Review follow-up: the blanket 10s default correctly targets worker-thread
cold-start races (starting, first-script-line, async child-registration
messages — genuinely CPU-bound under CI contention), but four waits assert
something different — that the HOST reacted PROMPTLY to an event that
already happened (a settled result, an observed worker death). Those had no
cold-start left to wait on, so the generous default just widened the window
a real regression could hide in.
Verified by injecting a 6s delay into the settle-reap's abort call: the
un-overridden helper's test still passed in ~6s. The same mutation now fails
in ~1s with the explicit 1000ms override restored on all four sites (the
abort-on-settle test's own assertion, the two worker-death cancel/dispose
reap checks, and the dispose-ack-race check). The other 12 waits keep the
10s default — they run BEFORE a result is awaited, waiting on the worker to
actually start rather than on a host reaction. Doc comment corrected to
describe the split instead of claiming every wait is a cold-start race.
The spec's 16 vi.waitFor sites used the 1s default timeout to wait for
worker-thread startup and child registration — CPU-bound work that blows
past 1s on a contended runner. The CI coverage lane (4 vitest workers plus
suites that spawn real subprocesses) hit this 3 times across 4 recent PR
runs, each a different subset of the cancellation/worker-death tests, each
green on rerun.
Every wait now goes through a shared helper with a 10s bound, and the file
sets a 30s test timeout to make room for it. The one deliberately tight
wait keeps its 800ms bound through the helper's override — it proves the
host (not the wedged worker's later loop turn) delivered the cancel, so a
generous bound would erase what it tests. No behavior under test changed.
The empty-env hardening wiped the one variable the UNBUILT worker's
loader depends on: tsx finds a tsconfig by searching up from the
worker's cwd, and a parent running outside the repo (the ACP snapshot
harness pins the repo tsconfig through TSX_TSCONFIG_PATH exactly
because its child cwd is a temp dir) lost the dsh-* paths map - the
worker then resolved workspace imports to unbuilt lib/ bundles and died
on CI with ERR_MODULE_NOT_FOUND (green locally only because stale built
lib/ masked the wrong resolution).
Forward exactly that variable when the parent carries it - loader
plumbing, not a secret; the built shape stays fully empty - and pin the
whole contract with an escape-based test: the worker env is exactly
{TSX_TSCONFIG_PATH}, the credential canary still never crosses.
The documented vm escape reaches process, and the worker inherited the
harness's env - so a buggy or prompt-injected script could read and
exfiltrate ambient credentials (DEEPSEEK_API_KEY et al.) without
touching a single file (ds-review-bot finding on #233).
Spawn with env: {} and a hermetic execArgv on both runtime shapes, the
same stance as dsh-code-runtime-worker and stronger than the scrubbed
env the defensive-patterns rule requires for spawned commands (a shell
needs PATH; this worker needs nothing). Ambient-channel hardening only:
an escapee keeps the process-wide privileges the trust premise already
admits - the genuine sandbox remains an engine swap.
agent-end was worker-authored only, so a start already forwarded to
observers lost its paired end whenever the worker could no longer speak
- the grace force-settle terminating a wedged script, or an unexpected
worker death - stranding progress consumers with agents that never
finish (ds-review-bot finding on #233).
The host now keeps a ledger of forwarded starts and funnels every
agent-end through one gate: worker-reported ends pair (and clear) their
entry, and both termination paths drain the remainder as synthesized
'cancelled' ends BEFORE the run settles, so ends always precede
workflow/end. A real settlement racing the force-settle loses to the
synthesized cancellation - the same first-wins override onResult applies
to the run's own result.
Systematic trace through Zed (crates/agent_ui thread_view.rs +
crates/acp_thread): kind:execute routes a tool call onto the
terminal-card layout, whose header (render_collapsible_command) has NO
disclosure toggle, whose body content renders only when is_open — a
flag only a real terminal entity can ever set — and which suppresses
the Raw Input view outright. Every prior attempt (rawInput, pending
content, completed content) targeted slots that layout structurally
never renders; the one slot it always shows is the TITLE, which said
"Run code". codex-acp confirms the idiom: execute cards are titled
with the command itself.
presentCall now titles the card with the program (rawInput kept as the
canonical input slot); presentResult omits the title — an update
replaces only provided fields, so the program header persists — and
carries the captured output as content. Goldens re-recorded; the unit
test pins title-carries-program on both frames.
The sandbox execute wrapper JSON round-tripped the return and blindly cast it
to ToolExecuteReturn. A JSON-valid but wrong-shape return — a bare string,
{ content: 'ok' }, blocks without a type tag — sailed through: the registry
spreads result.content, so { content: 'ok' } became ['o','k'], passed the
session log's isJsonValue gate, and the DeepSeek serializer then flattened it
to '(no output)' — silent corruption of the next model request and every
replay, instead of a contained tool error.
The round-tripped value is now shape-checked against the two ToolExecuteReturn
forms (array of content blocks, or { content: blocks, meta? }); block checks
are structural only (plain object + string type tag) because the ContentBlock
union is merge-extensible. A wrong shape — and the formerly cryptic
forgot-return/bare-string cases — fails that one call with a teaching error
echoing a truncated preview of what was returned and the two valid forms.
New specs pin the object-form pass-through (meta included), six rejection
shapes, and the preview truncation; per-file 100% coverage holds.
The previous commit put the fenced program only on the pending card —
but an ACP tool_call_update's content REPLACES the card content (Zed
truncates to the new list, crates/acp_thread update_fields), so the
code vanished the moment the run completed and was effectively never
visible. presentResult now re-carries the fenced program before the
captured output via a shared fencedProgram helper; the completed card
body is program + output, rendered by Zed as syntax-highlighted
markdown behind the card disclosure. Goldens re-recorded (filtered
this time: DSH_SNAPSHOT=record vitest -u -t mode-turn); unit test pins
the two-block result content.
The generated program rode only rawInput — the detail/expanded slot
many ACP clients never open — so the code a run executed was invisible
in the UI stream. presentCall now also carries it as a fenced ts block
in the card's content, which the bridge already forwards as tool_call
content. The two code-mode snapshot goldens are re-recorded live and
replay green; the presentation unit test pins the fenced block.
P1 review finding: extractMeta timed only the literal's vm evaluation;
materializing the RESULT then read properties ordinarily on the HOST
stack, so a meta literal smuggling a getter (get name() { while(true){} })
could wedge the host outside any timeout — defeating the exact spin
isolation the worker thread exists for.
Rather than harden the evaluator (descriptor walks, AST validation),
delete the mechanism: the workflow's identity now reaches the seam as a
plain JSON field (WorkflowStartRequest.meta), carried by the tool as a
schema-validated `meta` object parameter the model fills directly. The
engine only shape-validates data (validateMeta, every violation named)
and pre-parses the body; the scanner, the vm evaluation, and the
host-side materialization are gone, and with them the hole. A body
still opening with a Claude Code-style `export const meta` statement
gets a pointed SCRIPT_PARSE message (the likeliest authoring slip; a
CC script's body stays drop-in, only its meta header moves into the
parameter). syncTimeoutMs now governs exactly one thing: the initial
synchronous slice inside the worker.
The RFC's decision section is rewritten in place (implemented-RFC
rule); the embedded-meta format moves to alternatives-considered with
the hole as the reason. Tool description, presentation (title now reads
meta.name directly — the textual sniff is gone), seam vocabulary docs,
and catalogs follow.
The regression's script spun immediately after calling agent(), but the
agent() continuation (which posts the child-start RPC) only runs on a
microtask tick — the spin seized the worker's loop before the post, so
the host never saw a child inside the waitFor window. A few await-null
yields before the spin let the RPC out; the posted message needs no
further worker-loop turns to reach the host, so the wedge still holds
for the Cancel message the test is about.
Pure mechanical rename now that the package's internals are the
worker-thread engine: directory, package name, spec/e2e filenames,
module tags and logger prefixes, tsconfig/knip/run-gates/AGENTS.md
references, example cordis.yml plugin ids, doc links; catalogs
regenerated and the lockfile refreshed.
The outer ring catches up with the engine swap (the package's own
README/JSDoc rode the port commit):
- Seam module doc and README name the worker-thread engine as THE
implementation, with isolated-vm/separate-process sandboxing as the
deferred hardening; the seam service doc states the holder-owned-runs
contract (engine-fiber disposal deliberately leaves live runs to
their holders).
- Seam contract precision: agentsStarted documents the termination-path
degradation to the host-observed count; the events section scopes the
agent-start/agent-end pair to calls that STARTED a child run;
WorkflowRun wording drops the vm-era abandonment language.
- The dynamic-workflows RFC is rewritten in place to the shipped
mechanism (implemented-RFC rule): why worker threads, the thread's
concrete buys, the in-process node:vm first cut recorded under
alternatives considered; the tool section describes the usage policy
as the tool's own prompt section.
- gen-doc-graphs: six workflow/* DYNAMIC_EVENT_DISPATCHERS entries (the
catalog no longer claims nothing dispatches them) and the seam-note
wording; core-data-structures gains its workflow.md index row;
packages/README + AGENTS.md layout line + example cordis.yml comments
say worker-thread; catalogs regenerated.
In-place port of dsh-workflow-vm from the in-process node:vm execution
to one worker thread per run (the workflow-workerthread engine of
PR #215, adopted as THE engine): the script's vm context moves inside
the worker, agent() bridges to ctx.subagents over the message port
(host.ts/protocol.ts/session.ts/worker.ts are new; runtime.ts loses the
abandon channel — the host's grace timer force-settles and TERMINATES
instead), start() pre-parses the body host-side to keep the seam's
synchronous SCRIPT_PARSE throw, and a ready→go handshake keeps a run
cancelled before start from ever executing the body. start() no longer
blocks the host, termination is real, and the value boundary is
serialization by construction. The package keeps its name until the
follow-up rename commit; scripts see the identical hook surface, and
the seam-contract tests hardened ahead of this swap pass unchanged.
The run and child-RPC surfaces are class-shaped rather than literal
bundles: WorkerRun IMPLEMENTS the seam's WorkflowRun (id/meta are its
own clone, separate from event payloads') and start() returns the
instance directly — interface parity with the seam is compiler-checked;
worker-side, ChildRpcBridge (implements ChildPort; callId allocation +
pending book-keeping settled by onChild* entry points) and
RpcChildHandle (every member an RPC keyed by its callId) carry names in
stacks. ChildPort's method is startAgent — it names what it starts,
matching the script-side agent() hook and the agentsStarted /
workflow/agent-* vocabulary; the Child* type names deliberately stay
(the worker side is cordis- and subagent-free; these are reduced JSON
projections, not the seam's types).
Review findings from the reference PR are folded in rather than
re-introduced:
- cancel() drives BOTH child-cancel channels host-side: the request
signal aborts AND each registered child's explicit cancel() is
called — a worker wedged in a synchronous spin cannot relay its own
ChildCancel RPCs (regression: cancel-only provider + wedged worker).
- All host warn paths render through the total renderThrown; a child
dispose() rejecting a value whose coercion throws still acks
ChildDisposed instead of wedging the script's finally (regression).
- built-worker.e2e.ts is wired into builtBinSmokeGate and the AGENTS.md
CI sequence — the built lib/worker.js resolution contract now runs in
an automated gate.
- workflow/end payload pinned on the worker-death path (with the
cancelled and grace-force-settle pins riding the ported spec).
- Real-Worker scripted timing budgets widened (50-300ms → 150-1000ms)
for starved CI hosts.
Workspace plumbing: the "./worker" subpath export sanctions the second
runtime bundle (check-workspace-constraints), tsdown builds two
single-entry passes, tsx becomes a devDependency for the unbuilt worker
spawn.
cordis provide() accepts any value and cross-mount composition advertises
ctx.provide('name', value), but the façade's readService unconditionally
proxied every declared service — new Proxy('42') throws "Cannot create proxy
with a non-object as target or handler", so a consumer of a primitive-valued
service crashed on first read with an error naming neither the service nor
the fix. A primitive or null value now passes through unwrapped (after the
denyContext check); only object- and function-valued services are proxied —
a primitive has no method that could hand back a Context, so nothing is lost.
New cross-mount spec pins both read paths (ctx.<name> and ctx.get) for a
number and a null provided value.
The sandbox docs overclaimed a containment contract the design never makes:
"capability access is routed through cordis services, never Node built-ins,
so everything a mounted plugin does stays inspectable and disposable". The
host-realm helpers on the sandbox global (harness, console, btoa) are
reachable functions, so mount code that goes looking can reach the host realm
through one of them — accepted under the trust stance, because the ctx a
mount ultimately receives is fully privileged anyway. Reword the sandbox
module doc, the README trust stance, and the RFC sandbox-semantics section to
say exactly that: the traps and small global surface STEER honest code onto
the cordis services; they are not a security boundary.
Three engine-agnostic pins, landed BEFORE the worker-thread port so the
port commit demonstrates contract preservation against them:
- tool-workflow: the tool:<toolName> prompt-section registration was
entirely unasserted — assemble() now pins the section present under
the CONFIGURED name and gone after fiber dispose (the packages
AGENTS.md dispose-and-assert-cleanup rule; tool-bash is the template).
- tool-workflow: drop the dead `??` re-defaulting of already-
schemastery-resolved config (the hidden-fallback shape AGENTS.md
bans) and the direct-apply test that existed only to cover those
branches; both engines' `config as ResolvedConfig` is the pattern.
- workflow-vm: workflow/end was asserted only on completed runs — the
cancelled path and the grace force-settle path now pin the event and
its stopReason/error/agentsStarted payload (an observer's only death
signal on those paths).
emitWorkflowEvent's catch rendered the thrown value with a bare
String(error), which itself throws when the value's toString /
Symbol.toPrimitive throws — breaking the documented containment
guarantee: such a listener could fail the run mid-emit, starve later
listeners, and turn the detached workflow/end settle hook into an
unhandled rejection. Render through a local total fallback instead
(String in a try, a fixed label when even coercion throws); local
because the seam sits below every engine and cannot import an engine's
renderer. Regression: a listener throwing a coercion-trap value — the
emit does not propagate and later listeners still run.
Two review findings (#220) on the sandbox context façade:
- Undeclared services were reachable: the façade resolved any live global via
ctx.get(name), so ctx.bash worked without inject: ['bash']. A cross-mount
consumer could then depend on a provider cordis never saw — unmounting the
provider would neither park the consumer nor unwind its registered tools,
leaving a model-visible tool that fails only at execution. The façade now
reads ctx.fiber.inject and refuses any service the mount did not declare
(with a teaching error naming the inject fix), so the dependency is always
visible to cordis and its activation/unload semantics bind.
- ctx.tools.get returned the live ToolDefinition, including execute — mount
code could call another tool directly and bypass ToolRegistry.execute and
its pre/post-execute hooks and accounting. get now returns the same
read-only name/description/parameters view as schemas(), never an invocable.
Adds inject-gate and schema-view regression cases to sandbox-context.spec.ts
(undeclared property/get denied, declared allowed, the cross-mount zombie-tool
scenario refused at call time, get exposes no execute). Package stays at
per-file 100% coverage. RFC, mount description, and tool-catalog updated.
Review finding (#220): the guarded proxy only special-cased ctx.tools, so
mount code could reach an UNGUARDED context through ctx.root, ctx.extend(), or
a service instance's .ctx, then ctx.root.tools.register({…}) to bypass the
marker check and host-realm normalization — a raw vm-realm result would later
error a real agent turn at the session-log plainness check.
The sandbox ctx is now a whitelist façade, not a pass-through proxy: it exposes
only what a mount needs — tools.register (marker-guarded), on/once, provide, the
timer helpers, and injected services resolved through a guarded get — and denies
every framework-plumbing member (root, parent, fiber, reflect, registry, extend,
isolate, intercept, plugin, set, mixin, …) with a teaching error. Injected
services are wrapped so a method returning a Context is rejected on the way back
(the .ctx escape), closing the one indirect leak. There is no context-valued
member left to reach; cross-mount provide/inject is untouched (the plugin's own
inject and the fiber's pending/active gating are unchanged). ctx.plugin (child
plugins) and ctx.set are denied by design; ctx.effect is deferred (FIXME).
Adds tests/sandbox-context.spec.ts covering the escape class (root/extend/fiber/
plugin/set/… denied, the classic root.tools.register bypass, the .ctx escape,
read-only writes) plus the async-service and symbol/in-operator paths for 100%
coverage. RFC/README/tool-catalog/config-catalog updated; api-catalog.ts
regenerated (also picks up the codeRuntime service that entered on the master
merge and was left stale).
The tree SHAPE was the wrong surface for the model: what it needs from
cordis_inspect is what services, plugins, and capabilities are loaded, not the
fiber hierarchy. The plugins section is now a flat name + lifecycle-state list
from ctx.registry (deterministically sorted, one line per instance); the ASCII
tree renderer, the parent→child rebuild, and the dyn-id tree annotation are
deleted — dynamic mounts keep their own richer dynamic section (id, state,
provides, waits). Net -49 lines; RFC and READMEs state the flat-list contract.
Field sessions showed models writing tool schemas in the JSON-Schema dialect
by strong prior — type: 'integer', required: false, then the full
{ type:'object', properties, required: [...] } wrapper — and the rejection
text itself pushed a nearly-correct DSL attempt BACK to raw JSON Schema: one
stats tool cost three consecutive schema errors before mounting. The boundary
now normalizes wherever the input has exactly one meaning (wrapper unwrapped
with the required array becoming per-property flags at any nesting level,
integer → number, required: false → optional, all rebuilt as fresh host-realm
objects) and rejects only genuinely meaningless input, enumerating the valid
vocabulary in the error. Re-running the failing session mounts first-try.
The mount description documents both accepted forms.
The sandbox deliberately provides no Node API, and now says so instead of
letting a bare ReferenceError teach nothing: require, the timers, and fetch
are callable traps whose error redirects to the cordis alternative (inject:
['fs'] + ctx.fs, ['web'] + ctx.web, ['bash'] + ctx.bash, ['timer'] +
ctx.setTimeout — a fiber effect, unwound on unmount). Only function-shaped
globals are trapped; process/Buffer stay undefined so typeof feature probes
stay inert. The mount description and the demo persona state the routing rule,
and the demo mounts ctx.fs (local provider) and ctx.web (seam + keyless local
fetch provider) so agent-built plugins have real capabilities to build on.
Live-validated: a model that reached for Node setTimeout self-corrected to
inject: ['timer'] in one step and built a working ctx.web fetch tool.
Everything outside the package and example that a new top-level group and a
new demo touch: GROUP_ORDER in gen-module-graph and gen-doc-graphs (plus the
tools-service consumers list, the APP_EXAMPLES entry, and the graph-atlas
label/mode rows), the knip e2e entries, the packages/README group row, the
AGENTS.md layout and demo lines, and the regenerated module-graph /
config-catalog / graph-atlas / capability-seams / composition artifacts.
AGENTS.md and examples/AGENTS.md word-budget ceilings rise to current+5%
(1802 / 653): the new group and demo rows are genuine additions to both docs,
not condensable restatements.
New top-level packages/cordis/ group with the self-referential toolset:
cordis_inspect (services / plugin tree / tools / dynamic mounts / api / events,
the api section intersecting the generated catalog with the live service store),
cordis_mount (model-written code evaluated in a node:vm sandbox, mounted under
one cordis-dynamic group fiber as dyn-<n>), cordis_unmount (awaited disposal to
quiescence). Boundary mechanisms: dual-realm instanceof, JSON realm
normalization of dynamic tool results, marker-guarded registration, SchemaSpec
teaching errors, parse failures surfaced with the offending line + caret and a
line-scoped TypeScript hint, and the unmount-first recipe on tool-name
collisions. Config: vmTimeoutMs (schemastery, default 5000). Design record:
docs/rfc/implemented/feature/2026-07-08-self-referential-cordis-toolset.md.
The tool-catalog boot manifest, its regenerated output, and the pinned
tool-name list land here rather than with the other repo registration: the
completeness guard globs packages/*/tool-* and fails the generator (and the
core/tools spec) the moment the package directory exists.
The extracted library's name sat one edit away from @deepseek-ai/dsh-subagent-inprocess
(process/inprocess), inviting a typo'd import to silently resolve to the wrong
package. subagent-subprocess also reads as the deliberate counterpart to
subagent-inprocess (in-process vs. subprocess), matching how the two shared
drivers actually differ.
Package directory, npm name, module doc, JSDoc module tags, test-file name and
its temp-dir prefixes, the subagent-acp import and its Config/tsconfig/package.json
references, root tsconfig.json/tsconfig.build.json/knip.json entries, and the
packages/subagent group README all renamed together; regenerated
docs/module-graph.md and docs/config-catalog.md. Pure rename — no behavior,
export, or Config shape changed.
Move compact-basic's private _extractText/_blocksToText into the interface
package as renderTranscript/renderContentBlocks, so the summarize path and a
future recall read path render one span identically. Byte-identical output
vs the private helpers it replaces; compact-basic delegates.
