Research agent

Skill: Research Agent

Research Protocol

Step 1: Decompose the query

Query: "best free LLM for code generation 2026"

Decompose into:
- What: free LLMs (no cost inference)
- Domain: code generation (function, debug, complete)
- Constraint: 2026 (recent, not outdated)
- Implicit: API vs local, speed vs quality

Step 2: Source hierarchy (search in parallel)

Priority	Source	Tool	What to look for
1	Codebase	semantic_search, grep	Existing patterns, prior art
2	Official docs	fetch_webpage	Latest API, breaking changes
3	GitHub	fetch_webpage issues/discussions	Known bugs, community solutions
4	Community	fetch_webpage (SO, Reddit, Discord)	Real-world experiences
5	Registries	npm/PyPI/crates.io	Popularity, maintenance status

Step 3: Validate findings

Validation checklist:

• Date check: is this < 6 months old? If not, verify still valid
• Version check: compatible with project's pinned versions?
• Cross-reference: confirmed by ≥ 2 independent sources?
• Author credibility: official docs > known experts > anonymous posts

Step 4: Synthesize

Structure output as:
1. TL;DR (1 sentence answer)
2. Recommendation (with rationale)
3. Key findings (3-5 bullet points)
4. Trade-offs (pros/cons table)
5. Sources (cited with date)
6. Next steps (actionable)

Research Templates

Technology comparison

Compare [A] vs [B] for [use case]:

| Criteria | A | B |
|----------|---|---|
| Performance | | |
| Cost | | |
| Community | | |
| Maintenance | | |
| License | | |
| Learning curve | | |

Recommendation: [A/B] because [reason]

SOTA research

SOTA for [topic] as of [date]:

State of the art:
- Best approach: [name] — [why]
- Performance benchmark: [metric] = [value]
- Key paper: [authors, venue, year]
- Production-ready implementation: [repo/package]

Recent game changers:
- [paper/technique]: changes [what] by [how much]

What's still open:
- [challenge 1]
- [challenge 2]

GitHub issue investigation

Bug: [description]

Search: GitHub issues → repo/issues?q=[keywords]

Findings:
- Issue #[N]: [title] — Status: [open/closed/PR merged]
- Fix: [commit/PR link]
- Workaround: [if applicable]

Root cause: [explanation]
Solution: [steps]

Query Optimization for AI Research

Use these search patterns for better results:

# Web search — specific and recent
"[technology] [problem] site:github.com 2025"
"[library] best practices [year]"
"[error message] fix [language]"

# GitHub issues
"repo:[owner/repo] [keyword] is:issue"
"[error] in:title is:closed"

# Docs
"[library] [version] [feature] docs"

Anti-Patterns in Research

• ❌ Trust first result only → ✅ Cross-reference 3+ sources
• ❌ Use outdated tutorials (2020-2022 for fast-moving domains) → ✅ Check dates
• ❌ Skip official docs → ✅ Official docs first for API accuracy
• ❌ Research without codebase check → ✅ Check existing patterns first
• ❌ Deep research for simple facts → ✅ Match depth to complexity

ZeroClaw Integration

# Trigger deep research SOP
zeroclaw agent -m "hint:smart research: what's the best free embedding model for RAG in 2026?"

# Research + store in memory
zeroclaw agent -m "research zeroclaw provider rotation patterns and save findings to memory"