Files
deepseek-harness/packages/cordis/tool-cordis
kingwl 669771097d subagent: inherit parent sandbox/approval overrides in in-process children
Per-session policy overrides (sandbox/mode, approval/policy) never crossed
the delegation boundary: a spawn child of a read-only-switched parent ran
under the wider deployment default, and a fork child missed any switch made
after its seed boundary — delegation was a bypass channel for a user's
tightening.

The in-process driver now snapshots the delegating parent's override chain
and stamps it onto the child through the canonical write paths
(SandboxPolicyService.inheritOverride / ApprovalService.inheritOverride),
anchored inside the child's first turn via a one-shot agent/prompt-submit
listener: turn-enclosed (durable), ahead of the first request (an inherited
'never' reaches the child's first system prompt), and positioned after any
stale fork-seed switch so the ordinary last-event-wins fold resolves it.
Only overrides are copied — an unswitched parent stamps nothing and the
child follows the live deployment default; both services are consumed
opportunistically, so compositions without them delegate unchanged. Nesting
composes by construction (each stamp folds the already-stamped parent log).

Evidence: inheritance.spec.ts drives scripted-model children into the real
dsh-fs-sandbox fence through the real write tool (disk-state + denial-marker
assertions; spawn, stale-seed fork, grandchild, escalation fail-closed, and
no-stamp guards), inheritOverride contract tests in both service suites, and
the recorded subagent-sandbox-inheritance ACP snapshot (read-only preset →
delegate → child denied, replayed keylessly).

See .agents/notes/implemented/feature/2026-07-25-subagent-policy-inheritance.md.
2026-07-25 04:06:19 +08:00
..
2026-07-19 22:52:03 +08:00

@deepseek-ai/dsh-tool-cordis

The self-referential cordis toolset: three model-facing tools over the live runtime the agent runs inside. Design home — sandbox semantics, mount lifecycle, cross-mount composition, the generated API catalog, standing decisions: the toolset Agent Note.

What it does

  • cordis_inspect — read-only report over the runtime: services, the loaded-plugin list, registered tools, the dynamic-mount table, and the catalog-backed api / events references. An exact name with what: "api" or what: "events" narrows the report and adds the original source JSDoc.
  • cordis_mount — evaluates model-written JavaScript (the body of an async function) in a node:vm sandbox; the code must return a cordis plugin, which is mounted under the cordis-dynamic group fiber and tracked as dyn-<n>.
  • cordis_unmount — disposes one mount by id, returning only after quiescence.

Exact model-facing schemas: the generated tool catalog.

Trust stance

The sandbox isolates globals but is not a security boundary. Node globals are absent or redirect to Cordis services such as ctx.fs, ctx.web, and ctx.bash, and writes to globalThis stay local, but host-realm helpers make escape possible. Mounted plugins receive a façade without framework internals, yet its allowed services affect the live runtime. Treat this toolset like bash access; see the design and trust stance.

Config

Field Default Meaning
vmTimeoutMs 5000 Bound on the SYNCHRONOUS portion of mount-code evaluation; an async body escapes it

The generated API catalog

src/api-catalog.ts is generated by scripts/gen-cordis-api.ts from the same AST walk as docs/cordis-catalog and freshness-gated by pnpm run verify-cordis-api (in doc-sync) — never edit it by hand. cordis_inspect intersects it with the live service store at call time. Broad api / events reports render summaries and signatures only; an exact name opts into the retained method/event JSDoc, and unknown or non-running service targets fail loud.

Rendering

All three tools render generic cards (read / execute / delete); cordis_mount carries the mount code as rawInput. Presenters are pure functions of the args; results keep the default text rendering.

Export shape

Namespace plugin: named exports name / inject / Config / apply, no default export (docs/postmortem/0001).

Model Experience

Tool schemas

What the model sees

The conversation model sees the generated cordis_inspect, cordis_mount, and cordis_unmount schemas whenever this plugin is visible.

Token effect

Fixed schema cost on every request in that tool view.

KV Cache effect

Prefix-stable while this tool view is unchanged. Scoping or plugin lifecycle changes that hide these definitions may invalidate reuse from the first changed schema token.

Tool-call history and results

What the model sees

Inspect joins selected sections exactly as ## <section> then a newline and the data-dependent body, with one blank line between sections. Its broad API/event reports omit JSDoc; name with what: "api" or what: "events" returns one exact target with its original JSDoc. Mount returns mounted <id> (plugin "<name>", state: <state>), optionally inserting — waiting for service(s): <names> (activates when provided) before the closing parenthesis. Unmount returns unmounted <id> (plugin "<name>"); an unknown id becomes Error: no dynamic plugin with id "<id>" (list mounts with cordis_inspect what:"dynamic"). The submitted mount program remains in the assistant tool-call history.

Token effect

Inspect output and mount code are data-dependent and resent until compaction; lifecycle acknowledgements are small.

KV Cache effect

Append-only; newly visible content follows the reusable request prefix and does not invalidate existing KV-cache entries.

Later requests after a mount

What the model sees

A mounted plugin may register tools, prompt contributions, or listeners that change later requests for the scopes it targets; unmount removes those contributions after quiescence.

Token effect

Indirect token impact equals the mounted plugin's contributions and lasts only for the mount lifetime.

KV Cache effect

Mounting or unmounting a prompt or tool contribution changes later request prefixes and may invalidate reuse from the first changed contribution; an unchanged mount set remains prefix-stable.

Known Limitations and Deferred Work

  • The sandbox is containment for honest code, not a security boundary — host-realm helpers on the sandbox global are reachable, so mount code can reach Node; load this plugin as deliberately as you would grant a bash tool (see § Trust stance).
  • The ctx façade exposes no effect() — mount code cannot register a bespoke disposer; on/provide/tools.register cover every mount seen so far, and a guarded effect waits on a real need (FIXME(sandbox-effect)).
  • vmTimeoutMs bounds only synchronous evaluation — an async mount body escapes it; there is no async budget on mount code.