dsh-anchored-standard

中文说明

An experimental DeepSeek Harness agent preset that bootstraps the first model request with a Minimal-aligned prompt, the Minimal preset's real tool schema (bash + str_replace_editor), and no auto-injected workspace/skill context, then exposes the complete Standard tool catalog after the first durable tool call or reply.

This is a community project. It is not an official DeepSeek preset and is not affiliated with or endorsed by DeepSeek.

Why

DeepSeek V4 Pro conditions strongly on the API-visible tool catalog. In the Project2 evaluation, Standard and PTC produced scores of 91 and 92, while the official Minimal preset produced 99 and 96. Permanently staying on Minimal, however, gives up the Standard preset's broader tool set.

Anchored Standard separates initial trajectory selection from later tool use:

1. Keep the Minimal complete system prompt.
2. Expose the Minimal preset's REAL tool schemas — persistent bash + str_replace_editor, byte-identical to the official Minimal composition — on the first model request. Issue #11 measured this exact schema anchoring 5/5 runs at the adapter-default maxTokens (256000) with zero let me first-lines, while every standard-family schema (pwsh/read, pwsh only, sandboxed bash/read) fell into standard-like behavior 11/11. The tool schema identity is the decisive first-request variable at 256000, so no output cap is needed.
3. Strip the auto-injected context on that first request as well — the AGENTS.md/CLAUDE.md workspace digest and the available-skills reminder that true Minimal never mounts (suppressedContextSources in the tool-bootstrap row). User-initiated skill gestures are not filtered, and both injections return unchanged from request #2 on.
4. After the session records its first durable promotion signal — a tool/call or the first assistant/message, whichever comes first — expose all Standard tools. Request #1 always sees the bootstrap catalog; request #2 always sees the full catalog, so a text-only first reply can no longer trap the session in bootstrap. (promoteOn in the tool-bootstrap row selects the trigger: either default, tool-call, or assistant-message.)
5. Derive the phase from durable session events so resume and reload preserve it.

The bootstrap catalog is the same on every platform: the Minimal pair (bash/str_replace_editor). The preset's shell is the persistent PTY bash (the sandboxed Standard bash row is disabled — both register the bash name into the same layer, and the tools registry rejects duplicates; Windows never had the sandboxed bash anyway). pwsh remains available in the promoted catalog on Windows.

Results

Project2 V4.1b, DeepSeek V4 Pro, reasoningEffort=max, Windows native:

Run	Ability	Reasoning blocks	we	let's	let me	Visible replies
r1	98	193	179	88	1	1
r2	99	162	165	98	0	1

Both runs emitted exactly two tool-catalog snapshots: the two-tool Minimal bootstrap, followed by the 25-tool Standard catalog. The result is reproducible evidence for this task, not a claim of universal improvement across models or workloads.

Cross-version evidence (issue #11, Windows + official endpoint, first-request trajectory only): at the adapter-default maxTokens the Minimal tool schema anchored 5/5 (We need modify… first lines, we 1.4, let me 0.0), while pwsh/read, pwsh-only, and sandboxed bash/read all produced standard-like first lines 11/11 — the tool schema, not the output cap, is the decisive first-request variable at 256000.

Full methodology and aggregate evidence are in xiaobright/modeltest.

Compatibility

Developed and tested against:

• DeepSeek Harness 0.1.0-rc.5
• repository commit 47f9438
• Node.js 24 on Windows

On the 0.1.0-rc.5 source checkout, bootstrapMaxTokens reaches the actual first request (the first request/header records the cap, adapterDefaults stays empty), because llm.prepareCall only materializes a default maxTokens when the proposed config has none. One prebuilt profile package observed in issue #11 (CLI launcher reporting 0.1.0-rc.6) overwrote the proposed cap with adapterDefaults.maxTokens; there the cap is a no-op. The default composition therefore relies on the Minimal tool schema alone (which anchors at the adapter default with no cap) and leaves bootstrapMaxTokens as an opt-in for standard-schema bootstraps.

DeepSeek Harness is currently a developer preview and explicitly permits breaking changes. This preset is a full snapshot of the Standard composition, so review upstream changes before using it with a newer release.

Install

Clone this repository, then copy the entire preset directory into the user preset root under the id anchored-standard.

PowerShell:

$target = Join-Path $env:USERPROFILE '.dsh\.agent-presets\anchored-standard'
if (Test-Path -LiteralPath $target) { throw "Preset already exists: $target" }
New-Item -ItemType Directory -Force -Path (Split-Path -Parent $target) | Out-Null
Copy-Item -Recurse -LiteralPath '.\preset' -Destination $target

Linux/macOS:

dsh_home="${DSH_HOME:-$HOME/.dsh}"
mkdir -p "$dsh_home/.agent-presets"
test ! -e "$dsh_home/.agent-presets/anchored-standard"
cp -R preset "$dsh_home/.agent-presets/anchored-standard"

Fully restart DeepSeek Harness, create a blank session, and select Anchored Standard (experimental). Do not switch an active session from a different preset.

Verify

Export the session JSONL and inspect request/header events. Reproduction checklist (issue #11 asks for the first two explicitly, because both are the variables that decide the anchor):

• First-request config.maxTokens value: with bootstrapMaxTokens unset (the default), the first header records the adapter default (e.g. 256000 with adapterDefaults.maxTokens: true); with a cap configured it records the cap (e.g. 1024 with no maxTokens adapterDefault).
• First-request tool schema source: the first header's tools array must be exactly ["bash", "str_replace_editor"] — the official Minimal preset's real schemas, not Standard's pwsh/read.
• the first request's messages should contain no AGENTS.md/CLAUDE.md digest and no available-skills reminder — only the user message and the minimal persona system prompt;
• after the first tool call or the first assistant reply, the next changed header should contain the full Standard catalog;
• subsequent requests should keep that full catalog and restore the standard context injections.

Run the local zero-dependency tests with:

npm test

Important behavior

• With the default promoteOn: either, the session promotes after its first durable tool/call OR its first assistant/message, whichever comes first — request #1 sees the bootstrap catalog and every later request sees the full catalog. A text-only first reply therefore still promotes at request #2; set promoteOn: tool-call to restore the original behavior, where a first response that makes no tool call never promotes.
• A failed tool execution still promotes the session because the durable tool/call already exists.
• The first request's output budget is NOT capped by default: the Minimal tool schema anchors at the adapter-default maxTokens, so bootstrapMaxTokens is opt-in. When set, the first request is capped and the cap is explicitly stripped after promotion (the next request's seed proposal carries the previous header's maxTokens forward).
• The Minimal pair stays mounted after promotion, so the promoted catalog is the Standard catalog plus bash (persistent) and str_replace_editor — and the Standard sandboxed bash row is disabled in favor of the persistent shell (same tool name, same layer; see Why). The read/write/edit tools keep the sandboxed filesystem while str_replace_editor uses the preset's local fs.
• A missing bootstrap tool degrades to the full catalog with a one-time warning instead of failing requests, so a composition drift cannot brick a session; invalid promoteOn values fail at preset mount instead.
• Promotion decisions are memoized per session for the process lifetime; the durable event scan runs once per session per process.
• While a session is unpromoted, the pre-step filter strips messages whose source.kind is listed in suppressedContextSources (default: agent-instructions and skill-catalog, the two automatic injections Standard adds over Minimal). Set the list to [] to disable the context filter; add other source.kind values to suppress more. A filter failure degrades to keeping every message rather than eating context.
• The tool catalog changes once, so request-prefix cache continuity also changes once between the first and second model requests.
• The preset has the same trust level as shell access. Review its files before installation.
• The plugin performs no network requests and adds no telemetry.

Zero-Anchored Standard (experimental)

An extra test mode that does not change the Anchored Standard logic above. It uses the same Minimal-aligned system prompt, but instead of exposing two tools on the first request it injects one fixed zero-tool anchor turn:

1. When the user sends their first message, the anchor-turn plugin prepends a fixed user message — "This round is a test. Tools are not open yet; all tools will open next round." — ahead of it.
2. The first real model request carries ZERO tools, so the session's first reasoning chain follows the zero-injection "we" trajectory.
3. Once that anchor response is durable, the full Standard catalog is exposed and the real message proceeds with all tools.

Anchoring on the first message — not on session creation — keeps the blank-session preset switcher usable. Subagents always see the full catalog.

Measured behavior (opencode-go, DeepSeek V4 Pro, reasoningEffort=max): the anchor request is stable "we"-style with zero let me; the following tool-bearing requests return to the "The user wants…/Let me" style. This mode is a comparison point for whether the zero-tool first turn is worth the extra model call — not a claim that tool rounds stay "we"-style.

Install as a separate preset id:

dsh_home="${DSH_HOME:-$HOME/.dsh}"
mkdir -p "$dsh_home/.agent-presets"
test ! -e "$dsh_home/.agent-presets/zero-anchored-standard"
cp -R zero-anchored-standard "$dsh_home/.agent-presets/zero-anchored-standard"

Restart DeepSeek Harness, create a blank session, select Zero-Anchored Standard (experimental), then send your first message.

Official ecosystem guidance

DeepSeek currently asks community plugin authors to publish plugins in their own GitHub projects and add the dsh-plugin repository topic for discovery. The official repository does not currently accept external pull requests and does not mandate a community repository template. See the official CONTRIBUTING.md.

License

MIT. preset/agent.cordis.yml is derived from the DeepSeek Harness Standard preset; the original DeepSeek copyright and MIT notice are retained in NOTICE.