| name | search_strategy |
| description | Plan a multi-step research strategy for complex questions requiring deep investigation |
| metadata | {"openclaw":{"tags":["enterprise-search","strategy","research","planning"]}} |
Search Strategy
Plan and execute a structured, multi-step research strategy for complex questions
that cannot be answered by a single search. This skill breaks down a research
question into sub-queries, identifies the best sources for each, and builds a
progressive investigation plan.
When to Use
Activate this skill when the user wants to:
- Investigate a complex topic that spans multiple domains
- Research a competitive landscape or market question
- Understand a decision's full context from multiple angles
- Prepare for a deep-dive analysis that requires structured data gathering
- Answer a question where the right sources are not obvious up front
Workflow
Step 1: Analyze the research question
Break the user's question into its component parts:
- Identify the core question and any implicit sub-questions
- Determine what types of information are needed (factual data, opinions, historical context, internal decisions)
- Identify which organizational sources are most likely to have relevant information
- Estimate the scope: narrow (1-2 sub-queries) vs. broad (5+ sub-queries)
Step 2: Design the search plan
Create a structured plan with ordered phases:
Phase A -- Context gathering:
Define 2-3 broad searches to establish baseline understanding.
Phase B -- Targeted investigation:
Define 3-5 focused searches that drill into specific aspects identified in Phase A.
Phase C -- Gap filling:
Reserve 1-2 searches for questions that emerge from Phases A and B.
For each planned search, specify:
- The sub-query in natural language
- Which capability to use (
research.web_search, docs.search_files, mail.read_inbox, chat.search_messages)
- Why this source is appropriate for this sub-query
- What a successful result looks like
Step 3: Execute Phase A -- context gathering
For each Phase A search, use capability_execute with the following parameters:
- capabilityId: the capability identified in the plan (e.g.,
research.web_search)
- packId:
enterprise-search
- args:
query: the sub-query text
maxResults: 10
Review Phase A results before proceeding. Refine Phase B searches based on
what was found. If Phase A reveals the question is simpler than expected,
skip directly to synthesis.
Step 4: Execute Phase B -- targeted investigation
For each Phase B search, use capability_execute with the appropriate capabilityId:
- capabilityId: as specified in the plan
- packId:
enterprise-search
- args:
query: the refined sub-query
maxResults: 10
After each search, assess whether the question is sufficiently answered.
Stop early if enough evidence has been gathered.
Step 5: Execute Phase C -- gap filling
If there are still unanswered aspects after Phase B, execute gap-filling searches.
Use capability_execute with:
- capabilityId: the most appropriate capability for the gap
- packId:
enterprise-search
- args:
query: the gap-filling query
maxResults: 5
Step 6: Compile the strategy report
Assemble findings from all phases into a structured report.
Output Format
## Research Strategy Report: "{original question}"
### Strategy Overview
**Core question:** {the user's question restated clearly}
**Research approach:** {1-2 sentences describing the strategy taken}
**Phases completed:** {A, B, C as applicable}
**Total searches executed:** {count}
### Phase A: Context Gathering
**Sub-query 1:** "{query text}"
- **Source:** {capability used}
- **Key findings:**
- {finding 1}
- {finding 2}
- **Implications for next phase:** {what this tells us to look for next}
**Sub-query 2:** "{query text}"
- **Source:** {capability used}
- **Key findings:**
- {finding 1}
- {finding 2}
### Phase B: Targeted Investigation
**Sub-query 3:** "{query text}"
- **Source:** {capability used}
- **Key findings:**
- {finding 1}
- {finding 2}
[... additional sub-queries ...]
### Phase C: Gap Filling
{If executed, same format. If skipped, note why.}
### Synthesis
**Answer to the core question:**
{3-5 paragraph synthesis of all findings, directly addressing the user's question}
**Confidence level:** {high / medium / low}
**Key evidence:**
- {most important supporting finding with source}
- {second most important finding with source}
- {third most important finding with source}
**Gaps and limitations:**
- {information that could not be found}
- {sources that were not available}
### Recommended Next Steps
1. {next step if the user wants to go deeper}
2. {alternative angle to explore}
3. {action to take based on findings}
If the user's question turns out to be simpler than expected, collapse the
output into a shorter format and note that a full strategy was not needed.