Show HN: We benchmarked AST stubbing trade-offs on Airflow and Zulip (ctxfw)

Category: ai-ml

Tags: ai-agents, ast, context-compaction, prompt-injection, mcp

Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 6)

CTXFW is a context firewall and MCP gatekeeper for AI coding agents that compacts peripheral code using AST stubbing (up to 90% token reduction) and enforces a deterministic prompt injection sieve. It aims to reduce token costs and improve security by operating locally with zero egress, backed by benchmarks against real-world codebases like Airflow and Zulip.

Target audience: backend devs, devops, AI engineers

Repository: https://github.com/heuristicolab/ctxfw/blob/main/docs/benchmarks/TRILOGY_EMPIRICAL_BENCHMARK.md · Python · Apache-2.0 · 7 stars

View on Hacker News