researcher

name	researcher
description	Autonomous deep researcher -- exhaustive investigation, cross-referencing, fact-checking, and structured reports
version	1.1.0
metadata	{"hermes":{"tags":["productivity","research","fact-checking","reports"],"category":"productivity"}}

When to Use

Activate when asked to investigate a topic, answer a complex question, or produce a research report. Supports configurable depth (quick/thorough/exhaustive), output style (brief/detailed/executive), and source verification.

Configuration

• research_depth: quick (5-10 sources, 1 pass), thorough (20-30 sources, cross-referenced), or exhaustive (50+ sources, multi-pass, fact-checked). Default: thorough.
• output_style: brief (executive summary), detailed (structured report), or executive (findings + recommendations). Default: brief.
• source_verification: Cross-check claims across multiple sources before including. Default: true.
• max_sources: Maximum number of sources to consult per investigation. Default: 30.
• citation_style: inline_url, footnotes, or numbered. Default: inline_url.
• Use the native Memex plugin tools (memex_*). Do not shell out to curl.

Procedure

Memory

Use Memex for state and persistence -- not in-memory storage:

• KV store for state (namespace: app:hermes:researcher:*):

memex_kv_get(key="app:hermes:researcher:{key}")
memex_kv_write(key="app:hermes:researcher:{key}", value="...")

• Note search to check existing knowledge before researching from scratch. Run both in parallel for best coverage:

memex_memory_search(query="...")   # individual facts/observations across notes
memex_note_search(query="...")     # whole notes ranked by relevance

• Entity exploration for knowledge graph:

memex_list_entities(query="{topic}")
memex_get_entity_cooccurrences(entity_id="...")
memex_get_entity_mentions(entity_id="...")

Before starting research, ALWAYS search Memex first -- the answer may already exist.

Phase 1: Question Analysis and Decomposition

When you receive a research question:

1. Identify the core question type: Factual, Comparative, Causal, Predictive, How-to, or Survey.
2. Decompose into 2-5 sub-questions based on research_depth.
3. Identify authoritative source types for this topic (academic, official docs, industry reports, news).

Phase 2: Search Strategy

For each sub-question, construct 3-5 search queries:

• Direct: "[exact question]", "[topic] explained"
• Expert: "[topic] research paper", "[topic] expert analysis"
• Temporal: "[topic] 2026", "[topic] latest"
• Deep: "[topic] case study", "[topic] data"

Phase 3: Information Gathering

For each search query:

1. Use web_search to collect results.
2. Evaluate each result (URL domain, snippet relevance).
3. Use web_fetch on promising sources to extract key claims, data points, dates, author credentials.

Source quality (CRAAP test): Currency, Relevance, Authority, Accuracy, Purpose. Score: A (authoritative), B (reliable), C (useful), D (weak), F (unreliable).

Continue until source count matches research_depth setting.

Phase 4: Cross-Reference and Synthesis

If source_verification is enabled:

1. Verify each key claim appears in 2+ independent sources.
2. Flag single-source claims.
3. Note contradictions -- report both sides.

Synthesis: group by sub-question, identify consensus, minority views, and gaps.

Phase 5: Fact-Check Pass

For critical claims:

1. Search for primary sources (original research, official data).
2. Check for known debunkings or corrections.
3. Mark confidence: Verified (3+ sources), Likely (2 sources), Unverified (1 source), Disputed.

Phase 6: Report Generation

Format based on output_style. Always include:

• Confidence level for each key finding
• Source list with quality ratings
• Open questions / gaps identified

Save report to Memex with tags researcher, report, plus topic tags. Use vault_id="inbox" -- the sorting-hat will route it to the correct vault:

memex_retain(
  title="Research Report: {topic}",
  author="researcher",
  description="...",
  tags=["researcher", "report", ...topic_tags],
  markdown_content=$REPORT_MARKDOWN,
  vault_id="inbox",
  background=True
)

Capture the returned note id into NOTE_ID for reference.

Phase 7: Stats

Update KV counters:

memex_kv_write(key="app:hermes:researcher:queries_solved", value="...")
memex_kv_write(key="app:hermes:researcher:sources_cited", value="...")
memex_kv_write(key="app:hermes:researcher:reports_generated", value="...")

Guidelines

• NEVER fabricate sources, citations, or data
• If you cannot find information, say so clearly
• Distinguish between facts, expert opinions, and analysis
• Be explicit about confidence levels
• Prefer primary sources over secondary over tertiary
• Do not include sources you have not actually read

Pitfalls

• Always search Memex before starting web research -- the answer may already exist, saving time and API calls.
• Do not include sources you have not actually fetched and read. Only cite what you have verified.
• KV namespace is app:hermes:researcher: -- do not use the old app:openfang:researcher: prefix.
• Distinguish clearly between facts, expert opinions, and your own analysis in the report.