auto-paper-improvement-loop

Autonomously improve a generated paper via GPT-5.5 xhigh review → implement fixes → recompile, for 2 rounds. Use when user says "改论文", "improve paper", "论文润色循环", "auto improve", or wants to iteratively polish a generated paper.

Quality

—

Does it follow best practices?

Impact

—

No eval scenarios have been run

Securityby

Passed

No known issues

Quality

Content

77%

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

A highly actionable, well-sequenced workflow with strong validation feedback loops, but it is verbose for its length: the review prompt is duplicated, reviewer-independence is explained three times, and the large edit-whitelist spec lives inline instead of in a reference file.

Suggestions

Replace the duplicated Step 5 review prompt with 'use the Step 2 prompt unchanged, plus the zero-context preamble' to remove the verbatim repeat.

Consolidate the reviewer-independence guidance into one place (the Protocol section) and have the constant and Key Rules point to it instead of restating it.

Move the edit-whitelist schema/resolution/detector specification into a separate WHITELIST.md reference and keep only a short summary plus a link in SKILL.md.

Dimension	Reasoning	Score
Conciseness	Content is mostly actionable rather than explaining concepts Claude already knows, but the full review prompt is duplicated verbatim in Step 2 and Step 5, the reviewer-independence rationale is repeated in three places (REVIEWER_BIAS_GUARD constant, the Reviewer Independence Protocol section, and Key Rules), and several 'Empirical motivation' narratives pad the steps — fitting 'mostly efficient but could be tightened'.	2 / 3
Actionability	Provides copy-paste-ready bash (cp, latexmk, the section-concatenation loop, pdfinfo/grep format checks), an executable Python normalization script, concrete spawn_agent message templates, a YAML whitelist schema, and regex detectors — matching the 'fully executable code/commands; copy-paste ready' anchor.	3 / 3
Workflow Clarity	Steps 0–9 are clearly sequenced with explicit validation checkpoints (Step 4 'verify 0 undefined references/citations', Step 4.5 restatement regression test, Step 8 stop criteria) and feedback loops (recompile-after-fixes, whitelist reject→log→continue), matching the 'clear sequence with explicit validation steps' anchor.	3 / 3
Progressive Disclosure	The body is well-sectioned with clear headings (so above the monolithic anchor), but at ~570 lines it carries a 100+ line inline edit-whitelist specification and repeated review prompts that should be split into reference files, and no in-skill bundle files exist — fitting 'some structure but content that should be separate is inline'.	2 / 3
	Total	10 / 12 Passed

Description

92%

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: it is specific, gives explicit what-and-when guidance, and lists natural bilingual trigger phrases. Its only weakness is mild overlap risk from the generic 'auto improve'/'improve paper' triggers against a similar-sounding sibling skill.

Suggestions

Narrow the generic 'auto improve' trigger (e.g. 'auto improve paper' or 'auto paper improve') to reduce overlap with /auto-review-loop.

Add a one-clause distinction such as 'iteratively polish a *compiled/generated* paper (not research reruns)' to make the niche unambiguous against sibling review-loop skills.

Dimension	Reasoning	Score
Specificity	Names multiple concrete actions — 'GPT-5.5 xhigh review → implement fixes → recompile' — plus the model, effort level, and a fixed round count ('for 2 rounds'), matching the 'lists multiple specific concrete actions' anchor.	3 / 3
Completeness	Clearly states what ('Autonomously improve a generated paper via GPT-5.5 xhigh review → implement fixes → recompile, for 2 rounds') and when ('Use when user says...'), satisfying the explicit what-AND-when-with-triggers anchor.	3 / 3
Trigger Term Quality	Explicit natural trigger phrases users would actually say are listed verbatim — '改论文', 'improve paper', '论文润色循环', 'auto improve', 'iteratively polish a generated paper' — covering both English and Chinese variations, matching the 'good coverage of natural terms' anchor.	3 / 3
Distinctiveness Conflict Risk	The 'generated paper' / 'iteratively polish' niche and the Chinese triggers are distinctive, but the generic 'auto improve' and 'improve paper' triggers could overlap with the sibling /auto-review-loop described in the body, so it sits at 'somewhat specific but could still overlap' rather than the clear-niche anchor.	2 / 3
	Total	11 / 12 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
skill_md_line_count	SKILL.md is long (575 lines); consider splitting into references/ and linking	Warning
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

	Total	13 / 16 Passed

Repository: wanshuiyin/Auto-claude-code-research-in-sleep
Commit: fe5963c

Reviewed: about 18 hours 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.