Agent Skills › Prismer-AI/PrismerCloud › evidence-citations

evidence-citations

GitHub

构建基于来源的研究答案和文档，通过任务隔离的引用账本、验证过的引文和明确的证据缺口管理，确保事实准确性。

sdk/cloud/catalog/skills/evidence-citations/SKILL.md Prismer-AI/PrismerCloud

Trigger Scenarios

需要进行事实核查的研究任务生成带有明确引用的报告或比较分析

Install

npx skills add Prismer-AI/PrismerCloud --skill evidence-citations -g -y

More Options

Non-standard path

npx skills add https://github.com/Prismer-AI/PrismerCloud/tree/main/sdk/cloud/catalog/skills/evidence-citations -g -y

Use without installing

npx skills use Prismer-AI/PrismerCloud@evidence-citations

指定 Agent (Claude Code)

npx skills add Prismer-AI/PrismerCloud --skill evidence-citations -a claude-code -g -y

安装 repo 全部 skill

npx skills add Prismer-AI/PrismerCloud --all -g -y

预览 repo 内 skill

npx skills add Prismer-AI/PrismerCloud --list

SKILL.md

Frontmatter

{
    "name": "evidence-citations",
    "scope": "common",
    "license": "MIT",
    "metadata": {
        "upstream": "Hermes Agent skills\/research\/grounded-citations at 1a1f4a59e2",
        "nativeReplaces": [
            "grounded-citations"
        ]
    },
    "description": "Build source-backed research answers and documents with a task-isolated citation ledger, verified quotations, and explicit evidence gaps."
}

Evidence Citations

Use for research, comparisons, reports, and factual document generation. Retrieve through the available ingest/browser tools; a URL alone is not evidence. Read the actual source and distinguish its claims from your inference.

Resolve scripts/sources.py relative to this installed skill directory. Always pass an absolute, task-specific --ledger <scratch-dir>/citations.json. For Hermes, obtain the current scratch directory from execution context. Never reset a profile-global ledger or another task's ledger. Resume an existing ledger when continuing the same task.

Register retrieved URLs with sources.py --ledger <path> add <url> --title <title>.
Store fetched text in task scratch. Attach exact supporting quotations with quote <id> --text <quote> --from <source-file>; fabricated quotes fail.
Write a Markdown draft using the returned numeric citation IDs. Unknown facts stay unresolved; preserve contradictory sources.
Append the source list from render --cited-in <draft> to the draft.
Run verify <draft> --strict --evidence. This checks citation consistency and attached quotations, not whether a source is truthful or entails a claim. Add --min-coverage 1 when every claim needs declared provenance; CJK claims are counted, and an unmeasurable draft fails rather than reporting coverage. Only numeric [n] Markdown is verified; footnotes/LaTeX require their own checks.
For PKF, map verified ledger URLs to the citation format in pkf-writing; run the PKF validator separately. Do not claim the Markdown checker validates PKF.
Deliver requested files through assets/office-artifacts and cloud deliver. A short chat answer can cite the same verified sources inline.

No credentials are needed by the ledger. Missing retrieval tools are a coverage blocker, not a reason to fabricate sources. See references/citation-formats.md and references/grounding-rationale.md for background; this task-isolation and Prismer delivery contract takes precedence over upstream host-specific examples.

render --replace-in replaces only one unambiguous terminal Sources section and writes atomically. Fenced examples remain untouched; a non-terminal or duplicate heading fails without modifying the draft. A lock timeout is a failure, not permission to delete another writer's lock or continue without exclusivity.