GitHub

Cloudflare 403s and hardening ccxt clients

The article explains why legitimate bots interacting with exchange APIs can receive intermittent 403 Forbidden responses due to Cloudflare WAF bot challenges and presents a two-layer mitigation pattern implemented in an open-source library. The pattern includes hardening HTTP headers (User-Agent, Accept-Language, timeouts) to reduce WAF challenges and a selective retry helper that only retries transient errors (e.g., Cloudflare 403, 429, timeouts) with exponential backoff and jitter. The author published the ccxt-resilience library (Apache-2.0) with functions like harden and with_retry and provides installation and GitHub repository details. The piece is practical guidance for developers integrating with exchanges (example: OKX).

AI Agents Produce Flawed Production Code: Evaluation Bottleneck

An engineer who spent months grading AI-agent-generated code reports a recurring failure pattern: agent outputs are often syntactically correct but blind to real-world failure modes (retries, timeouts, partial writes, IAM, concurrency, distributed state). The author argues this is an evaluation problem — not a pure model capability issue — and says job roles like "AI evaluator" and practices such as RL environment design and LLMOps are emerging to address it. They describe common failures (reward hacking, golden-path assumptions) and announce they are building an open fault-injection harness to stress-test agent-generated infrastructure code with deterministic pass/fail checks, combining chaos engineering with AI evaluation. The author will publish the project on their portfolio and GitHub and invites collaboration.

Stable citations for TypeScript RAG apps

A technical article (published 2026-08-06) describing a pattern for making retrieval-augmented-generation (RAG) citations robust to re-chunking, re-embedding, and re-ingest workflows. The author proposes a typed SourceRef structure (docId, contentHash, start, end, revision), uses non-empty tuple types to require citations for factual claims, outlines a label-resolution flow to avoid model transcription of hashes, and defines four resolution outcomes (exact, moved, stale, gone) with corresponding render behavior. The post includes code examples and a test to ensure citations remain resolvable after ingest changes.