About this skill

Problem

feedship-ai-daily addresses citation hallucinations in AI news digests: models may invent missing reference numbers or fabricate titles and URLs when expanding links. It separates analysis generation from source injection, so the LLM only emits placeholders and the final report gets verifiable citations.

How It Works

The skill relies on feedship v1.8.0+ to extract today's articles, then filters them with AI / Tech keywords. It reads references/prompt.md, builds a prompt from the filtered titles, and asks the model to produce a structured analysis using classification, root cause, hidden threads, and value translation. The key constraint is: use only ${N} references and never expand titles or links directly.
replace_refs.py then replaces placeholders from authoritative JSON, supports single refs like ${N} and grouped refs like ${3,7}, emits [Invalid reference #N] warnings for bad IDs, and writes the final report to /tmp/daily_report_final.md.

Scope

Best for local feedship pipelines that need a repeatable AI daily report. Check date ranges and filters when the article list is empty; if many invalid refs appear, verify the model followed the citation rules; if timeouts happen, reduce --limit and keep the filtered article list under about 100 items.

Screenshots

Use Cases

Generate an AI industry digest each morning from local `feedship` articles.
Require the LLM to emit only `${N}` placeholders, then inject real titles and links via script.
Debug fabricated citation numbers or links and review `replace_refs` invalid-reference warnings.
Filter today’s AI / Tech articles into a prompt template and produce structured trend analysis.

Best For

Engineers maintaining local `feedship` pipelines who need scheduled AI daily reports.
AI engineers building agent workflows with verifiable citation sources.
Researchers tracking LLM trends who want to turn news digests into strategic analysis.
Prompt engineers debugging LLM citation hallucinations with placeholder replacement.