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

85%

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

71%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 a concrete reviewer call, schemas, and decision tables, and the workflow is well sequenced with a validation step. Its weaknesses are verbosity/redundancy in the soft-only and uncited sections and a monolithic structure with no actual bundled files despite referencing shared-reference documents.

Suggestions

Consolidate the uncited-entry protocol so it is defined once (Step 2 or its dedicated section) and merely cross-referenced elsewhere; remove the duplicated explanation under Submission Artifact Emission.

Tighten the soft-only section by moving the full augmented JSON schema and report example into a bundled reference file and keeping only the verdict-translation table and invariants inline.

Add an explicit 'if recompile shows new undefined warnings → fix cite keys → re-verify' feedback loop in Step 7 so the destructive bib/body-rewrite path has a clear validation checkpoint.

Dimension	Reasoning	Score
Conciseness	Mostly efficient and free of concepts-Claude-already-knows padding, but the soft-only section (~80 lines with a full augmented JSON schema and report example) and the uncited protocol — explained in Step 2, its own top-level section, and again under Submission Artifact Emission — are redundant and could be tightened.	3 / 5
Actionability	Fully executable: an exact `mcp__codex__codex` call with model (`gpt-5.6-sol`), reasoning effort (`xhigh`), sandbox, and a complete prompt template; a concrete JSON ledger schema; real-ish example keys (`madaan2023selfrefine`); and copy-paste bash (`latexmk -C && latexmk -pdf ...`) with explicit file paths.	5 / 5
Workflow Clarity	A clear 7-step sequence with a recompile/validation checkpoint in Step 7, but the error-recovery feedback loop (fix → re-validate) is more implicit than the 'If errors: fix and re-validate' anchor, and the bib-mutation/body-rewrite path lacks an explicit post-apply re-verification before submission.	4 / 5
Progressive Disclosure	Section headers and one-level-deep, clearly-signaled `shared-references/*.md` links are present, but no bundle files exist (references/scripts/assets absent), the referenced shared-reference files are not in the bundle to verify, and substantial content (full schemas, soft-only tables) is inlined into a single monolithic SKILL.md rather than offloaded to bundled references.	3 / 5
	Total	15 / 20 Passed

Description

100%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 a strong, third-person statement that crisply states what the skill does and when to use it, with concrete failure-mode coverage and bilingual natural trigger phrases. It hits every anchor at the top level with no fluff or over-claims.

Dimension	Reasoning	Score
Specificity	Names the domain (bibliographic entry verification) and multiple concrete actions — 'is real, correctly attributed, and used in a context the cited paper actually supports' — plus an enumerated list of concrete failure modes (hallucinated authors, wrong years, fabricated venues, version mismatches, wrong-context citations), matching the comprehensive-coverage anchor.	5 / 5
Completeness	Explicitly answers both 'what' (verify every bibliographic entry is real, correctly attributed, contextually supported) and 'when' ('Use when user says ... or before submission to ensure bibliography integrity') with concrete trigger phrases.	5 / 5
Trigger Term Quality	Comprehensive natural-term coverage in two languages — '审查引用', 'check citations', 'citation audit', 'verify references', '引用核对', 'before submission' — including synonyms and bilingual equivalents a user would actually say.	5 / 5
Distinctiveness Conflict Risk	A clear niche — bibliographic citation auditing — with specific triggers and a stated sibling-skill stack (experiment-audit, result-to-claim, paper-claim-audit) that distinguishes it; minimal overlap risk.	5 / 5
	Total	20 / 20 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: 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.