# Iterative Retrieval

> Dispatch subagents with automatic refinement loop until sufficient context is gathered

- Skill: `hubeiqiao/iterative-retrieval` (Agent Skill)
- Install (CLI): `npx skillmds@latest add hubeiqiao/iterative-retrieval`
- Raw SKILL.md: https://api.skillmd.com/api/skills/hubeiqiao/iterative-retrieval/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: hubeiqiao (https://skillmd.com/u/hubeiqiao)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/hubeiqiao/iterative-retrieval

---


# Iterative Subagent Retrieval Protocol

When dispatching a subagent to retrieve information from a source (transcripts, documents, codebases, external data), follow this protocol:

## Phase 1: Initial Dispatch

1. Define the PRIMARY OBJECTIVE - what decision or action does this information support?
2. Define INITIAL QUERIES - specific items to retrieve
3. Dispatch subagent with both the queries AND the objective context
4. Store the returned agent ID

## Phase 2: Sufficiency Evaluation

Upon receiving the subagent's summary, perform a critical evaluation:

### Ask yourself:
- Does this summary contain gaps that the objective requires filling?
- Did the subagent mention adjacent information I didn't ask for but now realize I need?
- Are there ambiguities in what was returned that require clarification from the source?
- Would I be confident proceeding with ONLY this information?

### Decision:
- If SUFFICIENT → proceed with task, note agent ID for potential future resumption
- If INSUFFICIENT → continue to Phase 3

## Phase 3: Refinement Request

Resume the subagent using its agent ID with:
1. Acknowledge what was useful from the initial return
2. Specify EXACTLY what additional context is needed and WHY
3. Ask if there's related information the subagent noticed but didn't include

## Phase 4: Loop

Repeat Phase 2-3 until:
- Sufficiency is achieved, OR
- Maximum 3 refinement cycles reached (prevent infinite loops), OR
- Subagent confirms source is exhausted

## Output Format

When reporting to user or continuing work, note:
- Total refinement cycles used
- What additional context was gathered beyond initial request
- Agent ID for future resumption if needed