From a senior to future senior Hackers

Latest Posts

View all posts →

Latest Videos

View all videos →
Auditing Python Install Hooks: Reproducing the agent-framework Credential Theft in a Lab

Auditing Python Install Hooks: Reproducing the agent-framework Credential Theft in a Lab

Public reporting in November 2026 said hijacked releases of Microsoft's agent-framework package on PyPI were used to steal Azure and GitHub credentials. The summaries leave the exact malicious versions, the compromise path, and whether the artefacts were sdists or wheels unknown. This short follows the post's defensive lab: a local PEP 503 index, a local collector, fabricated canary credentials, and a Docker network with no egress. It shows a minimal build hook that reads the install environment and posts to the collector, then covers the audit surfaces and the controls that reduce exposure: Trusted Publishing, scoped tokens, and lockfiles. The lab is intentionally local and uses fake credentials. Read the full walkthrough with commands and captured output at https://hackyjs.com/posts/auditing-python-install-hooks-reproducing-the-agent-framework-credential-theft-in-a-lab

pythonsupply-chain-securitypypi
Hy4 Flash on Long-Context RAG: What $0.07 per Million Tokens Actually Costs

Hy4 Flash on Long-Context RAG: What $0.07 per Million Tokens Actually Costs

Tencent lists Hy4 Flash with a 1M-token context at $0.07 per million tokens. This is what that price does to a long-context RAG pipeline once you work it through: cost per thousand requests for retrieval, stuffed context and agent loops, why prefill makes latency the binding constraint past 100K tokens, and the crossover point where retrieval stops paying for itself. None of the Hy4 Flash numbers here are measured. The cost figures are arithmetic from the published price, the time-to-first-token curve is explicitly modelled on assumed prefill throughput, and the harness is written so you can fill in your own endpoint numbers. Full write-up: https://hackyjs.com/posts/testing-tencent-hy4-flash-on-long-context-rag-cost-latency-and-recall-tradeoffs

llmraglong-context
Silent Configuration Drift: The Microsoft 365 Outage and the Missing Assertion

Silent Configuration Drift: The Microsoft 365 Outage and the Missing Assertion

A Microsoft 365 bug went undetected from July 2026 and contributed to a worldwide disruption, with the apparent peak on 13 November 2026. Coverage published on 16 and 17 November reported Microsoft's apology and three underlying causes. It did not name the affected component, the mechanism, or whether the July change was code, configuration, or data. That four-month window is the interesting part. Liveness monitoring asks whether a process is answering. Conformance monitoring asks whether state still equals what you decided. They fail independently, so a system can be fully live and silently wrong, with every uptime check green. The detector in this video is constructed to illustrate the failure shape. It is not a claim about Microsoft's internal configuration, which the public record does not describe. Full write-up: https://hackyjs.com/posts/silent-configuration-drift-what-the-microsoft-365-outage-reveals-about-detection-gaps

microsoft-365cloud-reliabilityincident-response