Proactive pressure scaled `thresholdRatio` by the routed model's entire
context window, but a request also reserves output tokens the provider
charges to that same window — the LLM service deliberately never clamps
them (`packages/llm/llm/src/index.ts`).
With a 1,000,000-token window and the default 256,000-token request cap the
gate therefore sat at 800,000 while the provider refused any prompt above
792,576. Observed in a real session: three requests died at 792,850 /
797,639 / 793,352 message tokens ("maximum context length is 1048576 ...
you requested 1048850 tokens"), no pressure compaction ever fired, and the
overflow recovery that did run could not condense an ~800k-token history
into the 8192-token summary cap.
Scale both ratios by the message budget left after the reserve, sourced from
the effective request envelope's `maxTokens` and falling back to the
adapter's per-request output default. A reserve that leaves no message
budget reports a target-specific pressure configuration error, matching the
existing per-route warning policy rather than silently disabling proactive
compaction. The headless compaction e2e advertises a coherent capacity/cap
pair so its 8000-token message budget and 4000-token threshold are unchanged.
Related #1841
The new master source-tool smoke reads isError and content from a V3
nested tool-result block. After integration with the V4 release, that
block no longer exists: the durable message itself has role tool,
toolCallId, isError and result content. The old assertions fail to
compile and cannot verify the current writer's output.
Read the existing V4 fields while retaining the real source-profile
launch, single Tools instance, call identity, sandbox error and shell
output checks. Do not impose the old single-wrapper cardinality on the
flat content array. This adapts the incoming regression test without
changing the V4 format, migration implementation or stored recordings.
Integrate the master tip observed for the final V4 landing preparation on
top of the already merged migration stack. Preserve the V4 writer, flat
message migration, extension policy, catalog validation and historical
recordings while bringing in upstream profile resolution and desktop/UI
changes.
Resolve four generated translation-pairing records from the combined
English/Chinese text. Keep this master checkpoint separate from any
integration fixes and retain all reviewed parent history. Follow-up
validation covers the affected launch/UI paths and migration replay.
Windows tmpdir spells the user directory with an 8.3 short name. Generation anchors and Node's own resolution report the long name, so the resolver, generation, and real-launch fixtures now create their roots through realpathSync.native and compute expected paths the same way.
The src/lib x installed/npm-link fixture adds a stale @deepseek-ai/dsh-tools link in profiles/node_modules, a plugin dependency absent from the generation with nested external leaves, and byte-for-byte preservation of profile files and link targets.
Give durable user context its producer-owned source kind instead of a generic plugin wrapper. The V3-to-V4 conversion preserves the frozen rename and collision rules, all own JSON metadata, event coordinates, and historical request-only kinds. Native catalog and JSONL admission retain source validation so unknown attribution can survive without its producer while malformed owned sources are refused.
Use an explicit user-only RequestUserInput for auxiliary LLM prompts that do not own durable message identity. Auto Review and compaction remove their request-only source registrations while retaining prompt content and freezing; provider serializers and projections keep the same wire data, assistant replay isolation, image ordering, and tool correlation. Session writes, Agent delivery, and recorded title requests continue to require identified durable messages.
Bind the saved compatibility policy only to user sources and qualify tmux location attribution without weakening fixed system, model, or tool semantics. Rebuild the current inventory and unpublished V4 acknowledgement on the actual tool-role parent, preserving its rationale and all accepted historical records. The catalog format version records tooling metadata; Session remains V4. Developer-role policy and activation belong to their own feature.
The latest-master Messages adapter, serialization, request-input typing, EpochHeader reservation, and attribution checks pass 103 tests across five files. The retained V3 Messages input-history snapshot passes through the source profile; Host/Client typechecks and all 41 documentation gates pass. The generated comparison records four changed roots and 447 unchanged type fingerprints. Latest-master ARIA changes are retained, with producer source labels applied independently; new Messages fixtures retain their producer names as declared source kinds and use the fixed system-message helper; recorded Session generations are unchanged by this regrouping.
Co-authored-by: Tianyi Cui <[email protected]>
Represent tool results as tool-role messages and migrate released wrappers
without losing outer own JSON fields, call identity, error state, or writer
field order. Refuse nested or wrapper-only information that has no lossless
current representation before successor publication.
Validate current event relationships directly instead of projecting V4 data
back into V3. Keep lifecycle, PTC, title/command, compaction, system-field and
retired-syntax checks on catalog restoration and ordinary JSONL opens. The
Worker JSON equality helper must inspect own keys so it cannot admit PTC
arguments that Node rejects.
Keep the source role change, required native admission and consumer fixture
updates in one complete review layer. Explain shared-type changes in the
persistence record and the native reader rationale in its existing note.
Use checked branded fixture identities and discriminant narrowing instead
of unknown double casts; fixture values and assertions remain intact.
This recomposes the original tool-role work by _Kerman with the independently
validated migration and admission repairs by Tianyi Cui. Frozen generations
and snapshot input version selection are retained.
Co-authored-by: Tianyi Cui <[email protected]>
Master's agent-target update reports non-running children as inactive. The
release-only descriptorless-child regression still expected ready, so CI
failed before comparing the persisted result. Update only that status in
the assertion and its expected output; direct catalog discovery and the
absence of corrupt-child diagnostics remain asserted.
The expected log retains version 3 and every other byte. A built assembled
headless run passes in normal comparison mode, preserving the older input
format and its upgrade/reconstruction coverage rather than refreshing the
snapshot generation.
The expected-output CI lane sets DSH_EXAMPLE_MODE=lib. Selecting tsx/esm
alone did not exercise the source bin there, and the compatibility test
only checked --help, so removing source profile link resolution could
silently restore duplicate source/lib Tools instances in a green CI run.
Run the existing real headless-profile tool round trip explicitly in both
src and lib modes. The explicit mode overrides the CI environment default;
the source case uses the supported tsx/esm entry. Both cases retain the
real profile, shell tool, persisted Session assertions, stdout, and stderr
expectations, with isolated temporary directories. No recorded generation
or output normalization changes are needed.
After a full build, both cases pass with DSH_EXAMPLE_MODE=lib. Negative
control: removing only the source bin link-resolution selection makes the
source case fail at scheduler.prepare while the built case still passes.
Restoring the implementation makes both cases pass again. Scoped lint
passes. This puts the regression in the existing post-build expected-output
CI lane without adding artifact requirements to compatibility-only jobs.
Address the review suggestion that a caller-provided `cwd` left
`tempDirPrefix`/`tempDirParent` as dead but still-required inputs. The options
are now a union of "expand a prefix into a fresh temporary cwd" and "reuse this
existing directory", so the type system rejects passing both; the two callers
that reuse a directory drop their unused prefix.
Address the ds-review-bot findings on 472554a29e:
- resolveAgent refuses an Agent already live under the requested id: its owner
may still drive it, and `whenIdle` is not a single-message signal, so the
one-shot runner cannot own an exclusive run interval over it. The adoptability
rules still run first so the refusal stays specific.
- Refresh the README intro, identity section, and limitations, and the Agent
Note live-identity rule and change scope, for adopt-only.
- Clarify the loader-smoke `cwd` JSDoc (the prefix/parent inputs are ignored)
and pin the resumed wake's final text in the two-wake e2e.
Address the ds-review-bot findings on 9aea4b0804:
- Drop the stale "freshly created" wording from the resolveAgent `@returns`
and rewrite its leading comment for the two remaining identity paths.
- Rewrite the README run-flow paragraph (EN and ZH), the intro sentence, and
the cordis.patch.yml header comment for adopt-only.
- Add a real-composition two-wake e2e: capture the generated id from the first
`--json` wake, resume it by `--session-id`, and assert one persisted log holds
both tasks. `runLoaderSmoke` gains an optional caller-owned `cwd` so both
wakes share one world; the harness leaves that directory in place.
`--session-id` now adopts only: an id with no stored Session fails before the
task runs instead of creating an empty history the caller believes it is
resuming. A first round omits the flag, and the runtime reports the minted
`session-<uuid>` identity in the opening `session` event, so a supervisor
persists it and names it on every later wake.
Updates the help text, the README pair, the Agent Note pair, the generated
config catalog, the unit suite, and the product-profile e2e.
Move exact user-patch watching into app boot and audit activation explicitly.
Wait for CLI fallback HMR and preset subtrees. Join the first fiber-disposal
result in the directory chooser and browser package runner. Keep async
update errors observable without restoring generic Loader rollback.
Restore the current workspace-link and Schemastery manifests and their gate.
Preserve unconditional patch-input cloning, awaited Include initialization,
and durable teardown writes. Update diagnostics, snapshots, bilingual docs,
and the exhaustive vendor divergence ledger.
Validation: focused Loader, watcher, preset, inventory, and cleanup tests;
100% watcher coverage; 32 Web preset cases; real CLI webhook model session;
SDK smokes; 13 headless expected cases; 3 recorded-session replays; build,
typecheck, lint contracts, and all 34 doc-sync gates.
A full Node 24/Python 3.12 Vitest run had 21806 passes and two failures:
a baseline-reproduced experimental CSS import failure and a Python probe
timeout that passed alone. The default Node 26/Python 3.9 run additionally
exposed unsupported Python and FileHandle-GC host failures.
Tianyi Cui