Trust the authoritative marker, not the plausible guess. In a Jeopardy!/Q&A transcript, a named speaker's response (e.g., "CORDRAY: What is gravity?") is often a WRONG guess; the real answer follows the "A:" marker. Extract the "A:" value, and discount conflicting forum posts when independent sources agree.
Resolve the prompt's actual answer type. When the "Question" is a bare declarative statement with an unresolved pronoun ("his", "her", "their") and no question word, it is not a True/False claim. The expected short answer is the named entity the pronoun refers to — resolve it against the context and emit that entity's name.
Match grammatical number and article to the prompt's phrasing. A blank introduced by "one of these [plural category]" ("...one of the ... holes") wants the singular count noun with its indefinite article ("a sinkhole"), not a bare plural ("sinkholes").
Before committing, check three things: what answer type the prompt requests (entity vs. boolean), which context region is authoritative vs. distractor, and what exact surface form (article/number) the key expects. Do not let the most salient or plausible token override these.
Answer Source Form Verification
Determine the authoritative source, answer type, and exact surface form before committing an exact-match short answer. Use when the retrieved context interleaves a speaker's guess with an 'A:' answer, when the prompt labelled 'Question' is a bare declarative statement with an unresolved pronoun, or when the prompt's 'one of these [plural]' phrasing dictates a singular noun with article.
Answer Source Form Verification by ruc-nlpir · f062a82
npx get-skill add ruc-nlpir/answer-source-form-verification File contents
---name: answer-source-form-verificationdescription: Determine the authoritative source, answer type, and exact surface form before committing an exact-match short answer. Use when the retrieved context interleaves a speaker's guess with an 'A:' answer, when the prompt labelled 'Question' is a bare declarative statement with an unresolved pronoun, or when the prompt's 'one of these [plural]' phrasing dictates a singular noun with article.flashrsi: algorithm: evo_harness level: general source_task_ids: [64cb77e3c59a40e1ae3a9502bd565867, 6072e63558b445ad9c01394dfed313b4, f6f35af39efd4a16ba440993d142349e]metadata: source: https://skillmd.com/skills/ruc-nlpir/answer-source-form-verification---1. Trust the authoritative marker, not the plausible guess. In a Jeopardy!/Q&A transcript, a named speaker's response (e.g., "CORDRAY: What is gravity?") is often a WRONG guess; the real answer follows the "A:" marker. Extract the "A:" value, and discount conflicting forum posts when independent sources agree.2. Resolve the prompt's actual answer type. When the "Question" is a bare declarative statement with an unresolved pronoun ("his", "her", "their") and no question word, it is not a True/False claim. The expected short answer is the named entity the pronoun refers to — resolve it against the context and emit that entity's name.3. Match grammatical number and article to the prompt's phrasing. A blank introduced by "one of these [plural category]" ("...one of the ... holes") wants the singular count noun with its indefinite article ("a sinkhole"), not a bare plural ("sinkholes").4. Before committing, check three things: what answer type the prompt requests (entity vs. boolean), which context region is authoritative vs. distractor, and what exact surface form (article/number) the key expects. Do not let the most salient or plausible token override these.
ruc-nlpir/flashrsi commit f062a82f55
Frequently asked questions
Run npx get-skill add ruc-nlpir/answer-source-form-verification in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.
Determine the authoritative source, answer type, and exact surface form before committing an exact-match short answer. Use when the retrieved context interleaves a speaker's guess with an 'A:' answer, when the prompt labelled 'Question' is a bare declarative statement with an unresolved pronoun, or when the prompt's 'one of these [plural]' phrasing dictates a singular noun with article. It is listed under AI & ML on SkillMD.
This skill has not completed SkillMD's automated safety review yet. Capability flags: reads secrets. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.
This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.
Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.
ruc-nlpir (@ruc-nlpir) published this skill. Their other Agent Skills are listed on their SkillMD profile.