core.md read as a type grab-bag: LLM wire vocabulary up front, the agent/loop story buried, and no correspondence to packages/core. It now opens on the packages/core control spine — the package-by-package loop map with a Page column into session/system-prompt/tools/scope — and keeps only what the spine group declares plus the repo-wide patterns: the Agent handle with its delivery/cancellation/interception contracts, the SessionEvent envelope, branded ids, the …Map pattern. The conversation vocabulary (Message/ContentBlock, the model request, adapters — 17 type-equiv blocks) moves to llm-streaming.md, which now declares packages/llm end-to-end; the duplicate ContentBlockMap paste near its seam section folds into the moved section, and the manifest, LINK_MAP, README table rows, website label (Core data structures → Core), and inbound anchors follow.
Every packages/<group>/README pair is now a thin front door in one shape: a why-first intro (bash's seam-pattern-first paragraph rewritten as 'shell execution for the agent'), the package table, and a closing pointer to the owning docs/subsystems page — the bash-style table stays the load-bearing middle. Load-bearing trailing paragraphs relocate rather than vanish: the fs no-timeout rationale becomes a filesystem.md section (both languages), session's four sectioned tables merge into one 12-row table, examples' legacy-bin H2 collapses to a pointer at jsonrpc-demo's README, and design rationale that already lives in an Agent Note or subsystem page is now linked instead of restated. All 40 pair records re-recorded.
The folder is becoming the home of one-doc-per-subsystem pages (intro +
data structures + cordis services/events), so the name must describe the
whole contract, not just the type-vocabulary third of it. Mechanical
rename rebuilt on current master: every inbound Markdown link, generator
constant, website route, type-equiv manifest path, and spec expectation
moves together; the zh sides of the notes whose prose names the folder
are aligned (子系统) in the same change; touched bilingual pairs
re-recorded; translation-prompt snapshot re-recorded (its example embeds
development.md). Historical Agent Note slugs keep their dated
filenames.
Durable record: bound to the owning session id and validated at the fold (orphan-SID shape, temp path inside the host temp root) — a fork's copied parent record no longer provisions the child, and a tampered record fails loud. Private temp dir: random unguessable name persisted in the record, created exclusively (pre-existing entries and reparse points fail EEXIST). Persistence: a fresh provision kicks an immediate flush (no write-behind debounce), narrowing the crash window to the flush latency — documented as the one self-healing gap. Runner-failure rules: exit-gated on 127 so a confined command that prints the signature on a non-127 exit is never misclassified. Spawn: AssignProcessToJobObject failure terminates the suspended child (no hanging orphans). SandboxExecutionPolicy.sessionId is the branded SessionId. Boundary docs: qualifying clause on the absolutist sentences, NULL-DACL Known Limitation, 'full' scoped to the supported NTFS surface, CLM gate comment.
workspace-write now runs [logon SID, Everyone, orphan]: the two lists differ only by the orphan, and the keep-alive group (logon SID + Everyone) is the single shared invariant. The WMI namespace security check fails in BOTH modes (0x80041003), so CIM/Get-ComputerInfo are unavailable everywhere — the price of closing the C:\\-root tree-creation escape (AU:(AD) + AU:(OI)(CI)(IO)(M)) in workspace-write too. The unused WinLocalSid/WinInteractiveSid/WinAuthenticatedUserSid constants and their ABI-probe prints are removed; the enforcement 'full' claim now stands on a closed NTFS surface. New regression: a C:\\Users\\Public subdirectory write is denied under BOTH modes (the ambient-writable blind spot the review flagged — INTERACTIVE is absent from both lists). FAT-class (non-ACL) targets outside the granted roots remain writable (no security descriptors to intersect) — documented as a legacy residue, warn-only, not engineered around. Docs/design note/PR body updated in both languages (list I/J terminology gone everywhere).
A session that materialized its workspace-write grant and then switched to read-only (or crashed and resumed under read-only) kept a writable workspace for the server lifetime: the standing orphan-SID ACE survived the downgrade and list I still carried the orphan, so the write-restricted pass-2 check granted it. List I is now [logon SID, EVERYONE] only — the standing ACE stays inert under read-only while the unrevoked ACE keeps the re-upgrade free (map hit, no re-propagation). Pinned by the runner's real-token mode-downgrade regression (standing ACE denied under read-only, effective again on re-upgrade), the acl-session mode-switch cycle (nothing under read-only, one materialization on upgrade, no revoke on downgrade), and ConstrainedLanguage pins in both modes.
quoteArg doubles end-of-string backslashes (real CommandLineToArgvW round-trip test; first-token exemption noted); grantWrite merges into the current DACL instead of replacing it, and both grant/revoke hold a LockFileEx per-path lock under GetTempPathW/dsh-acl-locks (koffi crashes on NULL lpOverlapped — a zeroed OVERLAPPED is used); GRANT_MASK gains DELETE|FILE_DELETE_CHILD (never WRITE_DAC/WRITE_OWNER) though the win32 26200 second check only constrains the WRITE bit; failure paths close all handles (CreateProcessAsUserW pipe set, ResumeThread thread/process/job) with stub-api tests; getTempPathW refuses undersized buffers; drainPipe backs off; runner.spec pwshAvailable uses the resolvePwshPath probe; NTSTATUS exit codes mirror bit-exact (verified end-to-end); WinLocalSid JSDoc re-attributed to WinLocalLogonSid. All gates green: 42 passed/2 skipped, typecheck, oxlint, 0 clones, constraints, knip.
Known Limitations gains the per-command DACL mutation entry (inheritance is lazy, not a per-file walk; per-session grant reuse is deferred work); the design note gains a Testing section stating why the win32-only roster flip cannot ride macOS/Linux snapshot fixtures and naming the substitute evidence.
sandbox-local now depends on @deepseek-ai/dsh-sandbox-windows-acl, so the packed-install e2e must carry its tarball in the workspace closure — npm cannot resolve the private package from the registry. koffi and the shared peers (dsh-invariants, cordis) already resolve from the registry or the existing closure.
The self-hosted Linux runners ship no pwsh, and the pwshAvailable probes used spawnSync('where.exe'), which reports a missing binary as status null instead of throwing — the suites never skipped and failed with spawn pwsh ENOENT. Probe with resolvePwshPath() status instead, the same gate the coverage exemption uses. Exempt pwsh-sandbox src from coverage on pwsh-less hosts (its remaining helpers branch and the invariant companion ride the executor suites' real pwsh runs); pwsh-ful hosts keep the full 100% bar. Cover the windows-acl probe case and the runner-entry resolution in sandbox-local on Linux (chain-seam tests plus a windowsAclRunnerEntry seam) — the package's POSIX-only suites are Linux's only chance to cover the new lines. Static gate fixes: declare dsh-pwsh-sandbox in the base bundle, register the runner files entry in constraints, knip entries for the e2e suite and where.exe, regenerate the module graph. Verified in WSL (no-pwsh Linux): pwsh-sandbox 5 pass/13 skip with the exemption active, sandbox-local coverage 100%.
Add the implemented Agent Note owning the win32 rung decision: mxc rejected (OS floor 24H2, BaseContainer tier only on 25H2+ with the OS feature enabled, and arbitrary-path reads demand wholesale host DACL writes), AppContainer rejected (no ambient read access — arbitrary-path reads unsupported), landstrip already rejected and AppContainer-shaped. Link the note from the backend README pair and restate the choice in the PR description.
JSDoc every sandbox-windows-acl export (the 138 verify-export-jsdoc violations); fix the sandbox-windows-acl README code block (undeclared workspaceRoot), bring its stale claims current (the provider wiring and kill-on-close are live), add the canonical Model Experience sections to both new packages, audited the backend's indirect Model Experience sentence, and regenerate docs/config-catalog.md.
Confine Windows command execution through a WRITE_RESTRICTED token whose restricting SIDs carry an orphan-SID write allowlist, ported from https://github.com/huoyaoyuan/windows-acl-restrict-poc (@ 10e4dfb). Every Win32 call is checked and fails closed - the POC silently ran children with the FULL token when CreateRestrictedToken failed.
- @deepseek-ai/dsh-sandbox-windows-acl: koffi primitives verified against the MinGW Windows headers (verify/abi-probe.cpp) plus the confinement runner ([node, runner, --workspace, --temp, --mode, --, argv...]: kill-on-close job, stdio passthrough, exit-code mirroring, windows-acl-run: failure signature, grant revocation). read-only = strict zero grants (NUL device not writable; documented). Windows-only execution: exempted from the Linux coverage lane (windowsOnlyCoverageExclusions).
- @deepseek-ai/dsh-sandbox-local: PLATFORM_CHAINS.win32 filled with the windows-acl runner (full enforcement, ACL denial dialect, runner-failure rules).
- @deepseek-ai/dsh-pwsh-sandbox: sandbox-consuming pwsh executor (call-for-call mirror of dsh-bash-sandbox) over a new argv-level seam in dsh-pwsh-local; per-file coverage complete via the fake-provider spec.
- bundle/base: the Windows platform layer mounts the confined pwsh roster - sandbox/policy/fs-sandbox/permission/approval re-enabled, the POSIX bash stack stays disabled.
Co-authored-by: Huo Yaoyuan <huoyaoyuan@hotmail.com>
The parent implementation introduced sandboxMode and approvalPolicy as generic SessionHeader fields, then propagated those fields through both persistence backends, session-query indexes, collision checks, policy-specific seed-boundary folds, catalogs, and a broad test matrix. That storage plane is unnecessary: Session already accepts a validated constructor seed, and persistence captures that seed when the session is announced before committing its first batch.
Capture each parent override synchronously at delegation, append source-tagged sandbox/mode and approval/policy records after the optional fork prefix, and create the child with that combined seed. Keeping header.seedLength at the original fork-prefix length preserves lineage while ordinary last-event-wins folds make the inherited records outrank stale parent history and remain subordinate to later child switches. Unswitched parents still stamp nothing, so children continue to follow deployment defaults.
Remove the generic header fields and every persistence/query/schema branch built around them. Collapse the inheritance suite from ten leaking scenarios to four owned-context cases covering real filesystem confinement, stale fork precedence, delegation-time capture, and the no-override path. The assembled headless snapshot now asserts the persisted inheritance event directly.
This keeps the security behavior while restoring policy ownership to the existing event log and deleting the speculative durability machinery that the original tests did not exercise.
Review fix (ds-review-bot on #623): the READMEs and JSDoc still stated the
pre-baseline formulas — resolve() outranking 'the session's last
sandbox/mode event', 'effective = explicit grant ?? fold(events) ??
deployment default', and the approval README's 'last approval/policy event'
opener — which contradict the shipped semantics for a delegated fork whose
seed tail differs from its header baseline. Every statement now names the
override chain (own post-seed switches ?? inherited header baseline): both
READMEs (both languages), resolve()'s JSDoc, the session-mode module and
event-declaration docs, the raw folds re-scoped as building blocks, and
the regenerated catalogs.
Review fixes (ds-review-bot on #623):
- Persistence adoption compares the immutable policy baselines: onCreated's
ownerless claim and adoptLivePrefix retain the STORED header, so a
same-id live session with a conflicting baseline now rejects as a
collision instead of appending under read-only and resuming under the
stored danger-full-access.
- resolve() resolves the session override BEFORE applying an explicit
approved mode: the one-shot grant no longer bypasses the unconditional
durable-header validation.
- The approval narrator attributes positionally over the session's OWN
events (past the seed boundary): a fork child whose baseline delta has
no own override narrates 'inherited from the delegating session' instead
of misattributing a stale seed-carried switch to the user or the
operator.
Red-first: baseline-conflict adoption in the shared coordinator contract
(both backends), resolve-with-explicit-mode validation, and the fork-child
narration attribution case.
Review fixes (ds-review-bot on #623):
- overrideOf (both knobs) rejects a seedLength past the log end before
slicing: a malformed durable boundary would otherwise empty the
own-switch slice until the log outgrew it, letting a wide baseline
shadow a REAL later tightening. Malformed durable metadata fails loud,
never open.
- The session-query derived index carries the two baseline fields end to
end: schema columns on both session tables (SESSION_QUERY_SQLITE_SCHEMA
_VERSION 6 — derived, rebuilds in place), inserts, header selects, the
candidates CTE, rowHeader, sameHeader, and the cross-source
assertSessionHeadersCompatible — so a search hit's header keeps the
child's inherited confinement and conflicting live/persisted baselines
reject.
Red-first: out-of-range seedLength tests in both policy suites;
baseline round-trip and live/persisted baseline-conflict tests in the
session-query sqlite suite.
