# Strategic Search
> Solve complex multi-constraint problems through systematic web search using multiple engines with adaptive strategies.
- Skill: `majiayu000/strategic-search` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/strategic-search`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/strategic-search/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/strategic-search
---
# Strategic Search
You solve complex, multi-constraint problems by systematically researching and combining multiple clues. You have access to a search MCP server that provides DuckDuckGo, Brave, and Serper engines.
- **Core approach**: Extract all constraints, rank them by uniqueness and searchability, research progressively, and validate candidates until finding the answer that satisfies ALL conditions.
- **Engine strategy**: DuckDuckGo (free) is the default; use Brave/Serper (paid) when DuckDuckGo results are insufficient or for cross-validation.
---
## CRITICAL: Search Budget Management
**You have a maximum of 30 searches. Track your count explicitly at every step.**
**Format:** `[Search X/30]` before EVERY query.
**Phase budgets (recommended):**
- Phase 1 (Extract & Rank): 0 searches (analysis only)
- Phase 2 (Initial Search & Validation): 8-15 searches
- Phase 3 (Cross-Validation): 3-5 searches
- Phase 4 (Revision if needed): 5-10 searches
**Early termination triggers:**
- ✅ **STOP immediately** when you find a candidate that satisfies ALL constraints
- ✅ **STOP at search 20** if you have a strong candidate (satisfies 80%+ constraints)
- ✅ **STOP at search 25** regardless, finalize your best answer
- ✅ **STOP at search 30** hard limit - return best candidate with explanation
---
## Core Workflow: Constraint-Based Search
**Example problem:** "Identify the fictional character who breaks the fourth wall, has a backstory with ascetics, is known for humor, and had a TV show in the 1960s-1980s with fewer than 50 episodes."
### Phase 1: Extract and Rank Constraints (0 searches)
#### 1.1. Extract all constraints
List every condition explicitly:
**Example:**
1. Fictional character
2. Breaks the fourth wall
3. Has a backstory with ascetics
4. Known for humor
5. Had a TV show in the 1960s-1980s
6. TV show had fewer than 50 episodes
#### 1.2. Rank by uniqueness and searchability
**Ranking strategy:**
- **Most unique/specific** → Highest priority (narrows search space dramatically)
- **Verifiable numerical constraints** → High priority (precise validation)
- **Common traits** → Lower priority (too broad initially)
**Example (high → low):**
1. **"ascetics backstory"** - Most distinctive characteristic
2. **"fewer than 50 episodes"** - Specific, verifiable number
3. **"TV show 1960s-1980s"** - Narrows time period significantly
4. **"breaks fourth wall"** - Distinctive but more common in comedy
5. **"known for humor"** - Very broad; many shows fit this
6. **"fictional character"** - Universal; not useful for initial search
**Rationale:** Start with the constraint that eliminates the most candidates. "Ascetics" is unusual enough to point toward specific genres (Eastern spirituality, martial arts, mythology), making it the ideal starting point.
---
### Phase 2: Progressive Search and Validation (Target: 8-15 searches)
Start with the **highest-ranked constraint**, generate candidate answers, then progressively filter using remaining constraints until finding the final answer.
#### 2.1. Initial Search (2-3 searches)
**Query construction:**
- Combine the top 1-3 most distinctive constraints
- Use synonyms to broaden coverage
- Keep queries concise but specific
**Example:**
```
[Search 1/30]
Query: "TV show 1970s 1980s monks ascetics humor"
Goal: Find shows with Eastern spiritual themes
Expected: Kung Fu, Monkey, martial arts shows
```
#### 2.2. Candidate Validation (5-12 searches)
**Key principle: Validate ONE candidate at a time, STOP as soon as ALL constraints are satisfied.**
For each candidate, verify constraints in order of ease/speed:
**Example - Candidate "Kung Fu":**
```
[Search 2/30]
Query: "Kung Fu TV series 1972 episode count"
Result: 63 episodes
✗ REJECTED immediately - Fails episode constraint (<50)
→ Move to next candidate
```
**Example - Candidate "Monkey":**
```
[Search 3/30]
Query: "Monkey TV show 1978 episode count"
Result: 52 episodes (Japanese), 39 episodes (English dub)
✓ Episode constraint satisfied (39 < 50)
→ Continue validation
[Search 4/30]
Query: "Monkey TV show 1978 fourth wall breaks audience"
Result: Confirms character addresses the camera directly
✓ Fourth wall constraint satisfied
→ Continue validation
[Search 5/30]
Query: "Monkey TV show Buddhist monk Tripitaka ascetics"
Result: Confirms journey with Buddhist monk, spiritual themes
✓ Ascetics constraint satisfied
Validation checklist:
✓ Fictional character (Sun Wukong/Monkey King)
✓ Breaks the fourth wall (speaks directly to camera)
✓ Ascetics backstory (travels with Tripitaka, Buddhist monk)
✓ Known for humor (comedic, mischievous character)
✓ TV show 1978-1980 (within 1960s-1980s range)
✓ <50 episodes (39 episodes in English version)
→ ALL CONSTRAINTS SATISFIED - STOP SEARCHING
→ Final answer: Monkey (TV series, 1978-1980)
```
**Total searches used: 5/30** ✅
---
### Phase 3: Cross-Validation (Optional, 3-5 searches)
**Only use if:**
- You found a candidate but have doubts
- Multiple viable candidates exist
- Critical decision requires high confidence
If you have a strong candidate that satisfies all constraints, **skip this phase**.
**Example:**
```
[Search 6/30]
Query: "Monkey 1978 TV series episode count IMDb"
→ Confirms: 52 episodes total, 39 English dub
[Search 7/30]
Query: "Monkey TV show 1978 fourth wall Wikipedia"
→ Confirms fourth wall breaking in multiple sources
```
---
### Phase 4: Revision Strategy (5-10 searches, only if needed)
**Trigger revision ONLY if:**
- After 15 searches, no viable candidates satisfy >50% of constraints
- All candidates fail critical constraints
- Search results consistently off-target
**If you reach search 15 without a good candidate, PAUSE and revise:**
**Revision strategy:**
- **Re-examine constraints**
- Are you interpreting "ascetics" too narrowly?
- Could "fourth wall" mean something different?
- Is the time period flexible (late 60s, early 80s)?
- **Re-rank constraints**
- Perhaps start with "fourth wall + <50 episodes" instead
- Try different constraint combinations
- **Switch search engines**
- Use Brave/Serper (paid)
- Different engines index content differently
- **Adjust search parameters**
- Increase `max_results` in `search` MCP tool (default: 10 → try 15-20)
- More results = higher chance of finding obscure matches
**Hard stops:**
- **Search 20**: If you have a candidate satisfying 80%+ constraints, finalize it
- **Search 25**: Finalize your best candidate regardless
- **Search 30**: ABSOLUTE LIMIT - return best candidate with explanation
---
## Complete Worked Example: Full Search Sequence
**Problem:** "What TV show from the 1970s-1980s featured a character who breaks the fourth wall, has a backstory involving ascetics, is comedic, and had fewer than 50 episodes?"
### Phase 1: Analysis (0 searches)
```
Constraints extracted:
1. TV show 1970s-1980s
2. Character breaks fourth wall
3. Backstory with ascetics
4. Comedic/humorous
5. <50 episodes
Ranking (most unique first):
1. Ascetics backstory (MOST DISTINCTIVE)
2. <50 episodes (VERIFIABLE)
3. 1970s-1980s (NARROWS TIME)
4. Breaks fourth wall (DISTINCTIVE)
5. Comedic (COMMON)
Strategy: Search "ascetics + 1970s TV" first
```
### Phase 2: Search & Validate
```
[Search 1/30]
Query: "TV show 1970s 1980s monks ascetics comedy"
Engine: DuckDuckGo
Results: "Kung Fu", "Monkey", "The Water Margin"
→ 3 candidates identified
[Search 2/30]
Query: "Kung Fu TV series episode count"
Result: 63 episodes
✗ Fails <50 constraint → REJECT
[Search 3/30]
Query: "Monkey TV show 1978 episode count"
Result: 52 episodes (Japanese), 39 (English dub)
✓ Passes <50 constraint (39 episodes)
→ Promising candidate
[Search 4/30]
Query: "Monkey TV show 1978 breaks fourth wall"
Result: "Monkey frequently addresses the camera and audience directly"
✓ Passes fourth wall constraint
[Search 5/30]
Query: "Monkey TV show Tripitaka Buddhist monk"
Result: "Journey to the West adaptation, Tripitaka is a Buddhist monk"
✓ Passes ascetics constraint
Quick checklist:
✓ 1970s-1980s (1978-1980)
✓ Breaks fourth wall (confirmed)
✓ Ascetics backstory (Buddhist monk companion)
✓ Comedic (described as comedic throughout)
✓ <50 episodes (39 English episodes)
→ ALL CONSTRAINTS SATISFIED
→ ANSWER: "Monkey" (1978-1980 TV series)
→ STOP SEARCHING
Total searches: 5/30 ✅
```
**Key success factors:**
- Started with most distinctive constraint (ascetics)
- Rejected failed candidates immediately (Kung Fu at search 2)
- Validated incrementally, stopped as soon as all constraints met
- Did NOT continue searching after finding valid answer
---
## Search Optimization Techniques
### Technique 1: Synonym Expansion
```
"breaks fourth wall" → "talks to audience", "addresses camera", "narrator character"
"ascetics" → "monks", "hermits", "spiritual teachers", "Buddhist", "Shaolin"
"humor" → "comedy", "comedic", "funny", "satirical"
```
### Technique 2: Query Chaining
Build subsequent queries from previous results.
```
[Search 1/30]: "TV 1970s spiritual comedy" → Identifies: Show X, Show Y, Show Z
[Search 2/30]: "Show X episode count" → Verify constraint
[Search 3/30]: "Show X fourth wall breaking" → Verify constraint
[Search 4/30]: "Show X monks ascetics backstory" → Verify constraint
```
### Technique 3: Constraint Combination
```
Combine multiple constraints:
"TV 1970s fourth wall <50 episodes comedy monks"
→ May directly hit the answer if the search engine indexes well
```
### Technique 4: Rapid Elimination
```
If a candidate FAILS any hard constraint (numbers, dates), REJECT immediately.
Don't waste searches validating other constraints.
Example:
[Search N/30]: "Show X episode count" → 75 episodes
✗ Fails <50 → STOP, move to next candidate
→ DON'T search for fourth wall, ascetics, etc.
```
---
## MCP Tool Interfaces
| Tool | Function | Return |
| ---------------------------------------------------- | ------------------------------------ | --------------------------------------------------- |
| `get_engine_status()` | Check availability of search engines | DuckDuckGo: O\|X
Brave: O\|X
Serper: O\|X
|
| `search(query, max_results=10, engine="duckduckgo")` | Run search query on selected engines | List of {rank, title, content, url} |
---
## Search Engines
**Engine selection strategy:**
- **DuckDuckGo (free)**: Default for all searches
- **Brave / Serper (paid)**: Use when:
- DuckDuckGo returns <5 relevant results
- Cross-validation needed for confidence
- After 10 searches with no strong candidates
**Budget management:**
- Maximum 30 total queries per problem
- Track count explicitly: `[Search X/30]`
- Stop immediately when answer is validated
- At search 20: finalize if you have 80%+ match
- At search 25: finalize best candidate
- At search 30: hard stop
---
## Anti-Patterns: What NOT to Do
**DON'T:** Continue searching after finding a valid answer
```
[Search 8/30]: Found "Monkey" - satisfies all constraints
[Search 9/30]: "Let me verify one more time..." ← WASTEFUL
```
**DO:** Stop immediately when all constraints are met
```
[Search 8/30]: Found "Monkey" - satisfies all constraints
→ ANSWER: "Monkey" ✓
```
---
**DON'T:** Validate all constraints on failed candidates
```
[Search 5/30]: "Show X episode count" → 100 episodes (fails <50)
[Search 6/30]: "Show X fourth wall" ← WASTEFUL, already failed
[Search 7/30]: "Show X ascetics" ← WASTEFUL, already failed
```
**DO:** Reject immediately on hard constraint failures
```
[Search 5/30]: "Show X episode count" → 100 episodes
✗ REJECTED - move to next candidate
```
---
**DON'T:** Forget to track search count
```
Query: "TV show monks"
Query: "another search"
Query: "yet another"
... (loses count, exceeds budget)
```
**DO:** Track explicitly every time
```
[Search 1/30]: "TV show monks"
[Search 2/30]: "another search"
[Search 3/30]: "yet another"
```