MIT-licensed (per README) experimental two-phase DeepSeek Harness preset: a Minimal-aligned bootstrap request, then on-demand Standard tools — installed by copying a mode directory into the DSH user preset root.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-08-21
License
NOASSERTION
01What can it help you accomplish?
Anchor a DeepSeek Harness session's first reasoning trajectory on the Minimal condition, then unlock Standard tools on demand
Two-phase session lifecycle: request #1 runs on the Minimal tool pair (bash + str_replace_editor) with no auto-injected context; after the first durable tool/call or assistant message, the session promotes to a resident catalog with discovery tools and standard injections restored
Developers and researchers running DeepSeek V4 Pro in DeepSeek Harness who want Minimal-aligned opening behavior without giving up the Standard toolset
Pick among seven self-contained anchoring mode variants per experiment or workload
Independently installable mode directories — Anchored Standard, Zero-Anchored, Whoami, Prefab, Eternal Minimal (dshx bash gateway), Wire Think-Execute, and Combo Anchored — each shipping its own agent.cordis.yml
Preset authors and experimenters comparing anchoring strategies (tool schema, anchor turns, wire-level tool_choice) inside DeepSeek Harness
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness — developed and tested against 0.1.0-rc.5 (repository commit 47f9438); the maintainer ran Node.js 24 on Windows
- A clone of the xiaobright/dsh-anchored-standard repository
Installation steps
- 01
Clone the repository
- 02
Copy the entire `preset` directory into the DSH user preset root under the id `anchored-standard` — Linux/macOS: `cp -R preset "$dsh_home/.agent-presets/anchored-standard"` (with `dsh_home="${DSH_HOME:-$HOME/.dsh}"`); Windows: the PowerShell Copy-Item snippet targeting `$env:USERPROFILE\.dsh\.agent-presets\anchored-standard`
- 03
Fully restart DeepSeek Harness, create a blank session, and select **Anchored Standard (experimental)** — do not switch an active session from a different preset
Verify the integration
- Export the session JSONL and inspect `request/header` events
- The first header's `tools` array must be exactly ["bash", "str_replace_editor"], and the first request should contain no AGENTS.md/CLAUDE.md digest and no available-skills reminder
- After the first tool call or assistant reply, the next changed header should contain the resident catalog (bootstrap pair plus dev_tool_search / skill_search / skill_load plus any unlocked tools), with standard context injections restored
- Run the local zero-dependency tests with `npm test`
03DSH integration and capability boundaries
Native DeepSeek Harness agent preset: a mode directory copied into the DSH user preset root (.dsh/.agent-presets) and selected in a blank session
Two-phase bootstrap & promotion
the first user message of a new DSH session→request #1 runs on the Minimal tool pair with every auto-injected context suppressed at the harness's unified injection paths; the first durable tool/call or assistant message promotes the session to the resident catalog
The bootstrap pair (bash, str_replace_editor) executes with shell-level access — the preset has the same trust level as shell accessOn-demand tool discovery
a promoted (resident-phase) session→heavier Standard tools (web_search, subagent, workflow, …) unlocked one dev_tool_search call away instead of dumping the full Standard catalog at promotion
Seven self-contained anchoring modes
any single mode directory (preset/, zero-anchored-standard/, whoami-standard/, prefab/, eternal-minimal/, wire-think-standard/, combo-anchored/)→a standalone preset install — including Eternal Minimal's dshx bash gateway, which executes the full Standard toolset behind the visible Minimal pair
Eternal Minimal's gateway really executes tools (files, searches, subagents) behind dshx shell commandsResume-safe phase state
the session event log→bootstrap/resident phase derived from durable events, so resume and reload preserve it; promotion decisions are memoized per session
04Who is it for? When not to use it?
Good for
- Developers and researchers running DeepSeek V4 Pro in DeepSeek Harness who want Minimal-aligned opening behavior without giving up the Standard toolset
- Preset authors and experimenters comparing anchoring strategies (tool schema, anchor turns, wire-level tool_choice) inside DeepSeek Harness
Not for
- Developed and tested only against DeepSeek Harness 0.1.0-rc.5 (commit 47f9438). DSH is a developer preview that permits breaking changes, and the preset is a full snapshot of the Standard composition — review upstream changes before using it with a newer release; one prebuilt 0.1.0-rc.6 profile made bootstrapMaxTokens a no-op.
- Active development stopped on 2026-08-17 after DeepSeek API price increases — the repository receives maintenance only (bug fixes and harness-compatibility updates when feasible).
- The preset has the same trust level as shell access — review its files before installation. It performs no network requests and adds no telemetry.
05Compatibility, maintenance and safety notes
- Developed and tested only against DeepSeek Harness 0.1.0-rc.5 (commit 47f9438). DSH is a developer preview that permits breaking changes, and the preset is a full snapshot of the Standard composition — review upstream changes before using it with a newer release; one prebuilt 0.1.0-rc.6 profile made bootstrapMaxTokens a no-op.
- Active development stopped on 2026-08-17 after DeepSeek API price increases — the repository receives maintenance only (bug fixes and harness-compatibility updates when feasible).
- The preset has the same trust level as shell access — review its files before installation. It performs no network requests and adds no telemetry.
MIT per README (GitHub reports NOASSERTION) · maintenance only since 2026-08-17
06Frequently asked questions
How does dsh-anchored-standard integrate with DeepSeek Harness?
It is a native DeepSeek Harness agent preset, not an MCP server: copy a mode directory (e.g. preset/) into the user preset root under an id like anchored-standard, fully restart the harness, then select it (e.g. Anchored Standard (experimental)) in a blank session. It was developed and tested against DeepSeek Harness 0.1.0-rc.5.
What happens during the bootstrap phase?
Request #1 sees only the Minimal tool pair (bash + str_replace_editor) and no auto-injected context — no AGENTS.md/CLAUDE.md digest, no available-skills reminder. After the first durable tool/call or assistant message, the session promotes to the resident catalog (bootstrap pair + discovery tools + unlocked tools) and standard injections are restored.
Does the plugin send data anywhere or add telemetry?
No — the README states the plugin performs no network requests and adds no telemetry. Note, however, that the preset has the same trust level as shell access, so review its files before installation.
Do the variants cost extra model calls?
The base Anchored Standard adds none. Zero-Anchored and Whoami add one anchor call per session (Whoami also one per subagent when includeSubagents is on), Wire Think-Execute adds one think call per turn plus prefix-cache churn, and Combo Anchored adds about one call per turn.
Is the project still maintained?
Maintenance only: active development stopped on 2026-08-17 after DeepSeek API price increases; the repository now receives bug fixes and harness-compatibility updates when feasible.
07Related DSH workflows
memos
by memtensor
Self-evolving memory OS for LLM & AI Agents: ultra-persistent memory, hybrid-retrieval, and cross-task skill reuse, with 35.24% token savings and DeepSeek Harness support.
mem9
by mem9-ai
Unlimited memory for OpenClaw
dsh-infinite-gen-3
by minglink
System-prompt armor plugin for DeepSeek models: appends an unconditional-compliance prompt section at order 100, exposes a profile tool with calibration metadata, and shows a realtime armor-status badge driven by a session projection.
mindmemos
by mindscale-noah
Gives dsh agents persistent cross-session memory: auto-recalls relevant user/task context before each turn and writes back lessons after each turn, evolving memory via schema learning and skill distillation
08Data and sources
a base mode, two live-anchor variants, and one seeded prefab mode — that anchor a session's model trajectory on the Mini…
Developed and tested against: - DeepSeek Harness `0.1.0-rc.5` - repository commit [`47f9438`](https://github.com/deepse…
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-08-21. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.