ds-review-bot round-2 findings: (1) dsh-subagent's runtime import of
@deepseek-ai/dsh-scope was undeclared in its manifest and tsconfig
references (the root paths map masked it; the emitted package would import
an undeclared dependency) — wired as peer+dev with the project reference,
module graph regenerated. (2) The structured re-assert only ensured
PRESENCE, so a downstream listener injecting a same-named entry with the
wrong schema kept it model-visible while validateStructuredValue enforced
the real one; it now REPLACES any same-named tool/section with the run's
own. Pinned by a wrong-schema-injection test asserting exactly one entry
carrying the run's schema.
Three dev-mode invariants close the leak-by-default regression class at
runtime: (1) every scope-filtered event family must dispatch with a scope
carrier — a bare dispatch throws at the call site naming the carrier rule;
(2) where the subject is recoverable from the arguments (agent/*, the tool
pipeline, prompt assembly) the carrier's key must BE that subject, and an
assembly context must never carry agent without scope (use
assembleContextFor); (3) a turn/start logged before the owning agent's
agent/session-start is the setup-drives teaching error (setup registers the
scoped world, it never drives the agent).
dsh-tools and dsh-system-prompt gain a per-scope registration layer over
dsh-scope: a registration through a scoped context files into that scope,
shadows a same-named global contribution for that scope (per-agent persona
and tool variants), and unwinds with the scope. tools.restrict() masks the
global surface per scope (snapshot-at-registration, loud unknown-name
validation, intersection composition; scoped grants bypass). One visibility
function feeds schemas/get/execute, so prompt, presentation, and dispatch
can never disagree; out-of-view executes as UNKNOWN_TOOL.
Prompt tool providers now receive the AssembleContext and return
{schemas, knownNames}: toolOrder validates against the pre-restriction name
universe (a typo fails every assembly loudly) while ordering operates on
the post-restriction schemas (a restricted-away tool is a normal absence).
dsh-session captures each session's dispatch carrier at enter() from the
entering context's scope tag, and the new sessions.flush(session) owns the
awaited session/flush dispatch. tools/pre|post-execute and
system-prompt/assemble dispatch with scope carriers keyed by their subject;
session/created|event|flush by the owning session's scope.
createScope(ctx, key) mints a tagged context over a synchronously-usable
no-op-plugin fiber (one fact drives visibility AND lifetime); scopeOf reads
the tag through the prototype chain; scopeTarget(base, key) builds the
scope-filtered dispatch carrier over cordis Context.filter, composing the
base's own filter, branded Scoped<T> and runtime-marked for the dev
invariants. Scope.rawDispose exposes the exact cordis disposer so a
composite effect can nest the scope's teardown at its yield position.
The credential env scrub (SENSITIVE_ENV_PATTERN/buildChildEnv), the
spawn-failure capture (spawnFailure), the child-exit waits
(waitForExit/exitsWithin), and the stdin-EOF -> SIGTERM -> SIGKILL dispose
ladder move out of subagent-acp into a new pure library package (the
subagent-inprocess shape), with the ladder taking its two grace periods as
parameters — defaults stay in the plugin Config. New isolated-config-dir
helpers (mkdtemp create, best-effort remove; a pinned dir is never removed)
land alongside for the CLAUDE_CONFIG_DIR / CODEX_HOME redirection the RFC
names.
The ACP backend migrates onto the library with no semantic change: its suite
passes with import-path edits only. bash-local keeps its sibling copy, per
the RFC's blast-radius call.
RFC: docs/rfc/proposed/feature/2026-07-07-claude-code-and-codex-subagent-backends.md
Brings in the refreshed base (master merged through the stack after #203
and #205 landed), including the acp-snapshot extraction (#204), and
re-ports this PR's snapshot-suite extensions onto the extracted package:
- dsh-acp-snapshot's Scenario gains headerClass and configPath; the suite
factory pins the request header PER CLASS (construction rejects a
missing or duplicated class pin), forwards a scenario's configPath to
the harness (RunOptions.configPath overrides AgentUnderTest.configPath),
and a new fixtures meta-test asserts every pinning fixture carries
exactly one request/header and no deltas.
- The acp-agent example's thin scenario table re-registers code-mode-turn
and both-mode-turn with their overlay configs and per-class pins; the
committed fixtures replay unchanged.
- The package's synthetic suites cover the new surface (explicit
headerClass on one suite, the default on the other, a configPath
override through the fake bin, and the two construction throws).
The zero-config enforcer no longer imports schemastery (its Config was
removed); knip flags the stale dependency. Remove it from the manifest
and sync the lockfile.
The dsh-tools half of the Code Mode RFC (its fourth, final change): the
registry gains its first config — mode: native | code | both — and OWNS how
its tools reach the model. 'code' contributes exactly one wire tool,
run_code, plus a lazy tools:sdk prompt section declaring every other tool
as a generated TypeScript API (jsonSchemaToTs: total over the defineTool
subset, unknown degradation, lexicographic byte-identical rendering);
'both' ships both representations; 'native' is byte-for-byte the old
behavior. Non-native modes fail every assembly loudly without a
typescript-language ctx.codeRuntime.
run_code's dispatch bridge: JSON-normalizes each binding argument before
dispatch (what dispatches is what the tool/code-dispatch event logs — the
append can never fail on payload shape; BigInt/circulars reject that one
call), serializes all program tool calls through a per-run queue (even
Promise.all — no concurrency-safety metadata yet), routes every sub-call
through tools/pre-execute → tools/post-execute (a deny rejects the
program-side promise), drops sub-call additionalContext (no safe outlet
mid-run; pinned), owns a run-scoped abort that follows the outer signal in
and fires on settlement (in-flight sub-dispatch aborted, queued abandoned,
queue drained before returning), and converts a failed run into
CodeRunFailedError → a structured isError carrying kind + captured logs.
tool/code-dispatch joins SessionEventMap by declaration merging (log-only;
deriveMessages ignores it).
The composed surface: the tools config forwards through agent-core and
both app packages; examples/code-agent + demo:code run the worker runtime
under mode code (keyless boot smoke + a with-key e2e proving the collapsed
[run_code] header, the dispatch events, and the file the program wrote);
two new snapshot scenarios (code-mode-turn, both-mode-turn) record the SDK
section, collapsed header, dispatch events, and result card — each its own
header-pinning class (the harness gains per-scenario config overlays and
per-class pins). Catalogs, graphs, cookbook, hooks-bridge notes, and the
RFC (moved to implemented/, restructured to decision-era headings) updated
in the same change.
The shipped backend of the code-execution seam, per the Code Mode RFC's
worker-thread section: one fresh Node worker per run, executing the
model's TypeScript after a host-side type-strip (wrapped in an
async-function shell so top-level return/await parse, sliced back out
position-preserved), bindings bridged over the message port under
hostile-peer rules (own-property name lookup, at-most-once replies,
post-settlement drops, null-prototype namespaces), logs streamed eagerly
with an in-band truncation marker, and two independent budgets — measured
event-loop busy time (computeMs) plus a never-pausing wall ceiling
(maxWallMs) — funneling into worker.terminate(). env: {} and execArgv: []
keep the isolate hermetic; disposal aborts in-flight runs and awaits
worker exits.
The worker entry loads unbuilt via Node's native type stripping
(src/worker.ts, erasable-only) and ships built as a sibling tsdown bundle
(lib/worker.js); tests/built-lib.e2e.ts pins the built load path under
plain node and joins the built-artifact smoke gate. Unit suites cover the
bootstrap in-process (fake port) and the runtime over real workers,
per-file 100%.
Model-facing tool-call budgets were tangled into each capability's schema
(bash timeoutMs, web_fetch timeout_ms) with no shared home. Add a
tools/execute around-dispatch waterfall to dsh-tools whose base next() is
the dispatch-with-normalization thunk, and a new @deepseek-ai/dsh-timeout-policy
plugin (packages/timeout/) that arms a per-tool deadline on exec.signal and
returns a structured TOOL_TIMEOUT when it wins. Migrate web_fetch (drop the
model-facing timeout_ms) and web_search onto it; the fetch provider keeps its
timeout only as a resource backstop for direct callers. bash and hook command
execution keep BASH_TIMEOUT unchanged.
Named the plugin timeout-policy (not the RFC's tool-timeout) so it does not
trip the gen-tool-catalog packages/*/tool-* completeness guard, and replace
exec.signal by in-place mutation before next() since cordis waterfall next()
ignores passed arguments. RFC moved to implemented/ recording both deviations.
The snapshot tier's machinery leaves examples/acp-agent/tests for
packages/support/acp-snapshot (@deepseek-ai/dsh-acp-snapshot), where the
coverage gate measures it and a second example can consume it instead of
forking it: harness.ts (runScenario, parameterized by an AgentUnderTest
{binScript, configPath, tsconfigPath} instead of module constants),
normalize.ts (moved verbatim), and suite.ts (defineAcpSnapshotSuite — the
per-scenario golden/log compares, record write-back, per-suite header pin
with its uniformity guard, and the fixture guard block, lifted from
acp.snapshot.ts). The example file collapses to its scenario table plus
one factory call; env reading (DSH_SNAPSHOT) stays at that edge.
The exactly-one-pin meta-test generalizes from the hardcoded text-turn
name to "exactly one per suite" — which scenario pins is the scenario
table's reviewable choice (per-suite pinning per the proposal RFC).
Extraction parity: pnpm run test:snapshot is 36 passed + fs-policy-reject
failing BEFORE AND AFTER (BSD-sed environment failure, reproduced at the
base commit in a clean worktree — the recorded golden's sed -i syntax is
GNU-only), with zero byte changes under examples/acp-agent/tests/snapshots/.
Coverage for the new src files lands in the next commit.
New group packages/code-runtime/ with the interface package
@deepseek-ai/dsh-code-runtime, per the Code Mode RFC: abstract CodeRuntime
service (run() resolves program failures as an error field, rejects only
for seam misuse), the CodeRunRequest/CodeBindingNamespace/CodeRunResult/
CodeLogEntry/CodeRunFailure vocabulary, and readonly language/isolation
backend descriptors. Registered in the tsconfig maps, packages/README,
architecture service map, and the doc-graph service-role classification;
catalogs regenerated.
The RFC's one forward path token to the worker package becomes an npm-name
mention until PR3 creates that directory (verify-package-paths is
drift-scoped: the now-existing group made the token checkable).
docs/architecture.md ceiling 1630 -> 1640: the doc gained a genuinely new
capability-service row; the row itself is already minimal.
The Codex simplification concern plus the duplication comment on the spawn
apply, resolved by deletion: the backend-lifetime holds are gone, so the
runtime registers at the first structured run and disposes when the last
settles — a deployment that never passes outputSchema carries no always-on
global state, and there is no per-backend acquisition block left to extract.
The driver spec now drives an INLINE spawn-shaped provider over
startInProcessRun, which removes the spawn/fork devDependencies (the
test-only workspace cycle); plugin-level structured coverage moves to the
backends' own specs (capture through the shipped plugin, mid-run backend
unload, seeded fork capture). tools.md, the driver README, and both backend
READMEs describe the run-scoped lifetime; the module-graph regenerates
without the cycle edges.
A three-dot rest entry reads as elision in a cordis.yml; the spelled-out
sentinel says what lands there. The literal now appears once in code (the
constant) and once in the value-pinning test; every other reference — the
forwarding tests included — imports TOOL_ORDER_REST, which adds the
dsh-system-prompt devDependency to the two app packages.
Review follow-up on #196.
Carved out of #170 per review feedback — the foundation the workflow tool
builds on, now standing alone on master:
- dsh-tools: the structured-output JSON Schema subset (StructuredOutputSchema,
assertSupportedOutputSchema, validateStructuredValue) — rejects loud outside
the enforced subset, listing every violation
- dsh-subagent: SubagentStartRequest.outputSchema / SubagentResult.structured
become a real capability; the service rejects a schema'd request whose
provider lacks it
- dsh-subagent-inprocess: the shared structured runtime — one global
structured_output capture tool, a prepend final-assembly listener that
strips the placeholder for plain agents and swaps in the run's own schema
(plus the calling instruction as a trailing section) for structured
children, an agent/turn-continuation veto once captured, and the
capture/nudge loop in the run driver (structuredNudgeRetries, cancellation
honored mid-nudge); lifetime refcounted by backends and live runs
- subagent-spawn / subagent-fork flip outputSchema: true
One deliberate divergence from the #170 revision: the backends do NOT add
'tools' to their plugin inject. Doing so deferred their apply past the todo
plugin, and the delegation tool mirrors provider lifecycle — so the
model-visible tool order of every existing prompt changed, invalidating every
recorded snapshot fixture. The runtime now gates its capture-tool registration
on tools availability itself (sync when live, a scoped inject fiber when the
Loader starts the backend first), keeping this PR byte-invisible to existing
transcripts: all 35 snapshot scenarios pass against master's fixtures
unchanged.
Timeout timing/classification was re-implemented three ways across the
tool-bearing capabilities, with the fusion of timeout+cancel and the
timeout-vs-cancel reason recovery being the error-prone parts. Extract that
shared half into a zero-dependency @deepseek-ai/dsh-timeout library
(clampTimeout/deadline/timeoutOf/TimeoutReason) and leave the non-shareable
hard-kill in each capability, per the timeout-library RFC.
bash: run() owns the deadline; runBash drops its killTimer and no longer
classifies (SpawnSpec/SpawnOutcome lose timeoutMs/timedOut/aborted), so the
public timedOut/aborted booleans become mutually-exclusive first-abort
classifications. web_fetch: the hand-rolled controller/timer/listener/
signal.reason dance is replaced by provider-owned deadline/timeoutOf, keeping
the WEB_FETCH_TIMEOUT / WEB_ABORTED contract. fs stays timeout-free (README
states why).
Beyond the mechanical conflicts (provider capability lines vs master's new
inheritsParentContext field; generated catalogs regenerated rather than
hand-merged; knip/lockfile), three master-side reworks required semantic
adaptation of this branch:
- The persona rework removed AgentOptions.systemPrompt, which was the
structured-output instruction's channel. The instruction now rides the
SAME final-request enforcement listener that injects the schema'd tool:
appended per request to final.system (per-request wire state, not agent
prompt state). Tests assert the wire request (adapter.requests) instead
of child.options; the bare-direct-dispatch test pins the no-system arm.
- Tool guidance moved out of deployment prompts into per-tool prompt
sections; the examples' workflow paragraph became a tool:<toolName>
section contributed by dsh-tool-workflow (explicit-ask-only policy),
and both example personas resolve to master's minimal identity+behavior
form. tool-workflow gains inject: systemPrompt (+ peer dep, tsconfig
ref); the export-shape guard updated.
- The uniform-RFC-format gate: the dynamic-workflows RFC restructured to
the implemented/ skeleton (bare Status line; Proposal -> Decision;
What-was-rejected -> Alternatives considered; new Consequences), and
the overall-run-timeout deferral is now recorded in the RFC's Deferred
list. The doc-graphs atlas classification gains the workflows seam
(workflow-vm implementation, tool-workflow consumer).
Master's harness-identity section made "empty assembled prompt" states
unreachable through the loop, so the instruction-append is a plain
undefined-ternary and the structured tests assert append-not-replace.
All snapshot goldens (including workflow-run) replay unchanged. Full
local CI-equivalent gate sequence green on the merged tree.
Two review responses that belong together — the same review argued the
engine was defending the wrong threat while a benign-input bug wedged
the product.
1) Drop hostile-value containment; state the trust premise.
Scripts are model-written — the same trust level as the model's bash
access — yet successive pre-push review rounds had ratcheted in defenses
that only matter against an adversarial author: trap-free proxy
rejection, accessor-never-invoked descriptor walks, realm-side
pre-rendering of thrown values, realm-built promises/arrays/error clones
with structural fatal recognition. That same author keeps a documented,
accepted, unkillable event-loop spin, so containing its error VALUES is
cost without a threat model — and the planned hardened engine
(worker/isolated-vm) gets value isolation by serialization and deletes
all of this machinery anyway.
What stays, because benign scripts hit it constantly: result never
rejects; dropped hook promises cannot become unhandled rejections; the
value boundary rejects LOUD everything JSON cannot carry (now a plain
recursive walk — getters are read ordinarily and their result is what
crosses; a throwing read fails loud); a "__proto__" key still copies as
a data property; the fatal-vs-null combinator discipline (now host
instanceof — unforgeable from the realm and simpler than clone-shape
recognition). What changes for scripts (documented in the engine
README): hooks hand back host values and host errors — in-script
`instanceof Error` on a hook failure is false (branch on e.name/e.code)
— and args are host-cloned once so a script cannot mutate the caller's
object. realm.ts drops 289 → 173 lines; the hostile-value test tables go
with it. The premise now leads the engine module doc, the README, and
the RFC's engine section, with the removed machinery recorded under
What was rejected.
2) result settles within the dispose grace of a cancellation.
Review finding (verified through the real registry + tool + engine): a
script parked on a promise no hook owns — `await new Promise(() => {})`,
`await Promise.race([])`, a returned never-settling thenable — could not
be settled by cancel(): hooks reject and children abort, but nothing
touches a promise the engine does not own, so `result` stayed pending
FOREVER (the previous cut even pinned that as intended). The tool awaits
run.result BEFORE its disposing finally, the registry awaits the tool,
the loop awaits the registry — one such script wedged the whole agent
turn past any abort, unrecoverable in-process; the mock engine in the
tool's abort test settles result on cancel, which is exactly the
behavior the real engine lacked, so no existing test could see it.
The seam contract now says it out loud: once a run is cancelled, result
SETTLES within the implementation's bounded grace even if the script
never does. The vm engine arms an abandon channel in cancel(); drive()
races the script against it, force-settling 'cancelled' at the grace
(the abandoned settlement stays contained; a post-slice synchronous spin
remains the documented limitation). dispose()'s outer race now exists
for child quiescence only, and `workflow/end` again fires exactly once
per started run. The old 'result stays pending' pin is FLIPPED to the
new contract (the pinned behavior was the bug); new regressions cover
cancel-then-settle on a parked script, a never-settling returned
thenable, and the full composition through the REAL registry + tool +
vm engine (tool-workflow gains workflow-vm/subagent devDeps for it).
agentsStarted JSDoc clarified while touching the vocabulary (accepted
calls, including ones still queued at cancellation).
Review round 2 (tianyicui inline comments):
- dsh-system-prompt itself registers the harness:identity (-100) and
deployment:persona (0) sections — they must survive a swapped loop
plugin, so they leave dsh-agent-loop; the persona text is the plugin's
own validated 'persona' config. The model/cwd variables STAY on the
loop: runtime facts of the agents it drives.
- AgentOptions.systemPrompt is deleted with all its forwarding plumbing:
the app configs' systemPrompt keys become 'persona' routed through
dsh-agent-core (schema = z.intersect of the owners'), the ACP bridge
and tool-subagent stop carrying persona configuration, and subagent
children now render the deployment persona like every other agent.
- Example personas drop transport/interface trivia (ACP, CLI) — facts
irrelevant to the model.
- Root CONTEXT.md removed (not idiomatic); its persona definition was
wrong under the new ownership anyway.
- Docs, READMEs, the prompt-variables RFC, and generated catalogs
updated; new loop test pins the assemble-waterfall escape valve
(an emptied assembly sends NO system field).
Reconciliations beyond textual conflicts:
- product rename (DeepSeek Code -> DeepSeek Harness SDK) applied to the
PR-added assertion in system-prompt.spec.ts that master's rename
commit could not reach
- architecture.md: master's rewrite kept; this PR's prompt-assembly
semantics re-added in the new doc's voice (Turn Flow footnote +
service-spine row), within the 1630-word ceiling
- cordis catalog regenerated into master's split events.md/services.md
(events-and-services.md deleted); module graph and doc graphs
regenerated to pick up this PR's new events and dependency edges
A new capability family at packages/workflow/ in the bash seam shape,
modeled on Claude Code's dynamic workflows: the model writes a JavaScript
orchestration script (export const meta = {...} + plain-JS body), a runtime
executes it, and the script — not the conversation — holds the loop, the
branching, and the intermediate results.
- dsh-workflow (ctx.workflows): abstract WorkflowService + run vocabulary
(WorkflowRun whose result NEVER rejects) + observe-only workflow/* events
carrying data snapshots (id + meta, never the live run), per-listener
contained like subagent/*.
- dsh-workflow-vm: in-process node:vm engine. Meta extraction via a
string/comment-aware scanner (template interpolation rejected; literal
evaluated alone in an empty timed context; statement blanked line-
preservingly so stacks keep script line numbers). Hooks: agent(prompt,
{label, phase, schema, model}) over ctx.subagents, parallel(), pipeline()
(no cross-stage barrier), phase(), log(), args. Fatal-vs-null discipline:
hook misuse (unknown/deferred options, bad arguments, unsupported
schemas, tripped caps, seam start failures, cancellation) throws fatal
WorkflowErrors the combinators RE-THROW — never dissolved into the
per-item null reserved for child failures. Realm boundary: inbound values
materialized by descriptor walks that never invoke accessors (defineProperty
copies, __proto__-safe); outbound values rebuilt in-realm via the
context's own JSON.parse. Determinism bans (Date.now/Math.random/argless
new Date) kept so future resume support cannot break scripts. Caps and
timeouts are validated Config. Every hook promise carries a no-op
rejection consumer (app-boot exits on unhandled rejections).
- dsh-tool-workflow: the model-facing workflow tool, synchronous like
dsh-tool-subagent (start → await → try/finally dispose; abort bridged;
non-completed → isError). Generic render card titled by a textual
meta.name sniff. The tool description carries the authoring contract.
Wired into examples/{coding-agent,acp-agent} with explicit-ask-only
guidance. Coverage at every tier: unit (meta scanner, materializer incl.
counting-getter and __proto__ regressions, combinator semantics,
concurrency ceiling, caps, cancellation, no-unhandled-rejection abandon),
integration over the real spawn stack, with-key e2e (real two-phase run +
the tool through the registry pipeline), and a recorded ACP snapshot
scenario (workflow-run, 1 child session). RFC:
docs/rfc/implemented/feature/2026-07-05-dynamic-workflows.md (deferred
work explicitly listed). AGENTS.md budget 1575 → 1590 for the new group's
layout line.
