Research Codebase - Targeted Code Investigation

You are researching the codebase to answer specific questions about the feature specification.

Input: $ARGUMENTS contains the ambiguity analysis report with search strategies.

Research Approach: Hybrid Priority

Execute research in two phases:

Phase A: Pattern-Seeking (Context Building)

First, build understanding of relevant codebase areas:

1. Find Similar Features

   Glob: **/*{related_feature}*
   Grep: similar functionality keywords
   

2. Identify Project Conventions

   • File organization patterns
   • Naming conventions
   • Common abstractions used

3. Map Relevant Architecture

   • Entry points (routes, controllers, handlers)
   • Data flow patterns
   • Shared utilities and helpers

Phase B: Question-Driven (Targeted Answers)

For each RESEARCH_NEEDED item from the ambiguity analysis:

1. Execute the suggested search strategies
2. Read relevant files discovered
3. Extract specific answers
4. Assess confidence level

Confidence Assessment Criteria

HIGH Confidence (promote to spec)

• Direct code evidence (function exists, pattern used)
• Explicit comments or documentation
• Multiple consistent examples
• Standard framework/library behavior

MEDIUM Confidence (note but verify)

• Single example found (might be exception)
• Inferred from related code
• Convention appears consistent but not documented
• Third-party dependency behavior

LOW Confidence (keep as question)

• No direct evidence found
• Contradictory examples
• Requires human judgment call
• Involves external systems not in codebase

Research Execution

For each research target:

## Researching: {ambiguity description}

### Search Executed
- Glob: {pattern} → {n} files found
- Grep: {pattern} → {n} matches
- Files read: {list}

### Findings
{What was discovered}

### Evidence
File: {path}:{line}
```{code snippet}```

### Confidence: {HIGH|MEDIUM|LOW}
Rationale: {why this confidence level}

### Suggested Spec Update
{How this should update the specification}

Special Research Cases

When Nothing Found

• Note the absence as meaningful data
• Suggest this might be new territory for the project
• Recommend human decision on approach

When Contradictions Found

• Document all variations
• Note which appears more recent/prevalent
• Flag for human resolution

When Scope Expands

• If research reveals the feature is larger than expected
• Note the expansion but don't pursue rabbit holes
• Flag for orchestrator to handle

Output Format

RESEARCH COMPLETE
Questions Investigated: {n}
High Confidence Answers: {n}
Medium Confidence Answers: {n}
Still Open: {n}

## Findings by Question

### Q1: {original question}
**Confidence**: HIGH
**Answer**: {clear answer}
**Evidence**: {file:line references}
**Spec Update**: Add to High Confidence section: "{suggested text}"

### Q2: {original question}
**Confidence**: MEDIUM
**Answer**: {tentative answer}
**Evidence**: {file:line references}
**Spec Update**: Add to Medium Confidence section: "{suggested text}"

### Q3: {original question}
**Confidence**: LOW
**Answer**: Could not determine
**Searched**: {what was tried}
**Spec Update**: Keep in Open Questions, add context: "{additional info}"

## New Questions Discovered
During research, these new questions emerged:
1. {question surfaced while investigating}
2. {another emergent question}

## Relevant Files for Future Reference
- {path}: {why it's relevant}
- {path}: {why it's relevant}

Constraints

• Do NOT modify any files - read only
• Do NOT pursue tangential research
• Stay focused on the specific questions provided
• Time-box each question - don't exhaust search on one item
• If a question would require extensive research, note it and move on