Master removed the TUI package, the `meta` and `upgrade` subcommands, and
`--config-replace`, and made raw `dsh` require a `--config` overlay. Resolved
onto that shape:
- Dropped this branch's TUI edits with the surface itself, including
`tui.cordis.yml`, `runTui`, and the TUI keyless PTY smoke.
- Dropped the `--config-replace` plumbing rather than reintroducing a flag
master deliberately removed. The gap this branch fixed remains: `dsh -p`
still could not name its composition, so it keeps `--config`.
- Kept this branch's deletion of the personal `$DSH_HOME/config.yaml` layer,
which master still carried, and provided the environment snapshot in the new
raw `runConfig` surface alongside web and headless.
- Ported the headless shutdown PTY test off the personal overlay onto a named
`--config` file, which is what proves that flag now exists on `-p`.
Review found five real defects in the configuration-source work, all confirmed
against the code rather than argued:
1. The note claimed --config outranks settings.yaml. It does not: the settings
seam registers a plugin's cordis entry config as the `base` layer and the
user section layers over it, and the seam cannot tell a shipped value from a
--config one. The note now states shipped reality and names --config-replace
as the lever for a deployment that must win. Separately, a literal `apiKey`
in settings outranked both the environment and .credentials.yaml — the field
is removed, so configuration carries a reference and nothing else.
2. DEEPSEEK_SEARCH_BASE_URL was functionally deleted: the shipped inline went
away without the provider learning to read it. It now resolves from the
environment snapshot, as the README always claimed.
3. The bootstrap deny list missed the interpreter start-up hooks. BASH_ENV is
the sharpest: `bash -c` sources it on every bash tool call, so a project
.env could run a file of its choosing before every command. The list now
covers BASH_ENV and its per-language siblings, the Git hook commands, and
the remaining preload and CA variables, organised by what a variable does
rather than which runtime owns it.
4. YAML parse errors quoted the offending source line — which in a credentials
document is the secret — into boot stderr and the watcher's logger. Only the
error code and position are reported now, in credentials-local and
settings-local alike, pinned by a test that asserts the secret is absent.
5. 0600 governed only files the harness wrote. A hand-created 0644 document was
read normally. POSIX now checks the mode before reading contents, at boot
and on every reload; Windows has no mode to inspect and is skipped rather
than faked.
The project a session is launched in is trusted by default, with no prompt and
no stored trust record: it may supply its own endpoint, ordinary variables, and
a key ranked below the managed store. Trust stops at the harness itself — a
discovered file still cannot set DSH_PERMISSION_MODE, PATH, BASH_ENV, or the
rest, because those take effect with no user action, before any turn, outside
the permission policy and the sandbox.
$DSH_HOME/.env had just become an ordinary environment layer, which left the
harness resolving user-facing values from a flattened process.env that could
no longer say where a value came from. A key stored through the web page
stayed shadowed by an older key in the user's own .env. An endpoint could be
redirected by the project: the invoking directory's .env is materialized like
every other layer, and a base URL decides where a resolved API key is sent, so
a DEEPSEEK_BASE_URL written into a model-editable workspace would send the
user's credential — and the prompts carrying their code — to whatever host
that file named.
Give every user-facing value one ordering, with four kinds of source:
explicit for this run per-operation override, CLI argument
> authored by deployment --config / --config-replace
> this launch's shell inherited process environment
> product-managed store settings.yaml, .credentials.yaml
> discovered file $DSH_HOME/.env
> defaults schema default, shipped base, public default
The domains differ only in which tiers exist. The earlier split — credentials
ranking the environment over the managed file while settings ranked over the
environment — was inconsistent: the distinguishing fact is who authored the
source, not the domain.
packages/util/environment owns an immutable snapshot with per-layer
provenance. getFrom(name, sources) searches only the layers a caller names,
and omitting one is a refusal rather than a demotion: the adapters ask for
['process', 'user-env'], so no reordering can let a project file back into a
decision it was excluded from.
isBootstrapOnly rejects, before anything is materialized, any .env setting a
variable that governs how a process launches (PATH, SHELL, NODE_OPTIONS,
LD_PRELOAD), where code or model-visible instructions load from (the whole
DSH_* namespace, HOME, XDG_*), or how the network is reached (proxy and CA
variables). The namespace is denied wholesale so a switch added later cannot
become settable by being forgotten, and there is no opt-out.
verify-config-source-ownership keeps both rules: no unregistered process.env
read under packages/*/*/src (26 allowlisted with reasons), and no apiKey,
baseURL, or headers inlined from the environment in shipped Cordis config —
removing those inlines is what makes the deployment tier meaningful.
$DSH_HOME/config.yaml was an implicit composition layer: if the file existed,
every launch applied an arbitrary Loader patch graph over the shipped tree,
kept live by a dedicated HMR watcher. Three costs came from the implicitness,
not the capability. A patch replaces its target row's whole config, so a file
written months ago pins that row to the field set it knew and every default
the shipped tree later adds silently stops applying. It competed with the
typed settings namespaces llm-deepseek and llm-pi-ai already register, so
which one wins was a function of layer order rather than meaning. And the
explicit escape hatch it was supposedly redundant with did not exist on every
surface: dsh -p, dsh meta, and dsh upgrade all rejected --config, so for them
the implicit file was the only composition route at all.
Complete the explicit layer first: --config and --config-replace now work on
every booting surface. A headless --config-replace tree must still mount a
webserver row, because that surface reaches its own agent over the same HTTP
gateway the browser uses; AppCLIEntry names that contract in the failure
instead of reporting a bare missing service.
Then delete the implicit one. PERSONAL_CONFIG_FILENAME, loadPersonalPatches,
watchPersonalPatches, and the config-only HMR row mounted for it are gone; a
file left at that path is inert, and --dump-config no longer reads the Harness
home. --config therefore stops *replacing* the personal overlay and simply
*is* the user overlay.
No migration: a user who wants the old behavior names the same file
(dsh --config ~/.dsh/config.yaml), which a shell alias makes permanent.
$DSH_HOME/.env carried two incompatible jobs. As credentials-local's writable
secret store it could not be hoisted into process.env — hoisting makes every
stored key read as a read-only launch override and blocks rotation from the
TUI and the web page. But its name and dotenv format promise an environment
file, so a DEEPSEEK_BASE_URL sitting beside a working DEEPSEEK_API_KEY in the
same file was silently ignored: only the credential provider read the
document, and it addresses credential references alone.
Split the two jobs into two files.
.credentials.yaml is the provider-managed store: a strict YAML mapping of
CredentialRef to non-empty string, no version field, no wrapper level. Because
it holds credentials and nothing else, a non-mapping root, a non-identifier
key, a non-string value, an empty string, a duplicate key, and malformed YAML
are all rejections rather than skipped entries — loud at boot and at a write,
warn-and-keep-last-good on a live reload. The dotenv physical-line editor
gives way to a patch of the parsed document, so comments and untouched entries
keep their formatting and any string value round-trips, multi-line included.
Writer lock, read-modify-write, atomic 0600 write under a 0700 directory,
watcher, self-write suppression, and quiescent disposal are unchanged.
$DSH_HOME/.env becomes the user's ordinary environment layer. app-boot's new
loadLayeredEnv loads the invoking directory's .env then the Harness home's,
giving user < project < inherited; the home resolves from the inherited
environment first, so a project .env cannot redirect it.
Credential precedence is unchanged: the live environment still wins read-only
over the file, and shadowed writes still reject. Whether a provider-managed
store should instead win over the environment is a separate decision.
No migration: a key already in $DSH_HOME/.env keeps resolving through the new
environment layer, as a read-only env source that shadows the stored one.
The goal scenario asserted a paused revision-2 goal, but the pause is
appended only after cancellation reaches idle — after turn/end. Under
parallel snapshot files the subprocess could dispose before the pause
record persisted, folding the log to an active revision-1 goal.
Add a waitForEventAfterTurnEnd input step (the turn-end/title waiters'
shape, parameterized by event type) and use it in the goal scenario to
hold the subprocess open until the goal-state record lands.
The injected instruction persists as a durable user message, so an
unscoped 'Do not call any more tools.' stays in every later request's
history. Scope it: '...in this run; further work waits for the user's
next instruction.' A/B probes on deepseek-v4-pro show the scoped wording
is non-inferior in-turn (4/4 zero tool calls, closing quality unchanged)
and next-turn tool use is unaffected under both wordings; the scoped
form states the instruction's actual lifetime.
A goal round reporting complete or blocked used to conclude the turn at
the tool result, so the model never spoke after the call and sessions
ended on a bare update_goal card. The terminal update now defers one
plugin-sourced <goal_complete>/<goal_blocked> instruction onto its
result asking for a grounded closing message without further tool
calls; the turn then ends through the ordinary no-tool-calls stop.
Direct-human mutations stay uninstructed. Wording chosen by A/B
sampling on deepseek-v4-pro; one extra request per goal lifecycle.
New keyless ACP snapshot goal-wrapup drives the shipped app through
create -> round one -> autonomous complete and pins the injection, the
same-turn closing message, and the completed turn end.
- listChildren() takes an optional AbortSignal and rechecks it after every
un-signalled session-query await (the cold-resume cooperative-cancellation
boundary); list_agents forwards exec.signal so the registry's drain of
started tool bodies cannot block on a slow or large catalog.
- The list_agents description now presents running/complete as a stored-record
snapshot and defers deliverability to send_message, matching the ownership-
conflict semantics the service tests pin.
Implements the durable-subagent-catalog RFC: SubagentControlService.listChildren()
enumerates a parent's direct continuable children from one sessionQuery trace,
validates each child's sole subagent/descriptor event (now carrying the durable
creation label), and returns one ordered SubagentListEntry[] with per-child
corrupt/unsupported/unavailable diagnostics. The list_agents tool ships as a
separately loadable plugin of dsh-tool-subagent-control requiring sessionQuery
at load; send_message stays usable without it.
- 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.
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.
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.
- 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.
Implement the continuable background subagents RFC: a durable child
session with a series of Task-backed activations, each disposing its
run before the Task settles.
- dsh-subagent: rename SubagentRun.sendMessage to strict steer, drop
run-level resume, add SubagentProvider.resume dispatch via
SubagentService.resume, the continuation start field, and the
versioned model-hidden subagent/descriptor session event.
- dsh-subagent-inprocess/-spawn/-fork: publish the control-allocated
child id, append the descriptor inside the initial turn, implement
cold resume from the child's own transcript under the live parent
scope, and strict running-only steer.
- dsh-subagent-control (new): SubagentControlService owning stable
child ids, descriptor snapshot/fold/authorization, Task-backed
activation with settle-then-dispose ordering, the process-local
active-run association, and steer-or-resume sendMessage routing.
- dsh-tool-subagent: background route branches on the provider's
resume capability (continuable via the control service; one-shot
task for ACP), returning both child and task ids.
- dsh-tool-subagent-control (new): the globally named send_message
tool rendering steered/started routes.
Keyless coverage spans Task ownership and disposal ordering, running
delivery, cold follow-up, descriptor rejection and rollback, known-id
reconstruction, kill during lookup, admission races, and a new
subagent-continuable ACP snapshot scenario.
ds-review-bot round 1 on the repository-plugin runtime:
- a manifest-declared skill root absent or non-directory in the installed
package now fails the plugin load (skill-local treats a missing root as
legitimately empty, which silently mounted a skill-less plugin)
- includeDefaultRoots: false no longer inherits $DSH_BUNDLED_SKILL_DIR, so
isolated repository providers see only their explicit roots
- prepared wrapper baseUrl schema requires the file: scheme, failing hostile
URLs at the declared validation boundary
- preparedPath reuses format.ts's isOutside; SERVER_NAME_PATTERN is exported
and pinned equal to dsh-mcp-client's, with the restatement justified (the
prepare bin keeps a zod-only module graph); the unexplained `as never`
cast now carries its schemastery rationale
- the import-free wrapper assertion also rejects dynamic import(
- the headless fixture wrapper is regenerated by the real prepareDshPlugin
and a drift test pins fixture == generator output
- prepareDshPlugin JSDoc states the non-atomic publish repair contract
The session-fixture-layout gate requires every session JSONL fixture in
the canonical packed layout (maximal delta runs per kind/block, as
migrate:packed-session-fixtures rewrites); the packChunks: false knob
violated that invariant and failed the snapshot job. Revert the knob and
canonicalize the recorded fixture instead: the live log's
eager-drain-packed rows migrate to the maximal-run layout a burst replay
reproduces, so the fixture stays replay-deterministic and canonical.
The scenario previously carried an authored fixture; W4 of #1119 review
requires a live transcript. Recording surfaced two composition bugs that
are fixed here alongside it:
- provider ids: the app and the replay catalog both named the old
'deepseek' provider, which no adapter registers; both now use
'deepseek-official'
- the live config lacked persistenceCompression: none, so record-mode
sessions were written zstd-compressed and could not be harvested
(the snapshot twin already forced plaintext)
Recorded logs also need deterministic replay:
- packChunks: false in both configs — the eager-drain batch boundaries
that split packed delta runs are timing-dependent, so a packed log of
a long reasoning stream cannot replay-match its live record
- the fixture's request/header config and request/context are normalized
to the replay-produced minimal shape (the live adapter logs model
capabilities llm-replay has no data for), and tool-result path
separators are canonicalized to '/' for the Linux golden
posixOnly is restored now that the fixture is recorded.
glob/grep now run the @vscode/ripgrep binary via ctx.subprocess with a plain
argv vector: no system rg install, no shell layer, unconditional registration.
The load-time command -v rg probe and the bash-seam coupling are removed;
timeouts ride the cooperative exec.signal plus the seam's terminate
escalation. The fs-glob-sampling ACP snapshot executes the real packaged
binary against an mtime-pinned fixture. Adds the packaged-ripgrep-search
Agent Note, updates the roster-note facts and both shipped-composition e2es,
and regenerates the doc catalogs and third-party notices (surfacing
pre-existing manifest drift plus the new @vscode/ripgrep row; the notices
generator also learns pnpm 11's truncated virtual-store names).
Yichen Jiang
Turtle
Huanqi Cao
ZiyaZhang
imccyu
Dudu-0223
Dudu
Tianyi Cui