citation-injector_skill

Overview

This skill applies a citation-diversifier budget report by injecting in-scope citation keys into an existing draft so the project meets a global unique-citation target. It edits the draft in place without adding new facts and produces a short injection report describing changes. The goal is to close citation gaps and reduce over-reuse while avoiding citation-dump voice and invented claims.

How this skill works

The skill reads the citation budget report and the draft outline to locate H3 subsections with suggested keys, then inserts short, evidence-neutral citation sentences using approved injection patterns. It only uses keys present in the provided bibliography and never alters substantive claims or adds new numbers. After edits, it writes a PASS/FAIL injection report that lists which H3s were touched and which keys were added.

When to use it

• When output/CITATION_BUDGET_REPORT.md exists and reports a positive global gap
• Before running draft-polisher or pipeline-auditor to meet the unique-citation gate
• When many subsections reuse the same few citation keys and you need more unique keys
• When citations/ref.bib contains the candidate keys to inject and you do not need new upstream literature search

Best practices

• Only inject keys listed for that H3 in the budget report; do not invent or substitute keys
• Prefer short, axis-anchored or parenthetical sentences that connect to the subsection lens
• Avoid placing a bracketed list as the only sentence in a paragraph (no citation dumps)
• Favor keys that are globally unused to maximize diversification; avoid reusing the same new keys across many H3s
• Insert the injection early in the subsection (often after paragraph 1) so it reads as positioning, not an afterthought
• Rerun the validator script to confirm the global target before finishing and then run draft-polisher to smooth voice

Example use cases

• A draft has many in-paragraph citations but only three unique keys; inject additional unique keys per the budget to pass the pipeline gate
• Several subsections reuse a classic survey paper; add different representative implementations per subsection to reflect contrasting design points
• A pipeline run flags ‘unique citations too low’; apply the budget report to add scoped citations and produce a PASS injection report
• Before final polishing, distribute unused keys from ref.bib into relevant H3s to reduce over-reuse and improve evidence spread