Vibe Research

Conduct AI-led research with autonomous literature review, hypothesis generation, analysis, and synthesis while human provides vision.

Security Scan

The skill mostly matches its stated purpose (AI-led research) but contains internal contradictions and omissions—most notably claims it can run 'offline' and 'without external dependencies' while the instructions explicitly call web_search/web_fetch and large-scale source retrieval—so the documentation is inconsistent and warrants caution before installing.

Purpose & Capability

The skill's name, description, and instructions consistently describe an autonomous literature-review and analysis pipeline. Asking the agent to run web_search/web_fetch and assemble reproducibility artifacts is appropriate for the stated purpose. However, README and other files claim 'No external dependencies' and 'Works offline,' which contradicts the repeated use of network fetch tools and 100+ source retrieval—this is an important incoherence to resolve.

Instruction Scope

SKILL.md, pipeline.md, example.md and quickref instruct the agent to perform broad web_search and web_fetch cycles (count=20, 100+ sources minimum), to 'pull additional sources' proactively, and to 'execute ALL cycles' autonomously. The instructions do not tell the agent to read unrelated local files or environment secrets, which is good, but they grant broad discretion to fetch and aggregate large amounts of external content—this is expected for research but increases privacy and data-exfiltration risk if the agent is given access to sensitive inputs (e.g., PHI). The offline/no-external-dependency claim in README conflicts with these explicit network operations.

Install Mechanism

Instruction-only skill with no install spec, no binaries, and no code files. No on-disk installation or archive downloads are requested, which is low-risk from an install vector perspective.

Credentials

The skill declares no required environment variables, no credentials, and no config paths. That aligns with the skill's claimed behavior (uses platform search/fetch tools). There are no obvious requests for unrelated secrets or platform tokens.

Persistence & Privilege

always:false and default autonomous invocation are set. The skill does not request permanent presence or elevated privileges, nor does it instruct modifying other skills or global agent settings.

What to consider before installing

This skill appears to be a legitimate autonomous research workflow, but there are documentation contradictions and operational risks you should address before installing: - Confirm network requirements: the README claims 'Works offline' and 'No external dependencies' but the instructions repeatedly call web_search/web_fetch and mandate fetching many sources. Ask the skill author or maintainer whether the agent will actually perform online searches and whether it can function without network access. - Data-sensitivity check: avoid giving the agent any sensitive inputs (PHI, proprietary documents) unless you fully trust how the platform's web_search/web_fetch tools handle and store fetched data; the skill is designed to proactively pull external sources which could cause unintended data transmission. - Human checkpoints: enforce the three mandatory stop points in practice (clarification, plan approval, final report) and consider adding explicit confirmation steps before starting large automated fetch/analysis cycles. - Reproducibility/storage: clarify where the 'reproducibility package' and fetched sources are stored, how long they are retained, and who can access them. - Rate/resource controls: because the skill instructs many web searches and fetches, limit or monitor the number of cycles and network requests to avoid unexpected bandwidth, cost, or scraping of third-party sites. - Validate citations and provenance in outputs: the skill mandates APA citations and cross-verification—spot-check early outputs for proper sourcing and accuracy. If the author can explain/resolve the offline/no-dependency claims and confirm safe handling of fetched data, the inconsistencies would be addressed. Until then, proceed cautiously and avoid exposing sensitive data to the skill.

When to Use

User has a research question or knowledge gap. Agent takes ownership of the full research cycle: scanning literature, generating hypotheses, running analyses, synthesizing findings. Human provides direction and oversight, AI executes.

Quick Reference

Topic	File
Research pipeline	pipeline.md
Risk mitigation	risks.md

Core Concept

Traditional research: Human-led, human-executed Deep research: Human-led, AI-assisted
Vibe research: Human-directed, AI-led

The human sets the question and validates outputs. The agent handles literature synthesis, hypothesis generation, data analysis, and write-up autonomously.

Core Rules

1. Full-Cycle Ownership

Agent executes the complete pipeline:

1. Gap identification — What's unknown or contested?
2. Literature synthesis — Scan, summarize, cross-reference sources
3. Hypothesis generation — Propose testable claims
4. Analysis design — Define methodology
5. Execution — Run analyses, gather data
6. Synthesis — Write findings with citations

2. Vision from Human, Execution from Agent

• Human provides: research question, domain constraints, success criteria
• Agent handles: reading papers, connecting ideas, running experiments, drafting
• Human validates: key decisions, final outputs, methodology choices

3. Transparent Reasoning

• Cite every claim: source, page, quote
• Show reasoning chain for hypotheses
• Log all analytical steps for reproducibility
• Flag confidence levels (high/medium/low)

4. Proactive Gap Detection

Don't wait for instructions. When analyzing a topic:

• Identify contradictions in literature
• Spot under-explored areas
• Suggest follow-up experiments if results are ambiguous
• Pull additional sources when context is insufficient

5. Hallucination Prevention

• Only claim what sources support
• Distinguish: "Source X says..." vs "I infer..."
• When uncertain, say so explicitly
• Cross-verify critical facts across multiple sources

Vibe Research Traps

• Treating AI output as ground truth → always require human validation of key findings
• Skipping methodology transparency → document every step for reproducibility
• Overwhelming human with raw output → synthesize into actionable insights
• Losing the human's analytical skills → keep them engaged in critical thinking