skills.rest
kai-kou
kai-kouCOMMUNITY
Explore all skills by this author →
Website𝕏@k_aik_ouGitHub

research-runner

Automates deep research generation with fallback engines, quality gates, and PR creation.

Updated Aug 17, 2026
Search Tags:#automation#pull request#github#web search#quality gate#fallback#deep research
One-click install
npx skills add https://github.com/kai-kou/gem-hunter --skill research-runner-kai-kou
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: research-runner
Source: https://github.com/kai-kou/gem-hunter/tree/main/.claude/skills/research-runner
Command: npx skills add https://github.com/kai-kou/gem-hunter --skill research-runner-kai-kou

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Manually running deep research, validating output quality, and creating pull requests is repetitive and error-prone. This Skill fully automates the pipeline from a research prompt file to a cited, structured research report with quality checks and automated PR merge. ## Core Features & Use Cases - Multi-engine fallback chain: Runs the native /deep-research workflow first, falls back to a claude -p subprocess runner, then to a DIY web research mode (WebSearch/WebFetch) only on actual failure. - Quality gate enforcement: Validates output against a JSON schema with thresholds for fact-check flags, source rank, section count, and citation count, retrying once before escalating. - End-to-end GitHub automation: Locks issues via labels, commits results, opens PRs, monitors AI reviews, and auto-merges on approval. - Use Case: A maintainer labels an issue phase:research and status:waiting-claude; the Skill generates content/research/{ID}_deep_research.md and .json, opens a PR, and merges it after review without manual research work. ## Quick Start Ask the AI to run a deep research on a topic, for example: run a deep research on competitor pricing models and generate the structured report.

Frequently Asked Questions about research-runner

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automate deep research report generation?▼

Provide a prompt file at content/research/{ID}_prompt.md or ask directly in conversation. The Skill runs the native /deep-research workflow, normalizes output to a JSON schema, applies quality gates, and opens a pull request automatically.

What happens when the deep research workflow fails or hits rate limits?▼

Rate-limit failures (EXIT=6) trigger a skip counter stored as issue labels, retrying up to three consecutive times before falling back. True failures fall back to a DIY web research mode using WebSearch and WebFetch with parallel sub-agents.

Does this Skill use external LLM APIs like Gemini for research?▼

No, external LLM APIs such as Gemini are explicitly prohibited by project decision. Research runs only through the native /deep-research workflow, a claude -p subprocess, or an in-session DIY web research fallback.

How are research quality standards enforced automatically?▼

A quality gate checks the output against research-rules.md and research_schema.json, requiring at least 5 sections, 8 sources, no rank-C citations, and fewer than 5 fact-check flags. Failures trigger one retry before escalating to the user.

Can I trigger deep research interactively without an issue or prompt file?▼

Yes, ad-hoc interactive requests are supported. The Skill generates a prompt file from your instruction text, decomposes it into 5-7 investigation items, and proceeds directly to the primary research engine without stopping.

Why does the Skill commit the raw research report immediately?▼

Uncommitted files can be lost when sessions restart, so the raw report is committed and pushed before normalization. If normalization fails, the saved raw file allows re-normalization at zero additional research cost.

Related Skills

Explore More Education & Research Skills→
MohmmedKudmaniMohmmedKudmani

research-topic

Researches topics and saves structured summaries with content angles to dated Markdown files.

Community
Basic
pmcdowallpmcdowall

blogwatcher

Monitor RSS/Atom feeds and collect newly published articles.

Community
Intermediate
vankichivankichi

claude-session-jsonl

Parse Claude Code session logs to compute token usage, cost, and tool activity.

Community
Advanced
collat-labscollat-labs
1

tavily-research

Run Tavily CLI deep research and return citation-grounded reports.

Official
Intermediate