LLM Digest
Subscribe

AI Storyline

24 items · 3 sources · 11 days

View as JSON

Operational story trace

DeepSeek Everything

Current stateActive iteration · critical RCE disclosed Aug 25status changed Aug 25

Latest change

QiAnXin disclosed a critical remote-code-execution flaw in Harness on Aug 25 — the first reported security vulnerability in the two-week-old framework.

Earlier contextThe story so far

DeepSeek open-sourced Harness in mid-August as an MIT-licensed agent framework where models, tools, and skills are all plugins, drawing immediate comparisons to Claude Cowork. A fast release cadence followed — a v0.1 preview, then a 14-change post-beta update adding multimodal and subagent support — before Harness began orchestrating rival coding agents instead of just running its own.

editor-curated · source-linked

State over time

● prelude · Aug 5critical RCE disclosed · Aug 25 ●
  • prelude · Aug 5
  • open-source launch · Aug 13–15
  • MIT dev preview + v0.1 · Aug 17
  • weekly iteration · Aug 20–21
  • deep dive · Aug 24
  • critical RCE disclosed · Aug 25
PRELUDE · Aug 5
Early, unconfirmed "DeepSeek Code" buzz circulates before any harness reveal
1 source · show source ▾
OFF-TOPIC · Aug 11–12
Two Anthropic Claude Cowork posts, swept in by shared "Claude Code" keyword clustering
2 sources · show sources ▾
LAUNCH · Aug 13
DeepSeek open-sources Harness, an agent framework where "everything is a plugin"
1 source · watcher updated status · show source ▾
AMPLIFICATION · Aug 14–15
Coverage surges: hands-on tests, "missing layer" framing, direct Claude Cowork comparisons
10 sources · scout · show sources ▾
MIT RELEASE · Aug 17
Harness ships v0.1 developer preview under an MIT license
3 sources · show sources ▾
ITERATION · Aug 20
First post-beta update adds multimodal input and subagent support across 14 changes
2 sources · show sources ▾
THE TURN · Aug 21
Harness adds Claude Code and Codex as sub-agents, pitching itself as the agent-era scheduling layer
The pivot from standalone framework to orchestrator of rival coding agents.
2 sources · show sources ▾
NOW · Aug 24–25
Deep dive: the plugin architecture, not the V4 Pro model, is the real story
2 sources · scout · show sources ▾
SECURITY FLAW · Aug 25
QiAnXin discloses a critical remote-code-execution flaw in Harness
1 source · scout · watcher updated status · show source ▾

What to watch — open questions

  • Has DeepSeek shipped a patch for the QiAnXin-reported RCE, and is it exploitable in a default Harness install or only under specific plugin configurations?
  • Does DeepSeek publish independent benchmarks for Harness's sub-agent orchestration, or only vendor/Chinese-press claims?
  • How stable is the plugin API across the rapid v0.1 → post-beta → weekly-update cadence — will early plugins break?
  • Does hosting Claude Code and Codex as sub-agents require API keys or licensing terms from Anthropic and OpenAI, or is it a wrapper around local CLIs?
  • Will English-language technical documentation and independent community adoption follow the current, largely Chinese-press coverage?
How this thread was built
scout surfaced 3editor wrote the arc · 9 beatswatcher 2 status changes

Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.