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

81%

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

63%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 content is highly actionable with concrete reviewer prompts, executable commands, and explicit schemas, and the workflow is well-sequenced with validation gates. Its main weakness is token efficiency: significant redundant restatement of the same invariants and large inlined opt-in sections bloat what could be a much leaner file.

Suggestions

Consolidate the repeated invariants (fresh reviewer thread, REPLACE/REMOVE approval, always-emit-never-block, soft-only no-bib-mutation) into a single canonical statement in Key Rules and reference it elsewhere instead of restating each 3–5 times.

Move the detailed soft-only mode spec and the full JSON schema blocks into one-level-deep reference files (e.g. references/soft-only.md, references/artifact-schema.md), keeping only a short summary and link in SKILL.md.

Make the Step 7 validation loop explicit ("if Citation/Reference undefined warnings appear, fix the affected keys and re-run latexmk until clean") so the feedback loop matches the destructive/batch cap guidance.

Dimension	Reasoning	Score
Conciseness	At ~500 lines the body restates the same invariants repeatedly — the fresh-reviewer-thread, REPLACE/REMOVE-approval, always-emit-never-block, and soft-only no-bib-mutation guarantees each appear 3–5 times across Constants, Key Rules, Workflow, and Submission Artifact Emission — which is noticeably verbose padding even though the content is accurate.	2 / 5
Actionability	Provides a copy-paste Codex MCP reviewer prompt with model/config/sandbox, an executable `latexmk` recompile command, concrete JSON ledger examples (madaan2023selfrefine), and a full verdict decision table — fully executable guidance covering the common cases.	5 / 5
Workflow Clarity	A clear 7-step sequence with explicit checkpoints (per-entry reviewer verdicts, aggregation decision table, interactive Step 6 approval gate, Step 7 recompile-and-verify), but the Step 7 validate→fix→retry loop is only implied ("Confirm: No new warnings") rather than explicitly stated, a minor validation gap.	4 / 5
Progressive Disclosure	Good section structure and external protocols are delegated to shared-references files, but the file is monolithic with no bundle files present; large opt-in sections (soft-only ~120 lines, full JSON schemas, uncited detection) are inlined that could be split into one-level-deep reference files.	3 / 5
	Total	14 / 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 excellent: it concisely states what the skill does, enumerates concrete failure modes it catches, and gives explicit bilingual trigger phrases plus a timing cue. It cleanly matches the top anchor on every dimension with no over-claims or fluff.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete verification actions (verify real, correctly attributed, supports context) and a comprehensive catch list (hallucinated authors, wrong years, fabricated venues, version mismatches, wrong-context citations), matching the comprehensive-coverage anchor.	5 / 5
Completeness	Explicitly answers both what ("Zero-context verification that every bibliographic entry...") and when ("Use when user says... or before submission"), with concrete trigger phrases — a textbook match for the top anchor.	5 / 5
Trigger Term Quality	Comprehensive natural trigger coverage including bilingual user phrases ("审查引用", "引用核对", "check citations", "citation audit", "verify references") plus the "before submission" context, hitting synonyms and natural wording users would actually say.	5 / 5
Distinctiveness Conflict Risk	Clear bibliographic-verification niche with specific, narrow triggers; sibling audit skills target different domains (code, numerical claims), so conflict risk is minimal despite the shared "audit" family naming.	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: 0472e53

Reviewed: 5 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.