From f2866508c4510109621657078d74a86a67005fe7 Mon Sep 17 00:00:00 2001 From: Eric Wang Date: Thu, 13 Aug 2026 10:18:17 -0700 Subject: [PATCH] docs: add 'dsc and dsh' positioning section DeepSeek launched dsh (deepseek-ai/deepseek-harness, developer preview) on 2026-08-13. State the relationship honestly: dsh is the official harness and dsc does not compete with it. dsc's lane is a minimal auditable agent, a research vehicle with measured (including negative) findings, and a cross-harness testbed. Interop: dsh reads SKILL.md skills from ~/.agents/skills and /.agents/skills (verified in dsh's skill-filesystem package); dsc's planned M7 skills support targets the same layout. Phrased as planned, not shipped. Co-Authored-By: Claude Fable 5 --- README.md | 35 +++++++++++++++++++++++++++++++++++ 1 file changed, 35 insertions(+) diff --git a/README.md b/README.md index 4dd49f4..53abc0f 100644 --- a/README.md +++ b/README.md @@ -16,6 +16,41 @@ sessions/compaction, sub-agents, and interactive mode work and are eval-gated; the job scheduler works and is test-gated (the default prompt is byte-unchanged by it). Interfaces will move. +## dsc and dsh + +DeepSeek released an official agent harness on 2026-08-13: +[dsh](https://github.com/deepseek-ai/deepseek-harness) +(`npx @deepseek-ai/dsh`, developer preview, breaking changes +promised). dsc does not compete with it. If you want the official, +full-featured harness — plugin architecture, Web UI, the surface +DeepSeek will keep building on — use dsh. + +dsc's lane is narrower and stays that way: + +- **A minimal, readable DeepSeek-native agent.** Zero runtime + dependencies, a few thousand lines, one protocol. +- **A research vehicle that publishes measured results, including the + negative ones.** Fan-out does not buy speed (33.0s → 96.7s and + 70.2s → 97.5s measured; what it buys is ~30% fewer parent input + tokens and N independent context budgets). Interactive mode is not + a token-cost win over `-p --resume` — the prefix cache is + server-side and cross-process. Delegation timing decides whether + sub-agents pay (before investigating: 1.30x faster; after: 2.08x + slower). Reasoning tokens bill against `max_tokens`. These are + properties of the API, not of dsc — they hold in any harness, + dsh included. +- **A cross-harness testbed.** Small enough to change wire-level + behavior in an afternoon and measure the difference. + +Interop note: dsh discovers SKILL.md skills in `~/.agents/skills` and +`/.agents/skills`; dsc's planned skills support (milestone 7 +in the build order) targets the same layout, so one skills directory +should serve both. Planned, not shipped. + +Short version: use dsh for the official, full-featured harness. Use +dsc when you want a minimal auditable agent, wire-level experiments, +or the measured findings in `eval/BASELINE.md` and `docs/devlog/`. + ## Install / run Requires [Bun](https://bun.sh) and a DeepSeek API key.