Better Dsh — Dashr

DASHR is a persistent-kernel REPL for the DeepSeek Harness (dsh): a stateful ctx.replRuntime provider (one persistent IPython kernel subprocess per session, the run's principal, held in a map inside the one service instance per mount). Each run() is one cell on the calling session's kernel; variables, imports, and definitions assigned in run N survive into run N+1 — state codification (blueprint §1.1 channel ②), deliberately NOT the per-run isolation a one-shot execution backend provides — and two sessions sharing one service instance never see each other's variables.

Naming (v0.1.5 layer model): the runtime class is DashrRuntime — the standing-mount-layer component (one instance per mount holding the cross-session kernel map), and therefore the de facto daemon while the profile-level DashrDaemon concept stays an empty shell; the session layer is the ipykernel subprocess itself, a pure interpreter with no harness awareness.

The presentation half — the eval transport tool, the Python SDK renderer, and the tool→binding bridge — lives in the SAME package (merged in v0.1.8; the pre-merge dsh-rlm-mode/dashr-presentation sibling split is gone). The runtime registers the service key replRuntime through its own vendored Service Definition (see src/vendored/repl-runtime.ts), so it carries zero dsh runtime package dependencies: only @deepseek-ai/cordis (peer), schemastery, and zeromq.

Package positioning

• npm name: better-dsh (unscoped; formerly @pgmi-builds/better-dsh, which now stays published only as a deprecated pointer to this name).
• A standard Cordis plugin (Context + schemastery Config, every tunable configurable from cordis.yml, no hardcoded tunables).
• "Registrations are effects": the kernel lifecycle (lazy spawn on a key's first run(), teardown on that session's agent/disposed, optional snapshot at either teardown) is effect-owned, so plugin disposal tears every subprocess down.
• Published surface: lib/ only (files: ["lib"], main/types/exports pointing at lib/index.js / lib/index.d.ts) — the build emits beside the manifest (outDir: 'lib'; the tsdown default dist/ left the exports map dangling, fixed in M2B). The root also re-exports the vendored Service Definition's public contract (ReplRuntime plus the CodeRun* / CodeBinding* / CodeJsonValue types) so consumers depend on the published shape instead of reaching into sources. The declaration keeps dependency imports external (dts: { resolve: false }): bundled copies would create duplicate type identities in a consumer's program.

Install

▼
sh
复制代码
npm install better-dsh

The runtime OWNS its kernel environment. With python left at the python3 sentinel (or absent), it provisions a managed venv under the package (.venv-kernel) on first use, installing ipykernel + dill (CPython 3.11) — no /tmp symlink, no blind trust in a host python3. An explicit python (or DASHR_KERNEL_PYTHON) is verified instead. For development and tests, pre-create the venv:

▼
sh
复制代码
npm run kernel:venv        # uv venv .venv-kernel --python 3.11 + ipykernel + dill

Tests pick the kernel interpreter from DASHR_TEST_PYTHON, falling back to ./.venv-kernel/bin/python, then python3 — see test/helpers.ts.

Configuration

Every field of the plugin Config (schemastery defaults shown):

Field	Default	Meaning
python	python3	Explicit interpreter with ipykernel; the bare python3 sentinel (or absent) selects a managed venv under kernelEnvDir.
startupTimeoutMs	30000	Budget for kernel spawn → ready, in milliseconds.
runTimeoutMs	120000	Wall budget per run; expiry interrupts the kernel then force-settles.
interruptGraceMs	2000	Grace between a timeout/abort interrupt and the force-settle.
interruptConfirmMs	250	Confirm window between the control-channel interrupt and the SIGALRM escalation (must be < interruptGraceMs); see "Interrupts" below.
disposeTimeoutMs	5000	Budget for graceful kernel teardown (shutdown_request → SIGKILL).
snapshotTimeoutMs	30000	Budget for internal snapshot/restore cells (dill dump/load).
maxOutputBytes	67108864	Hard cap for serialized log-array, completion-value, and failure-message payloads.
snapshotDir	(unset)	Base directory for per-session namespace snapshots (<dir>/<principal>/state.dill + manifest.json); none when absent.
snapshotSizeCapBytes	268435456	Serialized-size cap for a turn-end snapshot; over-cap snapshots are skipped (one-time model warning).
username	dashr	Jupyter username stamped on wire messages.
kernelEnvDir	(unset)	Managed venv directory (defaults to <package>/.venv-kernel).
kernelPythonVersion	(unset)	Preferred CPython version for a managed venv (default 3.11).
kernelAutoInstall	true	Provision the managed venv (ipykernel + dill) on first use.

Persistent-state semantics

• Cell semantics: each run({ program }) is one cell on the calling session's kernel namespace (user_ns) — a pure IPython REPL. Top-level await works; a top-level return is a SyntaxError, exactly as in a native IPython cell. The completion value is the LAST expression's value (REPL displayhook-style): a statement-ending cell or a None final expression yields no value field, and a non-JSON value comes back as its repr text.
• Session keying (M3-A): one kernel per distinct request.principal (the presentation bridge passes the calling agent's session id); runs without a principal share one default key, preserving M1 semantics. The service instance count is unchanged — one per mount — the keying is a Map<principal, kernel> inside the provider.
• Kernel lifetime: lazy start on a key's first run(); teardown when that session's agent is disposed (the dsh agent/disposed event, payload { agent: { id } }, listened through the untyped cordis event service to keep this package's zero-dsh-dependency rule) and on plugin disposal (shutdown_request, then SIGKILL after disposeTimeoutMs). A kernel that dies unexpectedly is never reused in-process: it respawns onto its nearest replayable snapshot (or a fresh empty kernel when none exists) and the run that observed the death gets an explicit worker-exit naming what was lost.
• Turn-end snapshots (M3-B): with snapshotDir configured, every successful run is followed by a size-capped snapshot cell that dumps the user namespace to <snapshotDir>/<principal>/state.dill + manifest.json (turn, pythonVersion, venvPath = the kernel's own sys.executable, skills, names, sizeBytes). A namespace whose serialized size exceeds snapshotSizeCapBytes is skipped — estimated BEFORE any dill IO by a bounded walk that reads numpy/pandas in-memory footprints, then confirmed against the actual .part dump — and the model is warned once through the run's own logs. Skipped snapshots never replace the previous good one.
• Restore-on-first-boot (M3-B): a key's first kernel boot restores its on-disk snapshot before running user code. The kernel validates the manifest itself (python version, interpreter identity, skills); a non-replayable snapshot degrades to an EMPTY namespace and the first run tells the model so. Variable state and the append-only transcript are NOT transactionally consistent (blueprint §8.3): the snapshot is a point-in-time namespace capture that can lag the transcript, and a degraded restore never fabricates variables the transcript once saw.
• Interrupts (M3-A hardened): aborts/timeouts escalate in two phases — the zmq control interrupt_request first, then SIGALRM only after interruptConfirmMs if the cell has still not settled. The kernel-side bootstrap installs a busy guard that only raises KeyboardInterrupt while a dashr cell is actually executing, so a signal landing on an idle or booting kernel is swallowed instead of terminating the process (the M1 same-tick dual send killed idle kernels deterministically — 10/10 same-tick, 8/10 at +1-2ms, 40/40 during cold boot; see test/interrupt-race.spec.ts). The hard-abort contract is intact: a busy while True: pass still breaks inside the grace (blueprint §10.4).
• Concurrency: the bridge serializes cells per kernel (executeCell awaits the previous cell), so concurrent run() calls on one session queue rather than interleave; see test/parallel.spec.ts. Runs on DIFFERENT principals execute on their own kernels concurrently.
• What snapshots do NOT carry: nothing outside the kernel namespace is ever in scope — the v0.1.8b removal deleted the Continual Harness (refine(), harnessDir) entirely, so the snapshot/restore cycle's only persistence channel is the kernel namespace itself (blueprint §8.4). Anything a cell stores in ordinary variables follows the snapshot rules above as before.

Testing

▼
sh
复制代码
npm install
npm run kernel:venv     # once; or export DASHR_TEST_PYTHON=/path/to/python
npm run typecheck       # tsc --noEmit
npm test                # vitest --run (fileParallelism: false)

Teardown discipline: every test context is disposed through onTestFinished, and CI must assert no orphan kernels remain:

▼
sh
复制代码
pgrep -cf -- '-[m] ipykernel_launcher' || echo no-orphans

(The -[m] trick prevents pgrep from matching itself; 208 orphaned kernels once exhausted machine memory while every unit test stayed green — blueprint §10.8/§10.9.)

---

Presentation half (eval transport, SDK, bindings)

The DASHR agent-plane presentation (blueprint §7.4): the plugin half that presents the persistent-kernel runtime to the model as cells on one persistent IPython kernel. It contributes:

• eval — the cell transport tool: one call = one cell on the calling session's kernel (the ctx.replRuntime service). Variables, imports, and definitions survive across calls. Nested tool calls ride the host registry's native scheduling pipeline as await tool.name({...}) inside the cell — one positional arguments object per call, keyword arguments rejected. Every registry-visible tool is bound as a member; the send_message bridge callable joins them in the same catalog block.
• dashr:tool-catalog — a generated Python SDK prompt section: one async def name(args) -> Output per visible tool (flat tool.* shape) plus the cell contract (persistent namespace, completion-value rules, ToolCallError, sub-call concurrency).
• The model-direct collapse — an assembly filter leaves eval the only contributed tool schema, and a monotonic guard denies a model-direct call naming anything else with the route back into a cell. Both are scoped to the mounting composition, so a PTC (native Code Mode) preset in the same process keeps its own presentation.
• Masking (ADR-0002) — exactly two upstream tool names are displaced from the model's surface (send_message and the child-scoped report), collapsed into the single dual-direction send_message bridge. Every other delegation tool (subagent, subagent_fork, list_agents, interrupt_agent, workflow, ralph) stays directly exposed as a native tool.* binding — the model calls it exactly as the host ships it. Masking is presentation-only: the registry is never touched.

Removed in v0.1.8b: the refine Continual Harness and the compact REPL bridge (harness/refine became third-party territory; context compaction is the host runtime's business, not the REPL's).

Install

The canonical path is the repo-root one-click installer (install.sh): it installs the plugin from the npm registry and notes the restart. The equivalent manual steps, for reference:

▼
sh
复制代码
# 1. plugin — the pnpm registry-metadata cache can lag a fresh npm publish
#    by minutes-to-hours, so drop it first, or `@latest` resolves the OLD
#    version right after a release:
rm -rf ~/.cache/pnpm
dsh plugin --profile web add --config.auto-install-peers=false better-dsh@latest

# 2. restart the running instance:  systemctl --user restart dsh

Notes: --config.auto-install-peers=false is MANDATORY — the profile already resolves @deepseek-ai/* peers through the harness install; letting the package manager auto-install them adds a second, divergent copy of cordis and friends. The version is deliberately unpinned (@latest).

Coexistence with a PTC Code-Mode session

eval is our own transport name (the registry reserves run_code), so a Code-Mode preset (@deepseek-ai/dsh-agent-tool-presentation with mode: code over the host-plane worker-thread codeRuntime) composes beside a DASHR session in one process: the PTC agent's assembly shows run_code plus the TS tools:sdk section, the DASHR agent's shows eval plus the Python dashr:tool-catalog, and neither execution path touches the other's runtime.

Composition

▼
ts
复制代码
import dashr from 'better-dsh'

ctx.plugin(dashr, { maxParallelSubCalls: 10 })

One plugin, one row: apply mounts DashrRuntime first (providing ctx.replRuntime), then the presentation half injects on replRuntime — a composition without the runtime fails AT MOUNT, named in the preset's activation audit, instead of at the first prompt. The dashr bundle patch mounts the row on the HOST plane, so every agent in every preset sees the eval tool (agent → preset → global).

Delegation and messaging

The upstream delegation tools stay REGISTERED, EXECUTABLE, and directly bound — subagent, subagent_fork, list_agents, interrupt_agent, workflow, ralph are ordinary tool.* members the model calls exactly as the host ships them. Two names are displaced by the single send_message bridge (ADR-0001): upstream send_message (the parent→child downlink) and the child-scoped report (the child→parent uplink) collapse into one dual-direction channel:

• await tool.send_message({"receiver": "child", "message": ..., "subagent_id": ...}) delivers down through the tool layer;
• await tool.send_message({"receiver": "parent", "message": ...}) reports up through the SERVICE layer (ctx.subagents.reportFrom(...)) — live continuable children only; a root agent gets a structured UNAUTHORIZED.

Every binding returns structured JSON; errors are a FIELD on the result, never a host crash — a missing ctx.subagents service, a depth cap, an unknown id, or an infrastructure rejection all map to an error string.

Config

Field	Default	Meaning
maxParallelSubCalls	10	Cap on one cell's overlapping sub-calls (native scheduler contract; 1 = strictly serial).
(runtime keys)	—	The runtime slice (python, snapshotDir, timeouts, caps, kernel-env knobs) is documented in the Configuration table above.

Tests

▼
sh
复制代码
npm install
npm run kernel:venv     # once; or export DASHR_TEST_PYTHON=/path/to/python
npm run typecheck
npm test                # vitest --run

The suite needs a Python interpreter with ipykernel (+ dill for the snapshot tiers) for the real-kernel specs; it resolves one from DASHR_TEST_PYTHON, then ./.venv-kernel/bin/python, then python3 — see test/helpers.ts.

Relationship to upstream

Structure mirrors @deepseek-ai/dsh-agent-tool-presentation and the Code Mode half of @deepseek-ai/dsh-tools (0.1.1-rc.2), re-pointed at the vendored replRuntime Service Definition. See the module docs in src/index.ts for the deliberate deltas (eval vs run_code, ordinary scoped registration, guard-based collapse, mirrored tools/code-dispatch-log waterfall).