CI caught two gaps in the review-fix commit: the two post-next() flush
listeners were literal clones (jscpd), and the disposed-skip branch inside
them had no covering test. Extract one shared flushAfter wrapper and pin the
skip with a captured-continuation disposal test (a downstream listener
disposes the fiber mid-waterfall; the resumed wrapper appends nothing).
Four ds-review-bot findings:
- examples/tui-agent composed dsh-mode without the now-required
modes.plan.section, so the TUI leaf failed at Loader startup (the keyless
smoke only asserts the banner and missed it); graft the same deployment
plan instructions the ACP leaf carries.
- The prompt-submit and turn-continuation flushes ran before next(), so a
session/set_mode arriving while a downstream async listener (the shipped
hooks listeners' shape) awaited applied one request late; both listeners
now prepend and flush after next(), matching the request-error wrapper,
with a regression test pinning the ordering.
- An HMR unload during the exit_plan_mode review let a later approval write
into the disposed service and claim an exit whose flush could never land;
the execute path now checks the fiber lifetime after the await and fails
the call (the mode stays plan; the model re-presents).
- resolveConfig accepted empty/untrimmed mode names that list()/ACP then
advertised while the package invariant rejected their selection,
desynchronizing the picker; names are validated non-empty and trimmed at
load, the same shape the invariant enforces.
Master arrivals re-grafted onto the plan-mode surfaces:
- Package invariant runtime contracts: every package owns ./invariant —
added the dsh-mode companion (mode/set payload validation: non-empty
trimmed name), its unit spec, exports/peer wiring, and the mode group in
the tsconfig invariant paths glob.
- Interception waterfalls gained an AbortSignal parameter
(agent/prompt-submit, agent/turn-continuation) and ToolExecutionInput.signal
became required: listeners and test dispatch updated; the exit tool now
forwards exec.signal unconditionally.
- TUI question dialog: master's position/unanswered header layout kept,
the plan-review detail block re-grafted between question and answers.
- LSP capability family arrived: lsp tool joins the catalog pin, dsh-lsp
examples deps beside dsh-mode; lsp-definition snapshot refreshed with
the mode command and modes advertisement.
- docs/architecture.md and packages/README.md regenerated tables re-grafted
(ctx.modes row, mode package row) and condensed within word ceilings.
- examples/acp-agent cordis.yml: kept the deployment-owned plan instructions
(incl. the conversational-agreement sentence) beside master's routed-model
compaction comment; snapshot expected outputs refreshed keyless and noise
fixtures restored.
Master landed the tool-subagent depth budget while this branch was in
flight: the numeric default maxDepth now fails the mount against the
ACP provider's depthLimit: false, so the composition fixture must state
maxDepth: 'provider-managed' — the documented opt-out for a provider
whose recursion budget lives in the child harness. The fixture also
moves off the retired stdio-demo REPL onto the current app-boot driver
pattern (runLoaderSmoke + one-shot cli), and the split isDirectory
statements gain the file-not-a-directory case the single-expression
form used to cover implicitly.
The inheritance branch ran only under hand-mounted plugin tests, and
the with-key e2e always configures an explicit cwd. Add a test-only
cordis.yml (scripted delegating model + the scripted mock ACP child)
booting the stdio app through the real Loader with cwd omitted: the
child proves it ran in — and was announced — the parent session's
workspace, keylessly.
SubagentStartRequest.parent, its core-data type-equiv mirror, and the
ACP backend Agent Note still said the out-of-process backend ignores
request.parent — no longer true: with cwd unconfigured it reads exactly
the session header's cwd, and a parent session without one now rejects
at start. State the one-field read at each owning contract, record the
workspace-resolution decision in the Agent Note, and align the README's
accessibility wording with the search-permission guard.
statSync().isDirectory() is true for a mode-600 directory, but a
subprocess cwd needs X_OK — spawn would fail EACCES, deferring a
self-contained configuration error past the load/start boundary the
resolution owns. Probe accessSync(X_OK) alongside the type check so
"not an accessible directory" is literally what the guard verifies.
The consolidated resolver scopes the anonymous id to $DSH_HOME rather
than the machine. Update the module contract, README, and Agent Note to
say per-harness-home explicitly instead of over-claiming a machine-global
identity, and record why DSH_HOME scoping is the intended single-root
meaning rather than a regression.
Review fixes for #462:
- resolveDshHome now treats an empty or whitespace-only $DSH_HOME as
unset, so a blank override never resolves the home to cwd via
resolve(''). Restores the guard telemetry's old resolver carried.
- The default-env telemetry test asserts only that globalConfigDir()
returns an absolute path, so a machine DSH_HOME without a .dsh suffix
cannot break it.
Automatic compaction fires mid-conversation, right after the loop warmed
the provider's KV cache with the last routed request. The default
summarizer then issued a separate request whose prefix shared nothing
with that warm request — a bespoke summarizer system prompt followed by
the older history flattened to one rendered transcript string — so a
differing first token invalidated the entire cached prefix and every
compaction re-processed the whole replayed history twice.
Move the compaction directive from the FRONT (a fresh system prompt) to
the END (a trailing user message), and replay the last routed request's
own system prompt, tools, message prefix, and shadowed-region messages
verbatim via session.requestHeader() + deriveEventMessage. The auxiliary
call is now a genuine prefix-extension of the warm request, so the
provider reuses the cached tokens up to the trailing instruction.
SummarizationInput carries the replayed prefix instead of a flat string;
the now-unused renderTranscript/renderContentBlocks path is removed with
its spec. Cache reuse is best-effort (head compaction guarantees a hit;
a mid-range compaction or a differently-routed summarizer forgoes it),
correctness is not.
Delete @deepseek-ai/dsh-home and make dsh-paths the sole owner of the
single-root harness home ($DSH_HOME || ~/.dsh). Migrate tool-bash,
skill-local, and agent-spine-demo off dsh-home, and fold telemetry's
divergent globalConfigDir onto the shared resolver, dropping its second
XDG/APPDATA policy and the deepseek-harness namespace so the anonymous
id lives under the harness home. Add dshHomeDisplay() for symbolic
user-facing paths, replacing workspace-context's bespoke check.
The test 'labels the default DSH home as ~/.dsh when HOME points at the
configured default' mocks node:os to return a temporary home directory
and expects discoverBaselineInstructionFiles to fall back to the default
~/.dsh path. However, when DSH_HOME is externally set (e.g. in the DSH
agent environment), resolveDshHome(undefined) reads that external value
instead of falling back to defaultDshHome(), causing the test to look for
AGENTS.md in the wrong directory and return an empty result.
Stub DSH_HOME to undefined so the test controls the full resolution path.
