Iterative Retrieval Protocol

When dispatching subagents that need codebase context, use this 4-phase loop (max 3 cycles) to ensure they find the right files.

The Problem

Single-pass searches miss codebase-specific terminology. Example: searching "rate limiting" finds nothing because BrickTrack calls it rate-limit.ts in src/server/middleware/. Cycle 2 discovers "middleware" and "throttle" as related terms.

The Protocol

Phase 1: DISPATCH (Broad Search)

Search with multiple keyword variants and broad glob patterns:

• Glob: src/**/*.ts matching *keyword*
• Grep: keyword1|keyword2|keyword3 across src/
• Include: type definitions, interfaces, barrel exports
• Exclude: *.test.ts, *.spec.ts, node_modules

Phase 2: EVALUATE (Score Relevance)

For each file found, assign a relevance score:

After scoring, identify: What context is STILL MISSING?

Phase 3: REFINE (Update Search)

Based on Phase 2 findings:

• Add new keywords discovered from high-scoring files
• Add codebase-specific terminology (e.g., uow instead of "unit of work")
• Narrow glob patterns to confirmed-relevant directories
• Exclude confirmed-irrelevant paths

Phase 4: LOOP (Repeat or Stop)

Stop when:

• 3+ files scored >= 0.7 AND no critical context gaps remain
• OR: 3 cycles completed (proceed with best available)

Continue when:

• Fewer than 3 high-relevance files found
• Critical vocabulary gaps remain

BrickTrack-Specific Vocabulary

Common vocabulary mismatches in this codebase:

• "auth" -> BetterAuth, getAuthSession(), authClient
• "database" -> Drizzle ORM, ctx.db, ctx.uow.repo
• "api" -> tRPC routers in src/server/routers/
• "components" -> src/components/my/ (app) vs src/components/ui/ (shadcn)
• "middleware" -> src/server/middleware/ (rate-limit, auth)
• "types" -> src/types/ + z.infer<typeof schema> inline

Adding to Subagent Prompts

Include this in PROMPT.md for implementation subagents:

Before implementing, locate all relevant files using iterative retrieval:
1. DISPATCH: Broad Glob + Grep (multiple keyword variants)
2. EVALUATE: Score each file 0-1 for relevance
3. REFINE: Adjust terms based on codebase vocabulary discovered
4. LOOP: Up to 3 cycles until 3+ files scored >= 0.7
Do NOT assume file names — search first.