capsule_runtime/types.rs
1use std::{
2 collections::{HashMap, HashSet},
3 path::PathBuf,
4 sync::Arc,
5};
6
7use murmur_artifact::{
8 ArtifactImplementation, ArtifactRuntime, ContextConfig, HookConfig, InferenceConfig,
9 LifecycleConfig, LifecycleOverride, Registry,
10};
11use wasmtime::{component::Component, Engine};
12
13use crate::{
14 bindings::host::murmur::tool::run::ToolResult,
15 hooks::ShellDispatchInfo,
16 limits::{EpochTicker, ExecutionLimits},
17 network_policy::{HookCapabilityGrant, ToolCapabilityGrant},
18 spawn_credential::{SpawnApproval, SpawnCredential},
19};
20
21pub(crate) struct DispatchOutcome {
22 pub result: ToolResult,
23 pub shell: Option<ShellDispatchInfo>,
24 /// Set instead of [`Self::shell`] when the command outran its grace period and was demoted
25 /// to the background. The two are never both `Some`: a demoted command has no exit code yet,
26 /// which is exactly why `HookEvent::Shell` does not fire for one.
27 pub detached: Option<crate::detached::DetachedDispatchInfo>,
28 /// True when the dispatch was served by reading a skill.md file rather than running
29 /// a binary or WASM component. Used by the agent loop to write a `skill_call` trace
30 /// event instead of a `tool_call` event.
31 pub is_skill: bool,
32 /// The fence source `result.data` was wrapped under, or `None` when the result carries no
33 /// fence. Set by [`crate::runtime::CapsuleStoreState::dispatch_agent_tool_async`] on the
34 /// same branch that applies the fence, so the label and the wrapping cannot drift: a caller
35 /// reading this never re-derives a source name by parsing the content back.
36 ///
37 /// `None` on every outcome that leaves the private unfenced dispatch — the skill branch,
38 /// and [`crate::runtime::CapsuleStoreState::dispatch_submit_plan`]'s route, whose step
39 /// output is fenced later as one field of a report.
40 pub fence_source: Option<String>,
41 /// Set when this dispatch failed in a way that ends the *session* rather than just this
42 /// tool call — today only a `sealed` composed-root construction failure
43 /// ([`crate::shell::ShellExecError::session_fatal`]). `result` still describes the failed
44 /// call so the trace records what was attempted; the agent turn loop then returns this
45 /// error instead of feeding the failure back to the model for another turn.
46 ///
47 /// A dispatch that merely failed leaves this `None`: a capsule reacting to a broken tool
48 /// call is ordinary, whereas a capsule continuing to run after its declared containment
49 /// boundary stopped being establishable is the silent-degradation failure mode the whole
50 /// containment-class mechanism exists to prevent.
51 pub fatal: Option<crate::errors::RuntimeError>,
52}
53
54impl DispatchOutcome {
55 pub(crate) fn tool(result: ToolResult) -> Self {
56 Self {
57 result,
58 shell: None,
59 detached: None,
60 is_skill: false,
61 fence_source: None,
62 fatal: None,
63 }
64 }
65
66 pub(crate) fn skill(result: ToolResult) -> Self {
67 Self {
68 result,
69 shell: None,
70 detached: None,
71 is_skill: true,
72 fence_source: None,
73 fatal: None,
74 }
75 }
76}
77
78#[derive(Debug, Clone, PartialEq, Eq)]
79pub struct ArtifactRequest {
80 pub name: String,
81 pub version: String,
82 pub runtime: ArtifactRuntime,
83 /// Optional local source path (skill artifacts only). When set, the runtime resolves the
84 /// skill's `skill.md` directly from this path and skips registry resolution entirely.
85 /// Relative paths resolve against `StageRequest::manifest_dir`.
86 pub source: Option<String>,
87 /// Per-artifact capability grant copied verbatim from this artifact's entry in the
88 /// capsule operator's own manifest (`murmur_artifact::RuntimeArtifact::capabilities`).
89 /// Consumed for `runtime: hook` (default-deny baseline, `None` = deny everything) and for
90 /// `runtime: tool`/`runtime: driver` (ceiling baseline, `None` = the capsule-wide policy
91 /// unchanged); the manifest parser rejects the key on `runtime: skill`. Never sourced
92 /// from the artifact's own bundled `murmur.yaml`, so an artifact cannot self-grant.
93 pub capabilities: Option<murmur_artifact::Capabilities>,
94 /// What the runtime does when this hook's async job queue is full, copied verbatim from
95 /// `on_overflow:` on this artifact's entry in the capsule operator's own manifest — the
96 /// same operator-only sourcing rule `capabilities` above follows. Consumed for
97 /// `runtime: hook` and inert for every other role (the manifest parser rejects the key
98 /// there); inert too for a hook that turns out to be `execution_mode: blocking`, which is
99 /// never queued.
100 pub on_overflow: murmur_artifact::HookOverflowPolicy,
101 /// This artifact's operator-declared `config:` block, copied verbatim from its entry in the
102 /// capsule operator's own manifest (`murmur_artifact::RuntimeArtifact::config`) — the same
103 /// operator-only sourcing rule `capabilities` above follows, so an artifact pulled from a
104 /// registry cannot configure itself.
105 ///
106 /// Lowered to compact JSON at staging by [`crate::artifact_config::lower_artifact_config`] and
107 /// delivered to this artifact alone as [`crate::artifact_config::ARTIFACT_CONFIG_ENV`]. `None`
108 /// means the variable is absent from the guest environment. Consumed for `runtime: hook`,
109 /// `runtime: tool` and `runtime: driver`; the manifest parser rejects the key on
110 /// `runtime: skill`, and a native tool warns `W-SEC-015` because it reads no per-artifact
111 /// environment at all.
112 pub config: Option<serde_yaml::Value>,
113 /// This artifact's operator-declared `gateway:` block, copied verbatim from its entry in the
114 /// capsule operator's own manifest (`murmur_artifact::RuntimeArtifact::gateway`) — the same
115 /// operator-only sourcing rule `capabilities` above follows. Built into the artifact's
116 /// credential gateway at staging, which is also where the artifact's own `upstream_auth:`
117 /// declaration is read; an artifact without this block is never asked for one.
118 pub gateway: Option<murmur_artifact::ArtifactGateway>,
119}
120
121/// What `murmur.lock` pins one artifact at, already narrowed to this host.
122#[derive(Debug, Clone, PartialEq, Eq)]
123pub struct LockExpectation {
124 pub name: String,
125 pub resolved_version: String,
126 /// The hash the resolved payload must have. Selected by the caller from the lock entry's
127 /// `sha256` for the platform staging runs on, so it is whichever of `any` or
128 /// `platforms.<platform>` applies to this host.
129 pub sha256: String,
130}
131
132/// What staging actually resolved, for the caller to write back into `murmur.lock`.
133#[derive(Debug, Clone, PartialEq, Eq)]
134pub struct ResolvedLockArtifact {
135 pub name: String,
136 pub resolved_version: String,
137 /// Hash of the payload that was resolved on this host.
138 pub sha256: String,
139 /// Which platform the resolved payload is filed under, `None` for a platform-independent
140 /// one. Decides whether the lock entry pins `sha256.any` or `sha256.platforms.<platform>`.
141 pub platform: Option<String>,
142}
143
144#[derive(Debug, Clone, PartialEq, Eq, Default)]
145pub struct CapabilityPolicy {
146 pub network_allow: Vec<String>,
147 /// From `capabilities.peer_fetch.allow`: the peers this capsule may redeem a peer-file handle
148 /// against. Empty by default, which is deny — and which is also why no `fetch-peer-file` tool
149 /// manifest is written and the capsule's model never sees the tool exists.
150 ///
151 /// A separate list from [`Self::network_allow`], enforced with the same rule matcher but
152 /// never merged into it. Ingesting a peer's bytes lands a file in this capsule's own workdir,
153 /// so it is a prompt-injection surface and deserves its own operator control; a destination
154 /// in `network.allow` is not redeemable unless it also appears here, and declaring it here
155 /// does not widen `network.allow`.
156 pub peer_fetch_allow: Vec<String>,
157 /// From `capabilities.network.unix_sockets`. `false` by default, which makes the shell
158 /// subprocess tree's `socket(AF_UNIX, ...)` calls fail with `EACCES` at the kernel level on
159 /// both Linux tiers — `network_allow` above governs IP destinations only and would otherwise
160 /// leave a local daemon socket (`/var/run/docker.sock` — host root) completely unmediated.
161 /// `true` simply omits that seccomp rule; it is capsule-wide, not per-socket-path. See
162 /// `sandbox::denied_socket_domains`.
163 pub unix_sockets_allowed: bool,
164 pub filesystem_scope: Option<String>,
165 /// From `capabilities.filesystem.workdir_exec`. `false` by default, which withholds the
166 /// Landlock `Execute` right from the workdir's own `PathBeneath` rule — see
167 /// `sandbox::linux_enforce::workdir_access_rights`. With it withheld the kernel refuses to exec
168 /// anything under the workdir on the resolved path itself, so `shell_allow` is enforced
169 /// completely: a binary copied or compiled into the workdir cannot run under *any* name.
170 ///
171 /// `true` gives the right back for compile-and-run workflows and accepts, explicitly, that
172 /// `shell_allow` is then unenforceable for anything inside the workdir. That is why this field
173 /// is the one *grant* in this struct that also lowers the session's achieved containment class
174 /// (see `crate::containment::achieved_containment_class`) — nothing else here can.
175 pub workdir_exec_allowed: bool,
176 /// From `capabilities.filesystem.read_only`: workdir subtrees the capsule may read but must
177 /// not write, in the same vocabulary [`Self::filesystem_scope`] uses. Empty by default, which
178 /// is the whole workdir writable.
179 ///
180 /// Carried here as the declared strings and lowered once at staging into
181 /// `crate::protected_paths::ProtectedPaths`, which is what the dispatch check consults — so a
182 /// malformed entry refuses the launch rather than one call at a time. Unlike every other field
183 /// here this one grants nothing: it can only subtract from what the workdir preopen already
184 /// allows.
185 pub read_only_paths: Vec<String>,
186 pub shell_allow: Vec<String>,
187 pub spawn_allow: Vec<String>,
188 pub shell_strip_env: Vec<String>,
189 pub shell_baseline_env: Vec<String>,
190 /// Typed interpreter-runtime grants from `capabilities.shell.interpreter_runtime`: each
191 /// names an already-allowlisted binary and the exact host directories outside the workdir
192 /// its import machinery needs, with a per-directory `list_dir` enumerability flag. Empty
193 /// unless the manifest declares them. Consumed on `KernelFull` to add narrow Landlock
194 /// grants (see `sandbox::resolve_interpreter_runtime_grants`) and to fire `W-SEC-009`.
195 pub shell_interpreter_runtime: Vec<murmur_artifact::InterpreterRuntimeGrant>,
196 /// Typed staged-runtime grants from `capabilities.shell.staged_runtime`: each names an
197 /// already-allowlisted binary, the absolute host path of a pinned runtime tree, and the `pin`
198 /// identifying that tree's build. Empty unless the manifest declares them.
199 ///
200 /// Consumed at staging to enforce that declaring one requires an effective `sealed` floor (see
201 /// `crate::staged_runtime::check_staged_runtime_floor`) and to render the `staged runtime`
202 /// section of `--explain-scope`. Per-binary mutually exclusive with
203 /// [`Self::shell_interpreter_runtime`], which the manifest parser already guarantees.
204 pub shell_staged_runtime: Vec<murmur_artifact::StagedRuntimeGrant>,
205 /// Host env var names a WASM guest may observe, from `capabilities.env.allow`. Empty by
206 /// default: a guest sees only the runtime's own `MURMUR_*` injection unless the manifest
207 /// names a variable here. `shell_strip_env` and the credential patterns still apply on
208 /// top, so a credential-shaped name declared here is stripped anyway.
209 pub env_allow: Vec<String>,
210 /// Execution limits applied to every guest store in the session. Always fully
211 /// resolved — `capability_policy_from_runtime_manifest` substitutes a default for
212 /// each field `capabilities.limits` omits, so there is no "unset" state here and a
213 /// silent manifest yields [`ExecutionLimits::default`] rather than no limits.
214 pub limits: ExecutionLimits,
215 /// OS-level bounds applied to every *native subprocess* the session spawns — rlimits, the
216 /// Linux cgroup v2 scope, and the workdir-size ceiling. Resolved on exactly the same terms
217 /// as `limits` above (each omitted `capabilities.resources` field replaced by its default,
218 /// no "unset" state), but a completely different subject: `limits` bounds a WASM guest
219 /// inside its wasmtime store, this bounds the processes the host forks. See
220 /// [`crate::resources`].
221 pub resources: crate::resources::HostResourceLimits,
222 /// The containment class this capsule asked for, from `capabilities.containment`.
223 ///
224 /// Unlike every other field here it is a *requirement*, not a grant — and unlike
225 /// `StageRequest::declared_containment_floor` it is only the manifest's own vote, before the
226 /// workspace config and `--containment` are folded in. `stage_session` overwrites it with the
227 /// combined floor as soon as that is known, so anything reading it off a `StagedSession` sees
228 /// the effective value; the manifest-only value exists so a `CapabilityPolicy` built straight
229 /// from a manifest is still self-consistent.
230 ///
231 /// Read by `sandbox::ShellEnforcement::resolve` to decide whether a sealed-capable host
232 /// actually installs a composed root for this session — a capsule that declared `scoped` must
233 /// keep getting `scoped`'s mechanism. See `sandbox::applied_tier`.
234 pub containment_floor: murmur_artifact::ContainmentClass,
235 /// Whether the capsule's own top-level `capabilities.state` block was declared.
236 ///
237 /// Not a grant, and the only field here that is not: a durable state store is granted per
238 /// *artifact*, so a capsule-wide declaration reaches nothing at all. It is recorded so
239 /// `stage_session` can say that once (`W-SEC-014`) instead of leaving an operator to discover
240 /// an empty directory, and it is read by nothing else.
241 pub state_declared: bool,
242 /// Recorded for the same reason [`Self::state_declared`] is, and applied just as little: the
243 /// `murmur:conversation/read` grant is per artifact, so a capsule-wide declaration reaches
244 /// nothing and `stage_session` says so once (`W-SEC-016`).
245 pub conversation_declared: bool,
246 /// From `capabilities.plan.submit`: whether this capsule's model is offered the
247 /// runtime-provided `submit-plan` tool. `false` by default, which is deny — and which is also
248 /// why no `submit-plan` tool manifest is written and the capsule's model never sees the tool
249 /// exists.
250 ///
251 /// It grants no reach of its own: every step of a submitted plan runs through this same
252 /// session's tool dispatch, its shell allowlist and its spawn grant, so a plan reaches exactly
253 /// what the model could already call one call at a time.
254 pub plan_submit: bool,
255 /// `capabilities.limits.deadline_seconds` exactly as the manifest declared it — `None`
256 /// when it declared nothing. Retained undefaulted alongside the fully-resolved `limits`
257 /// above purely so hook calls can apply their own, lower default without mistaking an
258 /// explicit `600` for silence. See [`Self::hook_limits`].
259 pub declared_deadline_seconds: Option<u64>,
260}
261
262impl CapabilityPolicy {
263 /// Execution limits for hook lifecycle calls — [`Self::limits`] with the hook-specific
264 /// deadline default substituted when the manifest declared none.
265 ///
266 /// A hook call is one bounded piece of work, so it does not inherit the capsule-wide
267 /// ten-minute budget that exists for a capsule's whole `run`: a wedged hook would
268 /// otherwise stall the session for most of that, every event. An explicitly declared
269 /// `deadline_seconds` still wins, for hooks as for every other guest.
270 #[must_use]
271 pub fn hook_limits(&self) -> ExecutionLimits {
272 self.limits.for_hook_calls(self.declared_deadline_seconds)
273 }
274}
275
276#[derive(Debug, Clone, PartialEq, Eq)]
277pub struct InstalledArtifactSummary {
278 pub name: String,
279 pub version: String,
280 pub runtime: ArtifactRuntime,
281 pub implementation: Option<ArtifactImplementation>,
282}
283
284#[derive(Clone)]
285pub(crate) struct StagedHookArtifact {
286 pub name: String,
287 pub version: String,
288 pub component: Component,
289 /// Behavioral contract (binding/execution_mode/commit_policy) from the hook artifact's
290 /// *own* bundled manifest. Carries no capability information — see `grant`.
291 pub config: HookConfig,
292 /// What this hook is allowed to do, lowered and validated at staging time from the
293 /// operator's own manifest entry. Applied identically by all three hook instantiation
294 /// paths in `hooks.rs`; `HookCapabilityGrant::default()` is full default-deny.
295 pub grant: HookCapabilityGrant,
296 /// What to do when this hook's async job queue is full — from the operator's manifest
297 /// entry (via [`ArtifactRequest::on_overflow`]), on the same terms as `grant` and never
298 /// from `config`. Read only when `config.execution_mode` is `Async`; a blocking hook has
299 /// no queue to overflow.
300 pub on_overflow: murmur_artifact::HookOverflowPolicy,
301 /// This hook's own credential gateway, when its operator entry declares `gateway:`. Attached
302 /// to every store the hook is instantiated on, `on-stage` included. Filled once the session's
303 /// gateways are staged, before any hook runs.
304 pub gateway: Option<Arc<crate::credential_gateway::CredentialGateway>>,
305}
306
307/// How `mur run --resume` puts the loaded conversation in front of the model.
308///
309/// `Full` is often the cheaper of the two: a verbatim reload can land on the provider's own
310/// prompt cache, while compaction changes the prefix from the first altered token, guarantees a
311/// cache miss, and costs an inference call to produce the summary. `Compact` is the answer when
312/// the conversation would not fit the context window at all.
313#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
314pub enum ResumeMode {
315 #[default]
316 Full,
317 Compact,
318}
319
320/// One operator act: continue *this* conversation, whatever the capsule's own
321/// `lifecycle.conversation` policy says about carrying one between tasks.
322///
323/// Sugar over [`StageRequest::context_id`], never a second load path: the CLI resolves a session
324/// address to the context id that session ran under and fills both this and `context_id`. What
325/// this adds is the override — the record is loaded even under `lifecycle.conversation:
326/// stateless` — and the provenance the trace records.
327#[derive(Debug, Clone)]
328pub struct ResumeRequest {
329 /// The prior session's id, recorded verbatim as `session_start.resumed_from`.
330 pub from_session: String,
331 pub mode: ResumeMode,
332}
333
334#[derive(Debug, Clone)]
335pub struct StageRequest {
336 pub manifest_dir: PathBuf,
337 pub capsule_name: String,
338 pub capsule_version: String,
339 pub capsule_component_bytes: Vec<u8>,
340 pub artifacts: Vec<ArtifactRequest>,
341 pub allowlisted_tools: HashSet<String>,
342 pub lock_expectations: Option<Vec<LockExpectation>>,
343 pub capability_policy: CapabilityPolicy,
344 pub inference: Option<InferenceConfig>,
345 /// The global config file a `${NAME}` in an artifact's `gateway.api_key` is looked up in first, as
346 /// `credentials.NAME`, and re-read from while the session runs. `None` looks in the launching
347 /// environment only.
348 pub credentials_file: Option<PathBuf>,
349 /// Whether `inference` above had its system-prompt declaration replaced by the caller
350 /// (`mur run --system-prompt`) rather than read from the manifest. Purely a provenance
351 /// signal for the trace: the override itself is already applied to `inference`, which by
352 /// this point is indistinguishable from a manifest that declared the same prompt inline.
353 /// `true` even when the override cleared the prompt to nothing.
354 pub system_prompt_overridden: bool,
355 pub context: Option<ContextConfig>,
356 /// Context id every `task.md` task of this launch runs under, from `mur run --context`.
357 /// `None` mints a fresh one per task. Validated by `stage_session` as one path segment: it is
358 /// a directory name in the conversation record path.
359 pub context_id: Option<String>,
360 /// `mur run --resume <session>`, already resolved: `context_id` above holds the context that
361 /// session ran under, and this carries the session's own id and the mode. `None` is every
362 /// ordinary launch. `stage_session` refuses a resume whose context kept no record on disk,
363 /// and a `--resume-mode compact` with no hook bound to `on-compaction`.
364 pub resume: Option<ResumeRequest>,
365 /// `mur run --forget-session`: drop the harness session `context_id` names before this
366 /// launch's first task, so that task starts a new conversation under the same context id.
367 /// `false` on every ordinary launch. `stage_session` refuses it for a capsule whose transport
368 /// keeps no harness session.
369 pub forget_session: bool,
370 /// OTLP/HTTP endpoint for span export; None = no external OTel emission.
371 pub otel_endpoint: Option<String>,
372 /// JSON-serialized EvalConfig injected into hook WASI env as MURMUR_EVAL_CONFIG.
373 pub eval_config_json: Option<String>,
374 /// Dataset case identifier injected as MURMUR_CASE_ID when running a dataset.
375 pub case_id: Option<String>,
376 /// Dataset identifier injected as MURMUR_DATASET_ID when running a dataset.
377 pub dataset_id: Option<String>,
378 /// Lifecycle config from the manifest (None = use LifecycleConfig::default()).
379 pub lifecycle: Option<LifecycleConfig>,
380 /// Optional override applied on top of lifecycle (e.g. from CLI flags or mur-roost).
381 pub lifecycle_override: Option<LifecycleOverride>,
382 /// Trace config from the manifest (None = defaults apply: input captured, output not).
383 pub trace: Option<murmur_artifact::TraceConfig>,
384 /// Optional user project directory to mount as the capsule's accessible workspace.
385 /// When set, session artifacts go into `<workdir>/.murmur/<session_id>/`.
386 /// When None, uses the default temp dir (existing behavior).
387 pub workdir: Option<std::path::PathBuf>,
388 /// Address to bind the HTTP server on. Defaults to "127.0.0.1" for local runs.
389 /// Use "0.0.0.0" when deploying to a VM so the capsule is reachable externally.
390 pub bind_addr: String,
391 /// Internal port from `network.internal_port` in the manifest. When `Some`, the runtime
392 /// binds strictly to this port and errors if it is already in use. When `None`, the OS
393 /// assigns a port.
394 pub internal_port: Option<u16>,
395 /// Minimum containment class this session must achieve, already combined across every
396 /// source that asked for one (manifest / workspace config / `--containment`) by taking the
397 /// strongest. Required rather than optional: "nobody declared anything" is
398 /// [`ContainmentClass::Advisory`], an explicit floor every host clears, not an absence.
399 /// `stage_session` refuses to stage when the host cannot meet it.
400 pub declared_containment_floor: murmur_artifact::ContainmentClass,
401 /// The manifest's top-level `exports:` block. `None` — and `Some` with no `files:` — means
402 /// the resource plane is not declared and every request to it is denied.
403 ///
404 /// Carried beside `capability_policy` rather than inside it: an export is a disclosure the
405 /// operator makes, not a capability the guest holds, and nothing derived from this field ever
406 /// reaches the achieved containment class.
407 pub exports: Option<murmur_artifact::Exports>,
408 /// The approval a delegated child was handed by its parent, presented once at registration.
409 ///
410 /// `None` for every top-level launch, which the daemon admits against its own `--spawn-allow`
411 /// instead. Unlike [`StagedSession::spawn_credential`], this exists before staging: it names
412 /// an artifact rather than a session, and the parent already held it when it started this
413 /// process.
414 pub spawn_grant: Option<SpawnApproval>,
415 /// `spend.machine_tokens_per_day` from the operator's effective `config.yaml`. `Some` makes
416 /// staging open the shared daily spend ledger for a `transport: http` session with a driver,
417 /// and refuse when it cannot; `None` — every launch that does not read the operator config —
418 /// keeps no ledger and is not counted.
419 pub machine_tokens_per_day: Option<u64>,
420}
421
422/// The `transport: process` driver this session runs its harness through, compiled and described
423/// once at staging.
424///
425/// Not a tool: it is granted nothing, never enters `tool_components`, and is reached only by the
426/// process runner, which instantiates a fresh [`crate::process_driver::ProcessDriver`] from
427/// `component` per harness run.
428pub struct StagedProcessDriver {
429 pub(crate) name: String,
430 pub(crate) version: String,
431 pub(crate) component: Component,
432 pub(crate) description: crate::process_driver::Description,
433 /// The absolute path of the executable to spawn, already resolved and checked.
434 pub(crate) binary: PathBuf,
435 /// Where [`Self::binary`]'s name came from, for diagnostics and the trace: `inference.command`
436 /// or the process driver's `describe()`.
437 pub(crate) binary_source: String,
438}
439
440pub struct StagedSession {
441 pub session_id: String,
442 pub workdir: PathBuf,
443 /// User project directory (or session dir if no --workdir). WASM tools see this at ".".
444 pub accessible_workdir: std::path::PathBuf,
445 pub(crate) manifest_dir: PathBuf,
446 pub(crate) capsule_name: String,
447 pub(crate) capsule_version: String,
448 pub(crate) capsule_url: String,
449 pub resolved_lock_artifacts: Vec<ResolvedLockArtifact>,
450 pub(crate) installed_artifacts: Vec<InstalledArtifactSummary>,
451 pub(crate) inference: Option<InferenceConfig>,
452 /// Every credential gateway of the session, built at staging from each artifact entry's
453 /// `gateway:` block and the artifact's own `upstream_auth:` declaration. Empty when no entry
454 /// declares one.
455 pub(crate) gateways: crate::credential_gateway::GatewayTable,
456 /// This session's spend account, built at staging from `inference.max_session_tokens` and
457 /// [`StageRequest::machine_tokens_per_day`]. Always present: a session with no ceiling gets a
458 /// meter that admits everything, so every driver call has an admission to open.
459 pub(crate) spend: Arc<crate::spend::SpendMeter>,
460 /// Copied from [`StageRequest::system_prompt_overridden`] — the only record left that the
461 /// prompt in `inference` came from `--system-prompt` and not from the manifest. Passed to
462 /// `TraceWriter::open`, which turns it into `session_start.system_prompt_source`.
463 pub(crate) system_prompt_overridden: bool,
464 pub(crate) context: Option<ContextConfig>,
465 /// Copied from [`StageRequest::context_id`] and already validated. Read by `launch_session`,
466 /// which uses it in place of a freshly minted id for every `task.md` task.
467 pub(crate) context_id: Option<String>,
468 /// Copied from [`StageRequest::resume`] and already checked. Read by `launch_session`, which
469 /// turns it into `session_start.resumed_from` and the agent loop's record-load override.
470 pub(crate) resume: Option<ResumeRequest>,
471 /// Copied from [`StageRequest::forget_session`] and already checked against the transport.
472 /// Read by `launch_session` as a one-shot: the launch's first task carries the forget, and no
473 /// later task of the same launch does.
474 pub(crate) forget_session: bool,
475 pub(crate) engine: Engine,
476 /// `None` for manifest-only agent capsules; `Some` for script capsules with a WASM component.
477 pub(crate) capsule_component: Option<Component>,
478 pub(crate) tool_components: HashMap<String, Component>,
479 /// The process driver, for a `transport: process` session that named one. `None` on every
480 /// other transport.
481 pub(crate) process_driver: Option<Arc<StagedProcessDriver>>,
482 /// Per-artifact narrowing for the tool/driver dispatch path, keyed by artifact name and
483 /// lowered at staging time from the operator's own manifest entries. Holds an entry only
484 /// for artifacts that actually declared a `capabilities:` block — a name absent from this
485 /// map (the overwhelmingly common case) runs on the full capsule ceiling, exactly as
486 /// before per-artifact narrowing existed.
487 pub(crate) artifact_grants: HashMap<String, ToolCapabilityGrant>,
488 pub(crate) hook_components: Vec<StagedHookArtifact>,
489 pub(crate) allowlisted_tools: HashSet<String>,
490 pub(crate) capability_policy: CapabilityPolicy,
491 pub(crate) otel_endpoint: Option<String>,
492 pub(crate) eval_config_json: Option<String>,
493 pub(crate) case_id: Option<String>,
494 pub(crate) dataset_id: Option<String>,
495 /// Resolved lifecycle config (manifest + override already applied).
496 pub(crate) lifecycle: LifecycleConfig,
497 /// How much of each turn's driver request this session's trace keeps (resolved from the
498 /// manifest's `trace:` block, or [`murmur_artifact::TraceCapture::default`] when it has
499 /// none).
500 pub(crate) trace_capture: murmur_artifact::TraceCapture,
501 /// What bounds this workdir's session directories, from the manifest's `trace.retain` block.
502 /// `None` — the overwhelmingly common case — prunes nothing, ever.
503 pub(crate) trace_retain: Option<murmur_artifact::TraceRetainConfig>,
504 /// Address the HTTP server is bound to (copied from StageRequest::bind_addr).
505 pub(crate) bind_addr: String,
506 /// Internal port from the manifest (copied from StageRequest::internal_port).
507 pub(crate) internal_port: Option<u16>,
508 /// Floor this session was staged against (copied from
509 /// StageRequest::declared_containment_floor).
510 pub(crate) declared_containment_floor: murmur_artifact::ContainmentClass,
511 /// The complete effective grant set this session was staged with — the same report
512 /// `mur run --explain-scope --json` prints, built by `containment::scope_report_for_tier`
513 /// from this session's own policy, declared floor and single host probe. Written verbatim to
514 /// `trace.jsonl`'s `session_start` as `effective_grants`, and the source of the
515 /// `containment_declared`/`containment_achieved`/`workdir_exec` fields alongside it, so the
516 /// summary and the full report cannot disagree.
517 ///
518 /// Replaces a former separate `achieved_containment` field: what the host actually provided
519 /// is `scope_report.achieved_containment`, and keeping a second copy beside it only created a
520 /// way for the two to drift.
521 pub(crate) scope_report: crate::containment::ScopeReport,
522 /// The single host reading this session was staged against, taken at the top of
523 /// `runtime::stage_session`. `launch_session` passes it to `sandbox::ShellEnforcement::resolve`
524 /// rather than taking a new one, so the mechanisms installed at launch are the ones
525 /// [`Self::scope_report`] already claims.
526 pub(crate) host_probe: crate::sandbox::HostProbe,
527 /// The declared read-only surface, lowered and validated by `stage_session` — a malformed
528 /// entry refuses the launch there rather than surfacing at the first call. Empty for every
529 /// capsule that declared no `capabilities.filesystem.read_only`.
530 pub(crate) protected_paths: crate::protected_paths::ProtectedPaths,
531 /// What each staged tool's own `input_schema` declared about where its inputs go, lowered by
532 /// `stage_session` from the manifests it wrote into `<workdir>/tools/`. Empty for a capsule
533 /// that declared no `read_only`, which reads no schema at all, and for one whose tools
534 /// annotated nothing.
535 pub(crate) tool_annotations: crate::tool_annotations::ToolAnnotationMap,
536 /// The declared read-only file surface, with its root already checked against this session's
537 /// accessible workdir by `stage_session` — a root that resolves outside it refuses the launch
538 /// rather than being served. `None` means no resource plane.
539 pub(crate) exports_files: Option<murmur_artifact::FileExport>,
540 /// The declared peer-handoff surface, with its root already checked against this session's
541 /// accessible workdir and its `max_ttl` already checked against `lifecycle.after_task` by
542 /// `stage_session`. `None` means no peer plane: nothing mints, and `/resources/peer/` answers
543 /// `no_peer_plane`.
544 pub(crate) exports_peer_files: Option<murmur_artifact::PeerFilesExport>,
545 /// Registry used to resolve this session's artifacts, retained so `manage.pull()` can
546 /// resolve additional artifacts at runtime after staging has completed.
547 pub(crate) registry: Arc<dyn Registry>,
548 /// Keeps this session's epoch ticker running: without it no epoch deadline set on any
549 /// store built from `engine` can ever fire. Held here (rather than detached) so ticking
550 /// stops when the session's `StagedSession` drops. `launch_session` moves the other
551 /// fields out one by one, which leaves this one in place until the session returns.
552 pub(crate) _epoch_ticker: EpochTicker,
553 /// The credential this session presents when it asks `mur-roost` to spawn a capsule.
554 ///
555 /// `None` for every session that cannot delegate, which is nearly all of them. Set by the
556 /// caller between `stage_session` and `launch_session` rather than carried in
557 /// [`StageRequest`]: the credential names a session id, and `stage_session` is what mints one,
558 /// so it cannot exist before staging returns.
559 pub(crate) spawn_credential: Option<SpawnCredential>,
560 /// Copied from [`StageRequest::spawn_grant`]. Read by `launch_session`, which presents it at
561 /// `POST /register` and never again — an approval covers one launch.
562 pub(crate) spawn_grant: Option<SpawnApproval>,
563 /// The handle the spawning capsule's launcher injected, read once per process at staging.
564 ///
565 /// `None` for every capsule nobody delegated. Its `session_id` and `delegation_id` become
566 /// `session_start.spawned_by` and `session_start.delegation_id`; its `report_to` — present
567 /// only for a parent that wants a completion — is what `launch_session`'s reporting guard
568 /// posts to. Read here rather than at each of those two points so one process reads
569 /// `MURMUR_SPAWNER` once and both readers see the same value.
570 pub(crate) spawner: Option<crate::delegation::SpawnerHandle>,
571}
572
573impl StagedSession {
574 /// Hands this session the credential it will present to the spawning daemon.
575 pub fn set_spawn_credential(&mut self, credential: SpawnCredential) {
576 self.spawn_credential = Some(credential);
577 }
578
579 /// The credential this session presents, if it was granted one.
580 pub fn spawn_credential(&self) -> Option<&SpawnCredential> {
581 self.spawn_credential.as_ref()
582 }
583
584 /// The grant set this session was staged with, as written to the trace.
585 pub fn scope_report(&self) -> &crate::containment::ScopeReport {
586 &self.scope_report
587 }
588}
589
590#[derive(Debug, Clone, PartialEq, Eq)]
591pub struct LaunchResult {
592 pub session_id: String,
593 pub workdir: PathBuf,
594}
595
596pub fn capability_policy_from_runtime_manifest(
597 manifest: &murmur_artifact::RuntimeManifest,
598) -> CapabilityPolicy {
599 let caps = manifest.capabilities.as_ref();
600 let network_allow = caps
601 .and_then(|c| c.network.as_ref())
602 .map(|network| network.allow.clone())
603 .unwrap_or_default();
604 // Absent `capabilities.peer_fetch` block means denied, on the same terms — and separately
605 // from `network.allow`, which this deliberately does not read.
606 let peer_fetch_allow = caps
607 .and_then(|c| c.peer_fetch.as_ref())
608 .map(|peer_fetch| peer_fetch.allow.clone())
609 .unwrap_or_default();
610
611 // Absent `capabilities.network` block, or absent key within it, both mean denied.
612 let unix_sockets_allowed = caps
613 .and_then(|c| c.network.as_ref())
614 .is_some_and(|network| network.unix_sockets);
615
616 // Absent `capabilities.plan` block, or `submit: false` within it, both mean denied.
617 let plan_submit = caps
618 .and_then(|c| c.plan.as_ref())
619 .is_some_and(|plan| plan.submit);
620
621 // Recorded, never applied: see `CapabilityPolicy::state_declared`.
622 let state_declared = caps.is_some_and(|c| c.state.is_some());
623 let conversation_declared = caps.is_some_and(|c| c.conversation.is_some());
624
625 let filesystem_scope = caps
626 .and_then(|c| c.filesystem.as_ref())
627 .and_then(|filesystem| filesystem.scope.clone());
628 // Absent `capabilities.filesystem` block, or absent key within it, both mean denied — the same
629 // shape as `unix_sockets_allowed` above, for the same reason: the widening has to be declared.
630 let workdir_exec_allowed = caps
631 .and_then(|c| c.filesystem.as_ref())
632 .is_some_and(|filesystem| filesystem.workdir_exec);
633 // Absent block, absent key and an empty list are one declaration: nothing is protected.
634 let read_only_paths = caps
635 .and_then(|c| c.filesystem.as_ref())
636 .map(|filesystem| filesystem.read_only.clone())
637 .unwrap_or_default();
638
639 let shell_allow = caps
640 .and_then(|c| c.shell.as_ref())
641 .map(|shell| shell.allow.clone())
642 .unwrap_or_default();
643 let shell_strip_env = caps
644 .and_then(|c| c.shell.as_ref())
645 .and_then(|shell| shell.strip_env.clone())
646 .unwrap_or_default();
647 let shell_baseline_env = caps
648 .and_then(|c| c.shell.as_ref())
649 .and_then(|shell| shell.baseline_env.clone())
650 .unwrap_or_default();
651 let shell_interpreter_runtime = caps
652 .and_then(|c| c.shell.as_ref())
653 .map(|shell| shell.interpreter_runtime.clone())
654 .unwrap_or_default();
655 let shell_staged_runtime = caps
656 .and_then(|c| c.shell.as_ref())
657 .map(|shell| shell.staged_runtime.clone())
658 .unwrap_or_default();
659
660 let spawn_allow = caps
661 .and_then(|c| c.spawn.as_ref())
662 .map(|spawn| spawn.allow.clone())
663 .unwrap_or_default();
664
665 let env_allow = caps
666 .and_then(|c| c.env.as_ref())
667 .map(|env| env.allow.clone())
668 .unwrap_or_default();
669
670 let limits = ExecutionLimits::resolve(caps.and_then(|c| c.limits.as_ref()));
671 let declared_deadline_seconds = caps
672 .and_then(|c| c.limits.as_ref())
673 .and_then(|l| l.deadline_seconds);
674
675 let resources = crate::resources::resolve(caps.and_then(|c| c.resources.as_ref()));
676
677 CapabilityPolicy {
678 network_allow,
679 peer_fetch_allow,
680 unix_sockets_allowed,
681 filesystem_scope,
682 workdir_exec_allowed,
683 read_only_paths,
684 shell_allow,
685 spawn_allow,
686 shell_strip_env,
687 shell_baseline_env,
688 shell_interpreter_runtime,
689 shell_staged_runtime,
690 env_allow,
691 limits,
692 resources,
693 containment_floor: caps.and_then(|c| c.containment).unwrap_or_default(),
694 state_declared,
695 conversation_declared,
696 plan_submit,
697 declared_deadline_seconds,
698 }
699}
700
701#[cfg(test)]
702mod tests {
703 use super::*;
704
705 fn policy_for(manifest_yaml: &str) -> CapabilityPolicy {
706 let manifest = murmur_artifact::RuntimeManifest::from_yaml_str(manifest_yaml)
707 .expect("manifest fixture must parse");
708 capability_policy_from_runtime_manifest(&manifest)
709 }
710
711 /// `read_only` reaches the policy in declaration order, trimmed, with empty entries already
712 /// dropped by the manifest parser — and a capsule that declares none gets an empty list, which
713 /// is what makes the dispatch check free for it.
714 #[test]
715 fn read_only_paths_are_carried_through_and_default_to_empty() {
716 let declared = policy_for(
717 r#"
718name: cap
719version: 0.1.0
720capabilities:
721 filesystem:
722 read_only:
723 - tests
724 - " bench/fixtures "
725 - " "
726"#,
727 );
728 assert_eq!(declared.read_only_paths, vec!["tests", "bench/fixtures"]);
729
730 let silent = policy_for(
731 "name: cap
732version: 0.1.0
733",
734 );
735 assert!(silent.read_only_paths.is_empty());
736
737 let no_key = policy_for(
738 r#"
739name: cap
740version: 0.1.0
741capabilities:
742 filesystem:
743 scope: work
744"#,
745 );
746 assert!(no_key.read_only_paths.is_empty());
747 }
748
749 /// A manifest that sets no deadline gives hook calls the hook-specific default while the
750 /// capsule's own guests keep the capsule-wide one — one manifest, two budgets.
751 #[test]
752 fn a_silent_manifest_gives_hook_calls_the_lower_deadline() {
753 let policy = policy_for(
754 r#"
755name: cap
756version: 0.1.0
757artifacts: []
758"#,
759 );
760
761 assert_eq!(
762 policy.limits.deadline_seconds,
763 crate::limits::DEFAULT_DEADLINE_SECONDS
764 );
765 assert_eq!(
766 policy.hook_limits().deadline_seconds,
767 crate::limits::HOOK_DEFAULT_DEADLINE_SECONDS
768 );
769 // The default policy a hand-built fixture gets must agree with the silent-manifest one.
770 assert_eq!(
771 CapabilityPolicy::default().hook_limits().deadline_seconds,
772 crate::limits::HOOK_DEFAULT_DEADLINE_SECONDS
773 );
774 }
775
776 /// An explicit `deadline_seconds` applies to every guest, hooks included — the hook
777 /// default only fills a silence.
778 #[test]
779 fn an_explicit_deadline_applies_to_hook_calls_too() {
780 let policy = policy_for(
781 r#"
782name: cap
783version: 0.1.0
784artifacts: []
785capabilities:
786 limits:
787 deadline_seconds: 120
788"#,
789 );
790
791 assert_eq!(policy.declared_deadline_seconds, Some(120));
792 assert_eq!(policy.limits.deadline_seconds, 120);
793 assert_eq!(policy.hook_limits().deadline_seconds, 120);
794 }
795
796 /// `capabilities.shell.staged_runtime` lowers onto the policy verbatim — binary, source path
797 /// and pin all survive, since the pin is what a human later compares across two hosts.
798 #[test]
799 fn staged_runtime_grants_lower_onto_the_policy() {
800 let policy = policy_for(
801 r#"
802name: cap
803version: 0.1.0
804artifacts: []
805capabilities:
806 containment: sealed
807 shell:
808 allow:
809 - python3
810 staged_runtime:
811 - binary: python3
812 source_path: /opt/testbed/conda/envs/django__django
813 pin: conda-4.10.3/python-3.9.19
814"#,
815 );
816
817 assert_eq!(
818 policy.shell_staged_runtime,
819 vec![murmur_artifact::StagedRuntimeGrant {
820 binary: "python3".to_string(),
821 source_path: "/opt/testbed/conda/envs/django__django".to_string(),
822 pin: "conda-4.10.3/python-3.9.19".to_string(),
823 }]
824 );
825 // The two mechanisms stay separate all the way down: lowering one must not populate the
826 // other.
827 assert!(policy.shell_interpreter_runtime.is_empty());
828 }
829
830 /// A shell block that declares no staged runtime lowers to an empty list, not to a default
831 /// grant of any kind.
832 #[test]
833 fn absent_staged_runtime_lowers_to_empty() {
834 let policy = policy_for(
835 r#"
836name: cap
837version: 0.1.0
838artifacts: []
839capabilities:
840 shell:
841 allow:
842 - bash
843"#,
844 );
845
846 assert!(policy.shell_staged_runtime.is_empty());
847 }
848
849 /// The default every existing manifest gets. Three separate "nothing was declared" shapes —
850 /// no `capabilities:` at all, no `filesystem:` within it, and a `filesystem:` that only sets
851 /// `scope` — must all lower to denied, because each one is a real manifest in the wild and any
852 /// of them silently keeping workdir `Execute` would reopen the rename-to-an-allowed-basename
853 /// bypass this default exists to close.
854 #[test]
855 fn workdir_exec_denied_for_every_shape_of_undeclared() {
856 for manifest_yaml in [
857 r#"
858name: cap
859version: 0.1.0
860artifacts: []
861"#,
862 r#"
863name: cap
864version: 0.1.0
865artifacts: []
866capabilities:
867 network:
868 allow:
869 - https://api.anthropic.com
870"#,
871 r#"
872name: cap
873version: 0.1.0
874artifacts: []
875capabilities:
876 filesystem:
877 scope: workdir
878"#,
879 ] {
880 let policy = policy_for(manifest_yaml);
881 assert!(
882 !policy.workdir_exec_allowed,
883 "an undeclared workdir_exec must lower to denied: {manifest_yaml}"
884 );
885 }
886 }
887
888 #[test]
889 fn workdir_exec_denied_when_declared_false() {
890 let policy = policy_for(
891 r#"
892name: cap
893version: 0.1.0
894artifacts: []
895capabilities:
896 filesystem:
897 workdir_exec: false
898"#,
899 );
900
901 assert!(!policy.workdir_exec_allowed);
902 }
903
904 #[test]
905 fn workdir_exec_allowed_when_declared_true() {
906 let policy = policy_for(
907 r#"
908name: cap
909version: 0.1.0
910artifacts: []
911capabilities:
912 filesystem:
913 scope: workdir
914 workdir_exec: true
915"#,
916 );
917
918 assert!(policy.workdir_exec_allowed);
919 // The two filesystem keys stay independent: lowering one must not disturb the other.
920 assert_eq!(policy.filesystem_scope.as_deref(), Some("workdir"));
921 }
922
923 /// A manifest with no `capabilities:` block at all denies unix sockets — the field must not
924 /// pick up any implicit widening from the "nothing declared" path.
925 #[test]
926 fn unix_sockets_denied_when_capabilities_block_is_absent() {
927 let policy = policy_for(
928 r#"
929name: cap
930version: 0.1.0
931artifacts: []
932"#,
933 );
934
935 assert!(!policy.unix_sockets_allowed);
936 }
937
938 /// A declared `network:` block that never mentions the key still denies.
939 #[test]
940 fn unix_sockets_denied_when_network_block_omits_the_key() {
941 let policy = policy_for(
942 r#"
943name: cap
944version: 0.1.0
945artifacts: []
946capabilities:
947 network:
948 allow:
949 - https://api.anthropic.com
950"#,
951 );
952
953 assert!(!policy.unix_sockets_allowed);
954 assert_eq!(
955 policy.network_allow,
956 vec!["https://api.anthropic.com".to_string()]
957 );
958 }
959
960 #[test]
961 fn unix_sockets_denied_when_declared_false() {
962 let policy = policy_for(
963 r#"
964name: cap
965version: 0.1.0
966artifacts: []
967capabilities:
968 network:
969 unix_sockets: false
970"#,
971 );
972
973 assert!(!policy.unix_sockets_allowed);
974 }
975
976 #[test]
977 fn unix_sockets_allowed_when_declared_true() {
978 let policy = policy_for(
979 r#"
980name: cap
981version: 0.1.0
982artifacts: []
983capabilities:
984 network:
985 unix_sockets: true
986"#,
987 );
988
989 assert!(policy.unix_sockets_allowed);
990 }
991
992 /// The opt-in is independent of the IP allowlist in both directions: a non-empty `allow`
993 /// does not imply unix sockets, and `unix_sockets: true` does not imply any IP destination.
994 #[test]
995 fn unix_sockets_opt_in_is_independent_of_the_ip_allowlist() {
996 let policy = policy_for(
997 r#"
998name: cap
999version: 0.1.0
1000artifacts: []
1001capabilities:
1002 network:
1003 unix_sockets: true
1004 allow:
1005 - https://api.anthropic.com
1006"#,
1007 );
1008
1009 assert!(policy.unix_sockets_allowed);
1010 assert_eq!(
1011 policy.network_allow,
1012 vec!["https://api.anthropic.com".to_string()]
1013 );
1014 }
1015
1016 /// `CapabilityPolicy::default()` is what the sandbox tests and every `..Default::default()`
1017 /// call site start from; it must deny, not inherit whatever a future field ordering implies.
1018 #[test]
1019 fn default_capability_policy_denies_unix_sockets() {
1020 assert!(!CapabilityPolicy::default().unix_sockets_allowed);
1021 }
1022}