Plan and execute mini literature reviews

Plan and execute mini literature reviews via Consensus API

Researchers needing a structured literature orientation (launching pad) when full comprehensive reviews are not required.

• · Research question, framework hint (optional), desired depth (5/10/20 Consensus queries)

• · Editable Word document (.docx) mini literature review (launching-pad orientation guide, not finished review)
• · Phase 2 checkpoint (framework + sub-areas + depth confirmed)
• · Source-provenance tracking (1 q/sec rate limit, three-count sent/received/cited)

• · Consensus API (1 q/sec rate-limited, configurable depth 5/10/20 queries)
• · Node.js docx library (DOCX generation)

• · Grill-me intake (Q1-Q3) must complete before Phase 1 recon search
• · Phase 2 checkpoint requires user confirmation of framework (second forcing point)
• · Configurable depth (5/10/20) controls coverage vs speed trade-off

• · Shallow depth (5 queries) may miss sub-areas or contradictions
• · Rate-limiting (1 q/sec) means deep review can take 20+ seconds
• · Three-count tracking can break if Consensus API returns inconsistent citation counts

• · Source spec in PR #657 (megaprompts/09-litreview-megaprompt.md)
• · Agent Integrity Rules implemented (rate-limiting, three-count tracking, retry-once-after-3s)
• · Two forcing checkpoints (intake + Phase 2) ensure coherence
• · Consensus API is production academic search integration
• · DOCX output is editable and source-tracked