Remove the host-user continuation capability and the public residency query,
then separate the seam's public event payloads from its internal lifecycle
control interfaces.
`followup()` now takes the exact live direct parent `Agent` instead of a
`SubagentAuthority` union. No production adapter ever supplied user authority,
so the `UserAuthorityGrant` brand token existed only to stop a forged
discriminant from bypassing the direct-parent check — deleting the branch
retires the token, its mint method, and that attack surface together.
Narrowing `parent` from `Agent | undefined` to `Agent` removes three special
cases, including the path where a parentless epoch dispatched its lifecycle
events unscoped. Scoped-versus-global dispatch is now decided by the event, not
by whether a caller happened to have a parent.
`activationState()` had no caller; `ActivationState`, `ActivationObserver`, and
`ContinuationHost` are package-private.
New `src/lifecycle.ts` owns the contained emitter, the one-shot run observer,
and the Activation observer, while `SubagentRunInfo`/`SubagentRunEndInfo` move
to `src/types.ts` beside the other consumer-facing contracts. Those payloads are
public API — dsh-jsonrpc, hooks-claude, and the package invariant all consume
them — whereas the observer is a contract between two in-package collaborators,
so they no longer share a home merely for both being lifecycle-shaped. The
service keeps ownership of the scope carrier: `scopeTarget()` composes the
service's own context filter, so a narrowed stand-in would silently change
scope filtering.
Also drops now-unused dsh-tasks-local and dsh-tool-tasks dev dependencies, and
corrects the README claim that a pre-residency failure emits a terminal edge —
that path only ever rethrew.
- Make host-user authority unforgeable. `{ kind: 'user' }` was a bare
discriminant, so any plugin holding `ctx.subagents` — including
model-generated cordis_mount code, which the advanced ACP composition ships
alongside continuable subagents — could construct it and skip the
direct-parent check for any known child id. It now carries an opaque grant
that only SubagentService.userAuthority() mints, which composition hands to
trusted host adapters; a model-facing tool uses parent authority from its own
execution context.
- Reconcile a delivery discarded inside its own admission window. An enqueue
listener that cancels fires the discard before followup() returns, so the
discard listener could not clear an id it had not seen; submit() retained it
and residency stayed `running` until an explicit drain.
- Recheck the caller signal after materialization. An abort landing between
publication and inbox acceptance still submitted the prompt and returned both
ids; it now rolls the child back.
- Stop promising the model transcript access that no shipped continuable config
mounts. The tools now state only that a background child does not report back.
- Restate the implemented note as shipped state rather than a proposal, so it
works as current authority.
Round 1 traded one teardown ordering problem for another. The observer now
splits capture from emission, which satisfies both consumers at once:
- Terminal facts are captured while the child is still registered, so consumers
that resolve it for the child's log and scope still work.
- The edge is emitted only after handle disposal settles, so a rejecting scoped
cleanup is reported as a failed epoch instead of a successful one.
Also:
- Keep the Activation in the map until disposal settles. Removing it first let a
racing followup() see no Activation and cold-resume into the still-registered
agent, and let a concurrent forest drain skip a still-disposing child and
release its parent first.
- Derive terminal telemetry from this epoch's event suffix rather than the whole
session, so a cold resume whose prompt is blocked no longer reports the
previous epoch's answer and turn reason.
- Cancel the ACP bridge's own prompts before awaiting the descendant drain: a
drain can block on persistence, and the top-level agents must not keep running
model and tool work for its whole duration.
All five findings were real:
- The terminal lifecycle edge derived its stop reason from teardown success, so a
child that errored, hit its token ceiling, or was cancelled reported as
completed once its checkpoint and disposal succeeded. It now reads the child's
own last message turn/end, which is authoritative.
- Live delivery never rechecked the caller signal after authorization yielded, so
an abort that won before acceptance still enqueued the message and returned an
id. Admission now re-checks at the boundary that owns the decision.
- Drain flushed before cancelling, letting a running turn keep appending events
the checkpoint could not cover and letting model work continue through a slow
flush. It now cancels to quiescence first.
- subagent/end fired after AgentHandle.dispose() unregistered the child, so the
hooks bridge could not resolve it for the child's cwd and scope. The edge now
publishes while the child is still registered.
- activationState() read Agent.status alone, which stays idle between an accepted
waking send and the microtask that admits it, so a synchronous inbox observer
could see settled with a queued turn. Residency now also counts messages this
manager admitted but has not seen leave the inbox.
Pins that the bridge releases the Activation forest before its own sessions, and
that a failed drain is reported without stranding that teardown. Reads the one
teardown method structurally so the bridge keeps no dependency on the subagent
seam.
Every scenario compares its live tool schemas and system prompt against the
shared header pins, so the Task-free subagent and send_message descriptions
change all 14 pin sidecars. The diff is only that wording plus the tools'
output-type shapes.
The authored transcript drove task_output, which no longer exists for a
continuable child and is not registered in this config, so the scenario hung.
It now demonstrates the RFC criteria directly: a delegation returning only the
durable subagent id, two send_message follow-ups queueing as later FIFO turns on
one inbox, an unknown id failing without delivery, and child-first disposal
despite a failed final durability checkpoint.
The snapshot-only overlay remaps the transcript's placeholder child id onto the
randomly minted live child, since the scripted model cannot know that id.
Also drops probe logging accidentally committed in cf0138258.
A continuable Activation outlives the turn that started it and owns descendant
teardown, so the bridge must drain that forest child-first before releasing the
top-level agents whose runtime the descendants depend on.
Also rewrites the authored continuable snapshot transcript for the Task-free
tool surface; the scenario's keyless replay is still under diagnosis.
Adds coverage for the fork-seeded descriptor turn numbering, omitted and declared
descriptor composition fields, a routeless cold resume, and the drain no-op.
Splits materialize's create-versus-resume inputs so the impossible
create-without-meta case disappears, drops the observer's unreachable
pre-residency guard, and annotates the three remaining paths that only a
non-deterministic send-versus-dispose race can reach.
Restores the one-shot settleRun coverage in its own file beside the helper,
covers fork's seed contribution, the post-transfer rollback, the descriptor
model route on cold resume, manager-unload drain, and a failing teardown branch.
Removes three redundant checks the surrounding contracts already own: the
duplicate-Activation and live-id pre-checks (AgentRegistry.enter is the
authoritative collision boundary) and a rollback lifecycle edge that could never
publish because the epoch had no start edge.
Deletes the in-process durability, resume, and steering tests whose premises the
seam no longer has, keeping a one-shot analogue for agent-option resolution, and
covers both a false and a rejecting final checkpoint in the manager spec.
Rewrites the service API table, authority-versus-provenance contract, residency
routing, and deferred-work list; scopes the in-process driver README to one-shot
runs; and restates both model-facing tools' outputs, which no longer carry a
task id.
Adds the new continuable types to the cordis-catalog type-link map and
regenerates the cordis api/service/event catalogs, tool catalog, config
catalog, and doc graphs.
Moves the RFC to implemented/, restates it as current-state prose under the
implemented note format, and records what the Task-backed continuable-subagents
note and the two subagent-service simplification notes retain versus what this
record replaces.
The continuable path has no Task, so send_message reports a queued next turn
and continuable delegation returns only the durable child id. Pins that a
follow-up queues behind an open turn rather than steering it, and that a
non-parent caller is rejected.
Also makes startContinuable/followup reject rather than throw synchronously
when continuation services are absent, so callers have one failure mode.
Covers the inbox-acceptance return boundary, pre-acceptance rollback,
caller-signal ownership on both sides of acceptance, residency-only routing,
single-inbox FIFO ordering across parent and user origins, waiting wakeup with
a retained handle, cold resume without the historical parent, ownership
registration and release, child-first disposal, send-versus-dispose races,
durability failure without an ownership leak, per-epoch lifecycle pairing, and
the absence of cancellation, steering, and report surfaces.
Replace the Task-backed continuation manager with one durable Session plus at
most one process-local Activation — a residency epoch for a reconstructed child
Agent, not a request, result, cancellation, or Task boundary. The manager owns
activation admission, authority, the live ownership graph, cold resume, and
child-first disposal; the Agent inbox is the only turn FIFO.
- startContinuable() is async and returns { childId, messageId } at inbox
acceptance; followup() takes a SubagentAuthority and returns AgentMessageId.
- SubagentProvider.resume?(), SubagentProviderResumeRequest, SubagentRun.steer?(),
SubagentProviderStartRequest and SubagentContinuation are deleted;
prepareContinuable?() is the continuable-creation capability.
- Cold resume calls ctx.agents.resume() from the manager through a private
activation-owner scope, never dispatching through a provider.
- Extract shared child composition, descriptor seeding, depth accounting, and
one-shot run settlement so the manager and one-shot driver keep one home
per fact.
Tests and docs follow in subsequent commits.
- Strict steer additionally requires an OPEN STEP: between steps the
loop may be awaiting its continuation/turn-stop checkpoints, where
pending steering was already folded and a terminal stop discards a
later arrival. A message accepted during an open step is drained and
recorded at that step's settlement before any terminal decision, so
the acknowledged-then-discarded window is closed. New keyless test
holds agent/turn-stop open and pins the rejection.
- tool-subagent-control README: distinguish synchronous not-delivered
errors from started-Task failures (unknown/foreign/descriptor-less
ids settle the started Task as failed), and drop the claim that the
completion notice carries the child's response.
- Wire the control service and send_message tool into every shipped
composition with a resumable provider and background enabled
(headless-agent, tui-agent, and the SDK helper's subagent feature
base resources); jsonrpc-agent disables background and is unchanged.
- Resolve the send_message availability check in the CALLER's tool
scope so a restriction that removes the follow-up tool from one
agent also blocks that agent's continuable start.
- Control-service disposal now cancels live activations and awaits
producer settlement instead of stranding them: TaskService keeps
producer Tasks across a reload, so the disposing service aborts each
activation-owned controller, resolves its terminal gate (the
effect-scoped onTaskDone listener is already gone), and awaits done.
A new test kills a mid-start activation through HMR disposal.
- Strict steer now rejects the two windows where an acknowledged
message would be silently dropped: the closed-turn durability-flush
window (status still running, loop strands drained steering) and a
committed structured capture (terminal turn-stop discards late
steering). Seam JSDoc, catalog doc, README, and the Agent Note
bilingual pair state the tightened contract; new keyless tests pin
both rejections.
- Continuable background delegation now fails loud when the advertised
send_message tool is not registered, instead of starting a durable
child the model cannot continue. The acp-agent example already loads
the control tool; the tool-catalog boot recipe is unaffected because
capability wording is harvested at mount.
macOS realpaths temp dirs into /private. The normalizer only stripped
the /private prefix for TMPDIR under /var; a TMPDIR under /tmp (any
explicitly relocated temp root) left one side canonicalized and the
comparison failing. Accept both shapes.
Dudu-0223
imccyu
Tianyi Cui
Dudu