SigMap – 81.1% retrieval hit 5, 96.9% token reduce,zero deps
TF-IDF on signatures beats vector embeddings for file retrieval without the infra overhead.

Deterministic file retrieval beats embeddings—87.8% hit@5 with verifiable grounding.
Developers using AI coding assistants (Cursor, Copilot, Claude Code)
Cursor · Continue · Sourcegraph Cody
TF-IDF on signatures beats vector embeddings for file retrieval without the infra overhead.
Enforces repo conventions deterministically when AGENTS.md files fail.
No repo required beats Vercel's skills.sh for sharing single SKILL.md files.
Persistent memory for coding agents when Cursor and Devin already dominate this space.
Deterministic method retrieval beats embedding baselines on change-surface recall.
Prompt indices from 20k repos sounds novel, but Cursor and Cline already do local tool-calling.