stack-researchlisted

# Stack Research > **CRITICAL:** Research is a FACT-GATHERING step, not a decision step. This skill returns data. The calling skill (project-planning or tech-spec-planning) decides based on the data. > **CONTEXT BUDGET:** subagents do all heavy reading. The orchestrator never loads individual reports into its own context — only aggregator output and short DONE lines. Orchestrate parallel documentation research for one or more stack candidates. Produce focused reports, optional comparison table, and update the stack registry. Version-based cache avoids redundant research. **Input (from caller):** - `decision_context`: one sentence — what is being decided - `candidates`: list of 1..N elements. Each: `{ name, type: external-api|library|service|tool }` - `depth`: `shallow` | `deep` - `project_context`: 2-3 sentences — what the project needs from this element - `feature_path`: absolute path to feature folder (per-feature reports live under `{feature_path}/logs/stack-research/`) **Output:** - Per-candidate report at `{feature_path}/logs/stack-research/{slug}.md` (written by `stack-researcher`) - If `candidates.length > 1` → `stack-comparison-{slug}.md` in PK (written by `stack-aggregator`) - If `depth=deep` → registry entry in `stack-research.md` in PK (written by `stack-aggregator`) - Short summary returned to caller (paths + DONE digest) ## Context Protocol 1. The orchestrator (this skill) NEVER reads per-candidate reports — only the aggregator does. 2. Every subagent retu