citation-audit

Zero-context verification that every bibliographic entry in the paper is real, correctly attributed, and used in a context the cited paper actually supports — catching hallucinated authors, wrong years, fabricated venues, version mismatches, and wrong-context citations. Use when user says "审查引用", "check citations", "citation audit", "verify references", "引用核对", or before submission to ensure bibliography integrity.

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

77%

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 a clear, well-validated workflow, but it is verbose through repetition and monolithic because detail that belongs in reference files is kept inline. Splitting the large edge-case sections into bundle files would materially improve it.

Suggestions

Move the full 'Optional: Soft-Only Mode' and 'Uncited Entry Detection' sections (including their JSON schemas and verdict translation tables) into separate reference files under references/, leaving a concise summary and a single pointed link in SKILL.md to eliminate the repeated restatements in Constants, Key Rules, and Steps.

De-duplicate the soft-only and uncited guidance that currently appears in both the Constants/Key Rules bullets and the dedicated sections — keep one authoritative location and reference it rather than re-explaining in three places.

Extract the 'Submission Artifact Emission' JSON schema and verdict decision table into references/artifact-schema.md, citing it from Step 4, so the core workflow reads as a lean overview rather than an inline spec.

Dimension	Reasoning	Score
Conciseness	The body is mostly efficient operational guidance, but the soft-only mode is restated in Constants, Key Rules, and a full 'Optional: Soft-Only Mode' section, and uncited detection is repeated across Step 2, Step 5, its own section, and Submission Artifact Emission — clear padding that could be tightened.	2 / 3
Actionability	Provides a concrete mcp__codex__codex call (model, config, sandbox) with a full reviewer prompt template, complete JSON schema examples, a copy-paste bash recompile command, and an explicit verdict decision table — actionable and specific.	3 / 3
Workflow Clarity	Seven clearly sequenced steps culminate in a recompile-and-verify validation checkpoint (Step 7), with explicit error-recovery fallbacks (bib unreadable -> BLOCKED; enumeration failure -> cited audit proceeds) and a decision-table checklist for verdicts.	3 / 3
Progressive Disclosure	No bundle files exist (references/, scripts/, assets/ absent), so large blocks that should be offloaded — soft-only mode, uncited detection, and full JSON schemas — remain inline in a monolithic SKILL.md; only one-level external shared-references links are signaled.	2 / 3
	Total	10 / 12 Passed

Description

100%

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 complete, clearly stating both capability and activation conditions in third person. It is a strong, low-conflict skill description.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'verification that every bibliographic entry... is real, correctly attributed, and used in a context' plus 'catching hallucinated authors, wrong years, fabricated venues, version mismatches, and wrong-context citations' — matching the multi-action anchor.	3 / 3
Completeness	Explicitly answers both what (zero-context bibliographic verification across three axes) and when ('Use when user says... or before submission to ensure bibliography integrity'), with explicit triggers present.	3 / 3
Trigger Term Quality	Covers natural trigger phrases a user would actually say, including bilingual variants: '审查引用', 'check citations', 'citation audit', 'verify references', '引用核对', and 'before submission'.	3 / 3
Distinctiveness Conflict Risk	Occupies a clear niche (bibliographic citation integrity) with specific, domain-locked triggers unlikely to fire for unrelated skills.	3 / 3
	Total	12 / 12 Passed

Validation

75%

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 — 12 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (503 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
relative_links	Relative link issues: 1 suspicious	Warning

	Total	12 / 16 Passed

Repository: wanshuiyin/Auto-claude-code-research-in-sleep
Path: skills/citation-audit/SKILL.md
Commit: c5f3d5b

Reviewed: about 1 hour 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.