Deep Research

Indexed Reference Map

• @[SKILL.md#L1-L56] Frontmatter, scope, trigger rules, SOP, error handling, output contract.
• @[references/INDEX.md#L1-L11] Reading order and document map.
• @[references/QUICK_REFERENCE.md#L1-L23] Fast trigger map and execution ladder.
• @[references/CORE_API.md#L1-L37] Research phases, source priority, and date handling contract.
• @[references/PATTERNS.md#L1-L28] Search/query patterns, validation, and scope control.
• @[references/EXAMPLES.md#L1-L16] Example asks and expected depth.
• @[references/AI_WORKFLOW.md#L1-L14] Bounded AI execution sequence.

Scope

• In-scope:
  • Questions requiring current or externally verified information.
  • Topic comparisons, ecosystem scans, trend checks, and evidence-backed explainers.
  • Pre-research before producing a report/article/plan/design that depends on outside facts.
• Out-of-scope:
  • Questions answerable directly from local context or stable internal knowledge.
  • Blindly replacing all normal answering with web research.
  • Claiming certainty when evidence is sparse or contradictory.

KB Mode Notes

• Reading order:
  1. references/INDEX.md
  2. references/QUICK_REFERENCE.md
  3. references/CORE_API.md
  4. references/PATTERNS.md
  5. references/EXAMPLES.md
  6. references/AI_WORKFLOW.md
• Coverage Matrix:
  • Triggering and stop conditions -> QUICK_REFERENCE.md
  • Research phases and source priority -> CORE_API.md
  • Query design and validation discipline -> PATTERNS.md
  • Expected usage examples -> EXAMPLES.md
  • AI orchestration guardrails -> AI_WORKFLOW.md
• Portability note:
  • Tool names and workflow assumptions here are source-of-truth for the OpenClaw-Module Android workspace; verify source-of-truth before porting elsewhere.

Trigger Rules

• Positive:
  • User asks to research, compare, investigate, explain with current data, or scan external sources.
  • The answer would be weak without current facts, examples, or corroboration.
  • A content-generation task explicitly depends on outside information.
• Negative:
  • The answer can be given reliably from existing local context or stable technical knowledge.
  • The user wants a quick opinion or local-only reasoning with no need for web evidence.
• Escalation:
  • Narrow the objective, timeframe, geography, or competitor set before broad searching if the question is too open-ended.
  • When evidence conflicts, surface the conflict and reduce confidence instead of forcing synthesis.

SOP

1. Bind the research to the actual user question, not a generic broad topic.
2. Start with broad discovery to identify dimensions, stakeholders, and key terms.
3. Run targeted searches for the highest-value subtopics.
4. Use web_fetch on authoritative/high-signal pages rather than relying only on snippets.
5. Cover both supportive and critical angles, including limitations.
6. Synthesize only after facts, examples, and current context are sufficient for the user’s need.
7. State uncertainty, freshness limits, and missing evidence explicitly.

Error Handling

• Search results are noisy:
  • tighten query terms, add source/type hints, or constrain by time/region.
• Evidence is thin or contradictory:
  • return a bounded answer with lower confidence and explicit caveats.
• Topic is too broad:
  • reduce scope before continuing instead of accumulating low-value searches.
• Freshness matters but date context is unclear:
  • use session_status to confirm the current date before time-sensitive queries.

Output Contract

• Artifact:
  • Research notes or a final answer grounded in multiple external sources, with citations/links when material claims depend on them.
• Validation:
  • Must reflect multiple angles when the question is multi-faceted.
  • Must not rely on a single superficial search when broader evidence is needed.
  • Must distinguish verified facts from interpretation.
• Compliance Verdict:
  • PASS when research scope, evidence quality, and uncertainty handling match the user’s real need; otherwise PARTIAL or FAIL.