-transition race prevention') and when ('Use when user says ... or when /run-experiment is insufficient for 10+ jobs') with concrete trigger phrases.	5 / 5
Trigger Term Quality	Comprehensive natural trigger coverage including synonyms and a bilingual term: "batch experiments", "run grid", "multi-seed sweep", "auto-chain experiments", and "队列实验" — these are phrases users would naturally say.	5 / 5
Distinctiveness Conflict Risk	A clear niche (SSH ML-experiment orchestration) with distinct triggers and an explicit contrast against the most likely competing skill (/run-experiment), giving minimal conflict risk.	5 / 5
	Total	19 / 20 Passed

Validation

81%

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 — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
allowed_tools_field	'allowed-tools' contains unusual tool name(s)	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
relative_links	Relative link issues: 1 suspicious	Warning

	Total	13 / 16 Passed

Repository: wanshuiyin/Auto-claude-code-research-in-sleep
Path: skills/experiment-queue/SKILL.md
Commit: f4f20f9

Reviewed: 1 day 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.