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 formation::FormationId,
16 formation_credentials::FormationMember,
17 hooks::ShellDispatchInfo,
18 limits::{EpochTicker, ExecutionLimits},
19 network_policy::{HookCapabilityGrant, ToolCapabilityGrant},
20 spawn_credential::{SpawnApproval, SpawnCredential},
21};
22
23pub(crate) struct DispatchOutcome {
24 pub result: ToolResult,
25 pub shell: Option<ShellDispatchInfo>,
26 /// Set instead of [`Self::shell`] when the command outran its grace period and was demoted
27 /// to the background. The two are never both `Some`: a demoted command has no exit code yet,
28 /// which is exactly why `HookEvent::Shell` does not fire for one.
29 pub detached: Option<crate::detached::DetachedDispatchInfo>,
30 /// True when the dispatch was served by reading a skill.md file rather than running
31 /// a binary or WASM component. Used by the agent loop to write a `skill_call` trace
32 /// event instead of a `tool_call` event.
33 pub is_skill: bool,
34 /// The fence source `result.data` was wrapped under, or `None` when the result carries no
35 /// fence. Set by [`crate::runtime::CapsuleStoreState::dispatch_agent_tool_async`] on the
36 /// same branch that applies the fence, so the label and the wrapping cannot drift: a caller
37 /// reading this never re-derives a source name by parsing the content back.
38 ///
39 /// `Some("tool:<name>")` on every non-skill outcome, `Some("skill:<name>")` on a skill whose
40 /// `murmur.lock` origin is `runtime`, and `None` on an operator-declared skill and on every
41 /// outcome that leaves the private unfenced dispatch — that is,
42 /// [`crate::runtime::CapsuleStoreState::dispatch_submit_plan`]'s route, whose step output is
43 /// fenced later as one field of a report.
44 pub fence_source: Option<String>,
45 /// Set when this dispatch failed in a way that ends the *session* rather than just this
46 /// tool call — today only a `sealed` composed-root construction failure
47 /// ([`crate::shell::ShellExecError::session_fatal`]). `result` still describes the failed
48 /// call so the trace records what was attempted; the agent turn loop then returns this
49 /// error instead of feeding the failure back to the model for another turn.
50 ///
51 /// A dispatch that merely failed leaves this `None`: a capsule reacting to a broken tool
52 /// call is ordinary, whereas a capsule continuing to run after its declared containment
53 /// boundary stopped being establishable is the silent-degradation failure mode the whole
54 /// containment-class mechanism exists to prevent.
55 pub fatal: Option<crate::errors::RuntimeError>,
56 /// The runtime's own words about this call, appended by
57 /// [`crate::runtime::CapsuleStoreState::dispatch_agent_tool_async`] after the fence's closing
58 /// marker, so the model reads them as the runtime's rather than as the tool's data.
59 ///
60 /// `Some` on these outcomes:
61 ///
62 /// * a `call-member` call that started, carrying [`crate::member_call::started_note`];
63 /// * an accepted `end-without-answer`;
64 /// * a pull through `pull-artifact`, and a `remove-artifact` that removed something, each
65 /// saying when the tool list changes;
66 /// * a shell call that finished and whose failure was attributed to a
67 /// `capabilities.resources` limit, carrying [`crate::resources::resource_limit_line`].
68 ///
69 /// `None` on every other outcome. The unfenced route of
70 /// [`crate::runtime::CapsuleStoreState::dispatch_submit_plan`] never reads it.
71 ///
72 /// The session's one deadline notice is not a note: it is appended after the note, by
73 /// [`crate::runtime::CapsuleStoreState::deliver_deadline_notice`], on whichever dispatched
74 /// outcome is first handed back inside the notice margin.
75 pub runtime_note: Option<String>,
76 /// The runtime's own text, unfenced, when it could not run the call: the
77 /// [`crate::shell::ShellExecError`] a native tool or shell dispatch returned, the same text
78 /// `result.data` carries before the fence wraps it. The agent loop records it as the
79 /// `tool_call` trace record's `error`.
80 ///
81 /// `None` whenever the tool ran, including when the tool itself reported failure or exited
82 /// non-zero.
83 pub runtime_error: Option<String>,
84}
85
86impl DispatchOutcome {
87 pub(crate) fn tool(result: ToolResult) -> Self {
88 Self {
89 result,
90 shell: None,
91 detached: None,
92 is_skill: false,
93 fence_source: None,
94 fatal: None,
95 runtime_note: None,
96 runtime_error: None,
97 }
98 }
99
100 pub(crate) fn skill(result: ToolResult) -> Self {
101 Self {
102 result,
103 shell: None,
104 detached: None,
105 is_skill: true,
106 fence_source: None,
107 fatal: None,
108 runtime_note: None,
109 runtime_error: None,
110 }
111 }
112}
113
114#[derive(Debug, Clone, PartialEq, Eq)]
115pub struct ArtifactRequest {
116 pub name: String,
117 pub version: String,
118 pub runtime: ArtifactRuntime,
119 /// Optional local source path (skill artifacts only). When set, the runtime resolves the
120 /// skill's `skill.md` directly from this path and skips registry resolution entirely.
121 /// Relative paths resolve against `StageRequest::manifest_dir`.
122 pub source: Option<String>,
123 /// Per-artifact capability grant copied verbatim from this artifact's entry in the
124 /// capsule operator's own manifest (`murmur_artifact::RuntimeArtifact::capabilities`).
125 /// Consumed for `runtime: hook` (default-deny baseline, `None` = deny everything) and for
126 /// `runtime: tool`/`runtime: driver` (ceiling baseline, `None` = the capsule-wide policy
127 /// unchanged); the manifest parser rejects the key on `runtime: skill`. Never sourced
128 /// from the artifact's own bundled `murmur.yaml`, so an artifact cannot self-grant.
129 pub capabilities: Option<murmur_artifact::Capabilities>,
130 /// What the runtime does when this hook's async job queue is full, copied verbatim from
131 /// `on_overflow:` on this artifact's entry in the capsule operator's own manifest — the
132 /// same operator-only sourcing rule `capabilities` above follows. Consumed for
133 /// `runtime: hook` and inert for every other role (the manifest parser rejects the key
134 /// there); inert too for a hook that turns out to be `execution_mode: blocking`, which is
135 /// never queued.
136 pub on_overflow: murmur_artifact::HookOverflowPolicy,
137 /// This artifact's operator-declared `config:` block, copied verbatim from its entry in the
138 /// capsule operator's own manifest (`murmur_artifact::RuntimeArtifact::config`) — the same
139 /// operator-only sourcing rule `capabilities` above follows, so an artifact pulled from a
140 /// registry cannot configure itself.
141 ///
142 /// Lowered to compact JSON at staging by [`crate::artifact_config::lower_artifact_config`] and
143 /// delivered to this artifact alone as [`crate::artifact_config::ARTIFACT_CONFIG_ENV`]. `None`
144 /// means the variable is absent from the guest environment. Consumed for `runtime: hook`,
145 /// `runtime: tool` and `runtime: driver`; the manifest parser rejects the key on
146 /// `runtime: skill`, and a native tool warns `W-SEC-015` because it reads no per-artifact
147 /// environment at all.
148 pub config: Option<serde_norway::Value>,
149 /// This artifact's operator-declared `gateway:` block, copied verbatim from its entry in the
150 /// capsule operator's own manifest (`murmur_artifact::RuntimeArtifact::gateway`) — the same
151 /// operator-only sourcing rule `capabilities` above follows. Built into the artifact's
152 /// credential gateway at staging, which is also where the artifact's own `upstream_auth:`
153 /// declaration is read; an artifact without this block is never asked for one.
154 pub gateway: Option<murmur_artifact::ArtifactGateway>,
155}
156
157/// What `murmur.lock` pins one artifact at, already narrowed to this host.
158#[derive(Debug, Clone, PartialEq, Eq)]
159pub struct LockExpectation {
160 pub name: String,
161 pub resolved_version: String,
162 /// The hash the resolved payload must have. Selected by the caller from the lock entry's
163 /// `sha256` for the platform staging runs on, so it is whichever of `any` or
164 /// `platforms.<platform>` applies to this host.
165 pub sha256: String,
166 /// Who wrote the lock entry. A [`murmur_artifact::LockOrigin::Runtime`] pin stages only as a
167 /// tool or skill without a `gateway:` block; staging refuses anything else with
168 /// [`crate::RuntimeError::RuntimeOriginNotDeclarable`].
169 pub origin: murmur_artifact::LockOrigin,
170}
171
172/// What staging actually resolved, for the caller to write back into `murmur.lock`.
173#[derive(Debug, Clone, PartialEq, Eq)]
174pub struct ResolvedLockArtifact {
175 pub name: String,
176 pub resolved_version: String,
177 /// Hash of the payload that was resolved on this host.
178 pub sha256: String,
179 /// Which platform the resolved payload is filed under, `None` for a platform-independent
180 /// one. Decides whether the lock entry pins `sha256.any` or `sha256.platforms.<platform>`.
181 pub platform: Option<String>,
182}
183
184#[derive(Debug, Clone, PartialEq, Eq, Default)]
185pub struct CapabilityPolicy {
186 pub network_allow: Vec<String>,
187 /// From `capabilities.peer_fetch.allow`: the peers this capsule may redeem a peer-file handle
188 /// against. Empty by default, which is deny — and which is also why no `fetch-peer-file` tool
189 /// manifest is written and the capsule's model never sees the tool exists.
190 ///
191 /// A separate list from [`Self::network_allow`], enforced with the same rule matcher but
192 /// never merged into it. Ingesting a peer's bytes lands a file in this capsule's own workdir,
193 /// so it is a prompt-injection surface and deserves its own operator control; a destination
194 /// in `network.allow` is not redeemable unless it also appears here, and declaring it here
195 /// does not widen `network.allow`.
196 pub peer_fetch_allow: Vec<String>,
197 /// From `capabilities.network.unix_sockets`. `false` by default, which makes the shell
198 /// subprocess tree's `socket(AF_UNIX, ...)` calls fail with `EACCES` at the kernel level on
199 /// both Linux tiers — `network_allow` above governs IP destinations only and would otherwise
200 /// leave a local daemon socket (`/var/run/docker.sock` — host root) completely unmediated.
201 /// `true` simply omits that seccomp rule; it is capsule-wide, not per-socket-path. See
202 /// `sandbox::denied_socket_domains`.
203 pub unix_sockets_allowed: bool,
204 pub filesystem_scope: Option<String>,
205 /// From `capabilities.filesystem.workdir_exec`. `false` by default, which withholds the
206 /// Landlock `Execute` right from the workdir's own `PathBeneath` rule — see
207 /// `sandbox::linux_enforce::workdir_access_rights`. With it withheld the kernel refuses to exec
208 /// anything under the workdir on the resolved path itself, so `shell_allow` is enforced
209 /// completely: a binary copied or compiled into the workdir cannot run under *any* name.
210 ///
211 /// `true` gives the right back for compile-and-run workflows and accepts, explicitly, that
212 /// `shell_allow` is then unenforceable for anything inside the workdir. That is why this field
213 /// is the one *grant* in this struct that also lowers the session's achieved containment class
214 /// (see `crate::containment::achieved_containment_class`) — nothing else here can.
215 pub workdir_exec_allowed: bool,
216 /// From `capabilities.filesystem.read_only`: workdir subtrees the capsule may read but must
217 /// not write, in the same vocabulary [`Self::filesystem_scope`] uses. Empty by default, which
218 /// is the whole workdir writable.
219 ///
220 /// Carried here as the declared strings and lowered once at staging into
221 /// `crate::protected_paths::ProtectedPaths`, which is what the dispatch check consults — so a
222 /// malformed entry refuses the launch rather than one call at a time. Unlike every other field
223 /// here this one grants nothing: it can only subtract from what the workdir preopen already
224 /// allows.
225 pub read_only_paths: Vec<String>,
226 pub shell_allow: Vec<String>,
227 pub spawn_allow: Vec<String>,
228 pub shell_strip_env: Vec<String>,
229 pub shell_baseline_env: Vec<String>,
230 /// Typed interpreter-runtime grants from `capabilities.shell.interpreter_runtime`: each
231 /// names an already-allowlisted binary and the exact host directories outside the workdir
232 /// its import machinery needs, with a per-directory `list_dir` enumerability flag. Empty
233 /// unless the manifest declares them. Consumed on `KernelFull` to add narrow Landlock
234 /// grants (see `sandbox::resolve_interpreter_runtime_grants`) and to fire `W-SEC-009`.
235 pub shell_interpreter_runtime: Vec<murmur_artifact::InterpreterRuntimeGrant>,
236 /// Typed staged-runtime grants from `capabilities.shell.staged_runtime`: each names an
237 /// already-allowlisted binary, the absolute host path of a pinned runtime tree, and the `pin`
238 /// identifying that tree's build. Empty unless the manifest declares them.
239 ///
240 /// Consumed at staging to enforce that declaring one requires an effective `sealed` floor (see
241 /// `crate::staged_runtime::check_staged_runtime_floor`) and to render the `staged runtime`
242 /// section of `--explain-scope`. Per-binary mutually exclusive with
243 /// [`Self::shell_interpreter_runtime`], which the manifest parser already guarantees.
244 pub shell_staged_runtime: Vec<murmur_artifact::StagedRuntimeGrant>,
245 /// The runtime directories of each `shell_allow` binary the runtime recognises as an
246 /// interpreter, as that interpreter reported them (see [`crate::interpreter_runtime`]). Not a
247 /// manifest field: empty from [`capability_policy_from_runtime_manifest`], filled once by
248 /// `stage_session` and by `mur run --explain-scope` through
249 /// [`crate::interpreter_runtime::derive_interpreter_runtimes`]. Each becomes a readable,
250 /// non-executable Landlock grant (`sandbox::resolve_derived_interpreter_grants`) and, on
251 /// `sealed`, an optional composed-root bind.
252 pub shell_derived_interpreter_runtime:
253 Vec<crate::interpreter_runtime::DerivedInterpreterRuntime>,
254 /// The host's TLS trust store, as staging found it for a capsule with a non-empty
255 /// `shell_allow` and a non-empty `network_allow` (see [`crate::trust_store`]). Not a manifest
256 /// field: empty from [`capability_policy_from_runtime_manifest`], filled once by
257 /// `stage_session` and by `mur run --explain-scope` through
258 /// [`crate::trust_store::derive_trust_store`]. Each becomes a readable, non-executable
259 /// Landlock grant (`sandbox::resolve_trust_store_grants`), listable for a directory.
260 pub shell_trust_store: Vec<crate::trust_store::TrustStorePath>,
261 /// Host env var names a WASM guest may observe, from `capabilities.env.allow`. Empty by
262 /// default: a guest sees only the runtime's own `MURMUR_*` injection unless the manifest
263 /// names a variable here. `shell_strip_env` and the credential patterns still apply on
264 /// top, so a credential-shaped name declared here is stripped anyway.
265 pub env_allow: Vec<String>,
266 /// Execution limits applied to every guest store in the session. Always fully
267 /// resolved — `capability_policy_from_runtime_manifest` substitutes a default for
268 /// each field `capabilities.limits` omits, so there is no "unset" state here and a
269 /// silent manifest yields [`ExecutionLimits::default`] rather than no limits.
270 pub limits: ExecutionLimits,
271 /// OS-level bounds applied to every *native subprocess* the session spawns — rlimits, the
272 /// Linux cgroup v2 scope, and the workdir-size ceiling. Resolved on exactly the same terms
273 /// as `limits` above (each omitted `capabilities.resources` field replaced by its default,
274 /// no "unset" state), but a completely different subject: `limits` bounds a WASM guest
275 /// inside its wasmtime store, this bounds the processes the host forks. See
276 /// [`crate::resources`].
277 pub resources: crate::resources::HostResourceLimits,
278 /// The containment class this capsule asked for, from `capabilities.containment`.
279 ///
280 /// Unlike every other field here it is a *requirement*, not a grant — and unlike
281 /// `StageRequest::declared_containment_floor` it is only the manifest's own vote, before the
282 /// workspace config and `--containment` are folded in. `stage_session` overwrites it with the
283 /// combined floor as soon as that is known, so anything reading it off a `StagedSession` sees
284 /// the effective value; the manifest-only value exists so a `CapabilityPolicy` built straight
285 /// from a manifest is still self-consistent.
286 ///
287 /// Read by `sandbox::ShellEnforcement::resolve` to decide whether a sealed-capable host
288 /// actually installs a composed root for this session — a capsule that declared `scoped` must
289 /// keep getting `scoped`'s mechanism. See `sandbox::applied_tier`.
290 pub containment_floor: murmur_artifact::ContainmentClass,
291 /// Whether the capsule's own top-level `capabilities.state` block was declared.
292 ///
293 /// Not a grant, and the only field here that is not: a durable state store is granted per
294 /// *artifact*, so a capsule-wide declaration reaches nothing at all. It is recorded so
295 /// `stage_session` can say that once (`W-SEC-014`) instead of leaving an operator to discover
296 /// an empty directory, and it is read by nothing else.
297 pub state_declared: bool,
298 /// Recorded for the same reason [`Self::state_declared`] is, and applied just as little: the
299 /// `murmur:conversation/read` grant is per artifact, so a capsule-wide declaration reaches
300 /// nothing and `stage_session` says so once (`W-SEC-016`).
301 pub conversation_declared: bool,
302 /// From `capabilities.plan.submit`: whether this capsule's model is offered the
303 /// runtime-provided `submit-plan` tool. `false` by default, which is deny — and which is also
304 /// why no `submit-plan` tool manifest is written and the capsule's model never sees the tool
305 /// exists.
306 ///
307 /// It grants no reach of its own: every step of a submitted plan runs through this same
308 /// session's tool dispatch, its shell allowlist and its spawn grant, so a plan reaches exactly
309 /// what the model could already call one call at a time.
310 pub plan_submit: bool,
311 /// From `capabilities.install.skill`: the entries naming which skill artifacts a pull
312 /// (`manage.pull` or the `pull-artifact` tool) may install into this session, verbatim and in
313 /// declaration order. Empty by default, which is deny. With [`Self::install_tool`] also
314 /// empty, every pull is refused.
315 ///
316 /// A pulled skill gains no grant of its own; see `install_grant`.
317 pub install_skill: Vec<String>,
318 /// From `capabilities.install.tool`: the entries naming which tool artifacts, WASM or native,
319 /// a pull may install, verbatim and in declaration order. Empty by default, which is deny.
320 ///
321 /// A pulled tool joins the session's dispatch allowlist and dispatches on this policy's own
322 /// ceiling; it gains no grant of its own.
323 pub install_tool: Vec<String>,
324 /// From `capabilities.install.from`: `true` for `[stores, sources]`, under which a pull the
325 /// stores cannot answer is fetched from the scope's GitHub source. `false` by default.
326 pub install_from_sources: bool,
327 /// `capabilities.limits.deadline_seconds` exactly as the manifest declared it — `None`
328 /// when it declared nothing. Retained undefaulted alongside the fully-resolved `limits`
329 /// above purely so hook calls can apply their own, lower default without mistaking an
330 /// explicit `600` for silence. See [`Self::hook_limits`].
331 pub declared_deadline_seconds: Option<u64>,
332}
333
334impl CapabilityPolicy {
335 /// Execution limits for hook lifecycle calls — [`Self::limits`] with the hook-specific
336 /// deadline default substituted when the manifest declared none.
337 ///
338 /// A hook call is one bounded piece of work, so it does not inherit the capsule-wide
339 /// ten-minute budget that exists for a capsule's whole `run`: a wedged hook would
340 /// otherwise stall the session for most of that, every event. An explicitly declared
341 /// `deadline_seconds` still wins, for hooks as for every other guest.
342 #[must_use]
343 pub fn hook_limits(&self) -> ExecutionLimits {
344 self.limits.for_hook_calls(self.declared_deadline_seconds)
345 }
346}
347
348#[derive(Debug, Clone, PartialEq, Eq)]
349pub struct InstalledArtifactSummary {
350 pub name: String,
351 pub version: String,
352 pub runtime: ArtifactRuntime,
353 pub implementation: Option<ArtifactImplementation>,
354 /// Who pinned the artifact: its `murmur.lock` entry's origin, both for an artifact staged
355 /// from the lock and for one `manage.pull()` installed mid-session. An artifact staged
356 /// without a lock, and a `source:` skill, are [`murmur_artifact::LockOrigin::Operator`].
357 pub origin: murmur_artifact::LockOrigin,
358}
359
360#[derive(Clone)]
361pub(crate) struct StagedHookArtifact {
362 pub name: String,
363 /// The name this hook's `artifacts:` entry declares, scope included: what a message about a
364 /// setting on that entry names it by.
365 pub entry: String,
366 pub version: String,
367 pub component: Component,
368 /// Behavioral contract (binding/execution_mode/commit_policy) from the hook artifact's
369 /// *own* bundled manifest. Carries no capability information — see `grant`.
370 pub config: HookConfig,
371 /// What this hook is allowed to do, lowered and validated at staging time from the
372 /// operator's own manifest entry. Applied identically by all three hook instantiation
373 /// paths in `hooks.rs`; `HookCapabilityGrant::default()` is full default-deny.
374 pub grant: HookCapabilityGrant,
375 /// What to do when this hook's async job queue is full — from the operator's manifest
376 /// entry (via [`ArtifactRequest::on_overflow`]), on the same terms as `grant` and never
377 /// from `config`. Read only when `config.execution_mode` is `Async`; a blocking hook has
378 /// no queue to overflow.
379 pub on_overflow: murmur_artifact::HookOverflowPolicy,
380 /// This hook's own credential gateway, when its operator entry declares `gateway:`. Attached
381 /// to every store the hook is instantiated on, `on-stage` included. Filled once the session's
382 /// gateways are staged, before any hook runs.
383 pub gateway: Option<Arc<crate::credential_gateway::CredentialGateway>>,
384}
385
386/// How `mur run --resume` puts the loaded conversation in front of the model.
387///
388/// `Full` is often the cheaper of the two: a verbatim reload can land on the provider's own
389/// prompt cache, while compaction changes the prefix from the first altered token, guarantees a
390/// cache miss, and costs an inference call to produce the summary. `Compact` is the answer when
391/// the conversation would not fit the context window at all.
392#[derive(Debug, Clone, Copy, Default, PartialEq, Eq)]
393pub enum ResumeMode {
394 #[default]
395 Full,
396 Compact,
397}
398
399/// One operator act: continue *this* conversation, whatever the capsule's own
400/// `lifecycle.conversation` policy says about carrying one between tasks.
401///
402/// Sugar over [`StageRequest::context_id`], never a second load path: the CLI resolves a session
403/// address to the context id that session ran under and fills both this and `context_id`. What
404/// this adds is the override — the record is loaded even under `lifecycle.conversation:
405/// stateless` — and the provenance the trace records.
406#[derive(Debug, Clone)]
407pub struct ResumeRequest {
408 /// The prior session's id, recorded verbatim as `session_start.resumed_from`.
409 pub from_session: String,
410 pub mode: ResumeMode,
411}
412
413#[derive(Debug, Clone)]
414pub struct StageRequest {
415 pub manifest_dir: PathBuf,
416 pub capsule_name: String,
417 pub capsule_version: String,
418 /// The full `<scope>/<name>` the capsule was resolved by, for one launched by name
419 /// (`mur run --capsule`). The session registers with mur-roost and reports its completion
420 /// under it. `None` for a capsule run from a project directory, which is `local/<capsule_name>`.
421 pub capsule_artifact: Option<String>,
422 pub capsule_component_bytes: Vec<u8>,
423 pub artifacts: Vec<ArtifactRequest>,
424 pub allowlisted_tools: HashSet<String>,
425 pub lock_expectations: Option<Vec<LockExpectation>>,
426 pub capability_policy: CapabilityPolicy,
427 pub inference: Option<InferenceConfig>,
428 /// The global config file a `${NAME}` in an artifact's `gateway.api_key` is looked up in first, as
429 /// `credentials.NAME`, and re-read from while the session runs. `None` looks in the launching
430 /// environment only.
431 pub credentials_file: Option<PathBuf>,
432 /// Whether `inference` above had its system-prompt declaration replaced by the caller
433 /// (`mur run --system-prompt`) rather than read from the manifest. Purely a provenance
434 /// signal for the trace: the override itself is already applied to `inference`, which by
435 /// this point is indistinguishable from a manifest that declared the same prompt inline.
436 /// `true` even when the override cleared the prompt to nothing.
437 pub system_prompt_overridden: bool,
438 pub context: Option<ContextConfig>,
439 /// Context id every `task.md` task of this launch runs under, from `mur run --context`.
440 /// `None` mints a fresh one per task. Validated by `stage_session` as one path segment: it is
441 /// a directory name in the conversation record path.
442 pub context_id: Option<String>,
443 /// `mur run --resume <session>`, already resolved: `context_id` above holds the context that
444 /// session ran under, and this carries the session's own id and the mode. `None` is every
445 /// ordinary launch. `stage_session` refuses a resume whose context kept no record on disk,
446 /// and a `--resume-mode compact` with no hook bound to `on-compaction`.
447 pub resume: Option<ResumeRequest>,
448 /// `mur run --forget-session`: drop the harness session `context_id` names before this
449 /// launch's first task, so that task starts a new conversation under the same context id.
450 /// `false` on every ordinary launch. `stage_session` refuses it for a capsule whose transport
451 /// keeps no harness session.
452 pub forget_session: bool,
453 /// OTLP/HTTP endpoint for span export; None = no external OTel emission.
454 pub otel_endpoint: Option<String>,
455 /// JSON-serialized EvalConfig injected into hook WASI env as MURMUR_EVAL_CONFIG.
456 pub eval_config_json: Option<String>,
457 /// Dataset case identifier injected as MURMUR_CASE_ID when running a dataset.
458 pub case_id: Option<String>,
459 /// Dataset identifier injected as MURMUR_DATASET_ID when running a dataset.
460 pub dataset_id: Option<String>,
461 /// Lifecycle config from the manifest (None = use LifecycleConfig::default()).
462 pub lifecycle: Option<LifecycleConfig>,
463 /// Optional override applied on top of lifecycle (e.g. from CLI flags or mur-roost).
464 pub lifecycle_override: Option<LifecycleOverride>,
465 /// Trace config from the manifest (None = defaults apply: input captured, output not).
466 pub trace: Option<murmur_artifact::TraceConfig>,
467 /// The operator's directory to use as the capsule's work folder, `mur run --workdir`; the
468 /// runtime creates nothing in it. `None` uses `<manifest_dir>/workdir/<session_id>`.
469 pub workdir: Option<std::path::PathBuf>,
470 /// Address to bind the HTTP server on. Defaults to "127.0.0.1" for local runs.
471 /// Use "0.0.0.0" when deploying to a VM so the capsule is reachable externally.
472 pub bind_addr: String,
473 /// Internal port from `network.internal_port` in the manifest. When `Some`, the runtime
474 /// binds strictly to this port and errors if it is already in use. When `None`, the OS
475 /// assigns a port.
476 pub internal_port: Option<u16>,
477 /// Minimum containment class this session must achieve, already combined across every
478 /// source that asked for one (manifest / workspace config / `--containment`) by taking the
479 /// strongest. Required rather than optional: "nobody declared anything" is
480 /// [`ContainmentClass::Advisory`], an explicit floor every host clears, not an absence.
481 /// `stage_session` refuses to stage when the host cannot meet it.
482 pub declared_containment_floor: murmur_artifact::ContainmentClass,
483 /// The manifest's top-level `exports:` block. `None` — and `Some` with no `files:` — means
484 /// the resource plane is not declared and every request to it is denied.
485 ///
486 /// Carried beside `capability_policy` rather than inside it: an export is a disclosure the
487 /// operator makes, not a capability the guest holds, and nothing derived from this field ever
488 /// reaches the achieved containment class.
489 pub exports: Option<murmur_artifact::Exports>,
490 /// The manifest's top-level `control:` block. `None` means the session has no control
491 /// surface: no token is minted and every request under `/control` is answered `404`.
492 ///
493 /// Beside `exports` and on the same terms: what a controller may change is not a capability
494 /// the guest holds, and nothing derived from it reaches the containment class.
495 pub control: Option<murmur_artifact::ControlConfig>,
496 /// The manifest's `network.authentication` block. `None` is a public door, which answers every
497 /// caller that reaches the port; `Some` makes staging mint this session's door key and tokens.
498 pub door_authentication: Option<murmur_artifact::NetworkAuthentication>,
499 /// The approval a delegated child was handed by its parent, presented once at registration.
500 ///
501 /// `None` for every top-level launch, which the daemon admits against its own `--spawn-allow`
502 /// instead. Unlike [`StagedSession::spawn_credential`], this exists before staging: it names
503 /// an artifact rather than a session, and the parent already held it when it started this
504 /// process.
505 pub spawn_grant: Option<SpawnApproval>,
506 /// `spend.machine_tokens_per_day` from the operator's effective `config.yaml`. `Some` makes
507 /// staging open the shared daily spend ledger for a `transport: http` session with a driver,
508 /// and refuse when it cannot; `None` — every launch that does not read the operator config —
509 /// keeps no ledger and is not counted.
510 pub machine_tokens_per_day: Option<u64>,
511 /// The formation this session is a member of.
512 ///
513 /// `None` is the ordinary case: a session nobody placed in a formation, including a
514 /// standalone capsule that delegates and every child it spawns. `mur run` sets it from
515 /// [`crate::formation::FormationId::from_env`], which reads what the launcher of a formation's
516 /// members, or a member's runtime for its child, put in the environment.
517 pub formation_id: Option<FormationId>,
518 /// What this session's formation launcher handed it: its roster name, what its door checks a
519 /// formation token with, and the tokens and addresses of the members it may call.
520 ///
521 /// `Some` only for a session `mur run --roster` launched as a member: `mur run` reads it from
522 /// the channel [`crate::formation_credentials::FormationMember::from_env`] names, and refuses
523 /// the launch when that cannot be read. The door verifies formation tokens with it, every WASM
524 /// guest is handed its callees' names, and the runtime's egress resolves and authorizes calls
525 /// to them. No native process and no delegated child is handed any of it.
526 pub formation_member: Option<Arc<FormationMember>>,
527 /// Where a pull the stores cannot answer is fetched from, for a capsule granted
528 /// `capabilities.install.from: [stores, sources]`. `None` fetches from no source: every pull
529 /// is answered by the stores alone.
530 pub source_supply: Option<Arc<dyn crate::artifact_supply::SourceSupply>>,
531 /// The `(st_dev, st_ino)` the accessible workdir must have, compared with `fstat` of the
532 /// directory `stage_session` opens at [`Self::workdir`] before anything is written beneath
533 /// it; a mismatch refuses the launch with [`crate::RuntimeError::WorkdirIdentityMismatch`].
534 /// Set by a delegating parent through `mur run --workdir-identity`, for the child directory it
535 /// created. `None` checks nothing.
536 pub expected_workdir_identity: Option<(u64, u64)>,
537 /// The hard limit `mur run --deadline` set, counted from the moment `mur run` started. Only
538 /// [`crate::launch_session_handling_sigterm`] on an agent capsule enforces one: reached, it
539 /// ends the session through the termination routine a first `SIGTERM` begins. Every other
540 /// launch of a session staged with one is refused with
541 /// [`crate::RuntimeError::SessionDeadlineUnenforceable`]. `None` sets no limit.
542 pub session_deadline: Option<crate::session_status::SessionDeadline>,
543 /// The manifest file this launch read, copied into the runtime directory under its own file
544 /// name: `--manifest`, or the project's `murmur.yaml`. `None` for a launch that read none from
545 /// disk — an installed capsule run by name — which copies nothing.
546 pub manifest_file: Option<PathBuf>,
547}
548
549/// The `transport: process` driver this session runs its harness through, compiled and described
550/// once at staging.
551///
552/// Not a tool: it is granted nothing, never enters `tool_components`, and is reached only by the
553/// process runner, which instantiates a fresh [`crate::process_driver::ProcessDriver`] from
554/// `component` per harness run.
555pub struct StagedProcessDriver {
556 pub(crate) name: String,
557 pub(crate) version: String,
558 pub(crate) component: Component,
559 pub(crate) description: crate::process_driver::Description,
560 /// The absolute path of the executable to spawn, already resolved and checked.
561 pub(crate) binary: PathBuf,
562 /// Where [`Self::binary`]'s name came from, for diagnostics and the trace: `inference.command`
563 /// or the process driver's `describe()`.
564 pub(crate) binary_source: String,
565}
566
567pub struct StagedSession {
568 pub session_id: String,
569 /// The work folder: the `--workdir` directory, or `<manifest-dir>/workdir/<session-id>`
570 /// without one. Every command's current directory and every WASM tool's `"."`. The runtime
571 /// creates nothing beneath it, and reads beneath it only through [`Self::accessible_root`].
572 pub accessible_workdir: std::path::PathBuf,
573 /// [`Self::accessible_workdir`], held open.
574 pub accessible_root: crate::workdir_fs::WorkdirRoot,
575 /// `<manifest-dir>/workdir`, which holds the work folder the runtime created for a launch
576 /// without `--workdir`, and those of earlier such launches that `trace.retain` prunes. `None`
577 /// under `--workdir`, whose folder is the operator's and is never pruned.
578 pub(crate) created_workdirs: Option<PathBuf>,
579 /// The session's runtime directory, `~/.murmur/sessions/<session-id>/runtime`: the value of
580 /// `$MURMUR_RUNTIME_DIR`, holding `MURMUR.md`, `task.md`, `murmur.yaml`, `tools/` and
581 /// `plans/`. Removed when the session ends; the runtime reads and writes beneath it only
582 /// through [`Self::runtime_root`].
583 pub runtime_dir: PathBuf,
584 /// [`Self::runtime_dir`], held open, sharing [`Self::accessible_root`]'s refusal latch.
585 pub runtime_root: crate::workdir_fs::WorkdirRoot,
586 /// The three held directories beneath the session's runtime directory, with the records
587 /// directory beside them.
588 pub(crate) runtime_dirs: crate::runtime_dir::RuntimeDirRoots,
589 /// The session's records directory, `~/.murmur/records/<workdir root>/<session-id>`: the
590 /// value of `$MURMUR_RECORDS_DIR`, holding `trace.jsonl`, `blobs/`, `logs/`,
591 /// `completion.json`, `hooks/`, `out/`, `tmp/`, `home/` and `children/`. Kept after the
592 /// session ends; the runtime reads and writes beneath it only through
593 /// [`Self::records_root`].
594 pub records_dir: PathBuf,
595 /// [`Self::records_dir`], held open, sharing [`Self::accessible_root`]'s refusal latch.
596 pub records_root: crate::workdir_fs::WorkdirRoot,
597 /// The file name the launched manifest was copied into the runtime directory under, which
598 /// the `[Capsule]` notice and `MURMUR.md` name. `None` when the launch read no manifest file.
599 pub(crate) manifest_file_name: Option<String>,
600 /// What `MURMUR.md` is rendered from, for an agent session; `None` for a script capsule,
601 /// which gets no `MURMUR.md`. Its identity's capsule URL is empty until launch.
602 pub(crate) murmur_md: Option<crate::murmur_md::MurmurMdInputs>,
603 /// Owns the runtime directory: dropping it removes the directory, so it is held until the
604 /// session ends — `launch` takes it and drops it on return, and a `StagedSession` dropped
605 /// unlaunched drops it with itself.
606 pub(crate) runtime_dir_guard: Option<crate::runtime_dir::RuntimeDir>,
607 pub(crate) manifest_dir: PathBuf,
608 pub(crate) capsule_name: String,
609 pub(crate) capsule_version: String,
610 /// [`StageRequest::capsule_artifact`], or `local/<capsule_name>` for a project run.
611 pub(crate) capsule_artifact: String,
612 pub(crate) capsule_url: String,
613 pub resolved_lock_artifacts: Vec<ResolvedLockArtifact>,
614 pub(crate) installed_artifacts: Vec<InstalledArtifactSummary>,
615 /// Where staging took the facts about each allowlisted binary it examined, for
616 /// `session_start.host_facts`.
617 pub(crate) host_facts: Vec<crate::host_facts::HostFactsUse>,
618 /// Each staged artifact's full name, by the tool name the session knows it by.
619 pub(crate) artifact_full_names: std::collections::HashMap<String, String>,
620 pub(crate) inference: Option<InferenceConfig>,
621 /// Every credential gateway of the session, built at staging from each artifact entry's
622 /// `gateway:` block and the artifact's own `upstream_auth:` declaration. Empty when no entry
623 /// declares one.
624 pub(crate) gateways: crate::credential_gateway::GatewayTable,
625 /// This session's spend account, built at staging from `inference.max_session_tokens` and
626 /// [`StageRequest::machine_tokens_per_day`]. Always present: a session with no ceiling gets a
627 /// meter that admits everything, so every driver call has an admission to open.
628 pub(crate) spend: Arc<crate::spend::SpendMeter>,
629 /// What a controller has changed on this session, built at staging from
630 /// [`StageRequest::control`]. `None` for a capsule with no `control:` block, which mints no
631 /// control token and answers `404` under `/control`.
632 pub(crate) control: Option<Arc<crate::control_plane::ControlState>>,
633 /// The session's driver choices, the primary first, each with the credential staging left
634 /// it and, under `transport: http`, its model limits. `None` for a capsule with no inference
635 /// block.
636 pub(crate) driver_choices: Option<crate::driver_choice::DriverChoices>,
637 /// The output cap a hook's `run-inference` sends for each model it may call.
638 pub(crate) hook_output_caps: crate::model_limits::HookOutputCaps,
639 /// The `MURMUR_INFERENCE_DRIVER_CONFIG` a hook's `run-inference` driver sees while the hook
640 /// serves `on-compaction`: `inference.driver.config` with `inference.compaction.driver_config`
641 /// laid over it. `None` when the manifest sets no overlay, so the call sees the session's.
642 pub(crate) compaction_driver_config: Option<String>,
643 /// Copied from [`StageRequest::door_authentication`], which the cards are built from.
644 pub(crate) door_authentication: Option<murmur_artifact::NetworkAuthentication>,
645 /// This session's door key and tokens, minted at staging when
646 /// [`Self::door_authentication`] is declared. The key never leaves this value.
647 pub(crate) door_auth: Option<Arc<crate::door_auth::DoorAuth>>,
648 /// Copied from [`StageRequest::system_prompt_overridden`] — the only record left that the
649 /// prompt in `inference` came from `--system-prompt` and not from the manifest. Passed to
650 /// `TraceWriter::open`, which turns it into `session_start.system_prompt_source`.
651 pub(crate) system_prompt_overridden: bool,
652 pub(crate) context: Option<ContextConfig>,
653 /// Copied from [`StageRequest::context_id`] and already validated. Read by `launch_session`,
654 /// which uses it in place of a freshly minted id for every `task.md` task.
655 pub(crate) context_id: Option<String>,
656 /// Copied from [`StageRequest::resume`] and already checked. Read by `launch_session`, which
657 /// turns it into `session_start.resumed_from` and the agent loop's record-load override.
658 pub(crate) resume: Option<ResumeRequest>,
659 /// Copied from [`StageRequest::forget_session`] and already checked against the transport.
660 /// Read by `launch_session` as a one-shot: the launch's first task carries the forget, and no
661 /// later task of the same launch does.
662 pub(crate) forget_session: bool,
663 pub(crate) engine: Engine,
664 /// The compiled-form cache handle built for [`Self::engine`] and the capsule's writable root.
665 /// A mid-session pull compiles through it, so a pulled component is stored and
666 /// looked up under the same name a staged one is.
667 pub(crate) compiled_forms: crate::compiled_forms::CompiledForms,
668 /// `None` for manifest-only agent capsules; `Some` for script capsules with a WASM component.
669 pub(crate) capsule_component: Option<Component>,
670 pub(crate) tool_components: HashMap<String, Component>,
671 /// What each native tool installed at staging needs to start, keyed by tool name. Derived
672 /// from the image the runtime wrote; see [`crate::sandbox::NativeEntrypoint`].
673 pub(crate) native_entrypoints: HashMap<String, crate::sandbox::NativeEntrypoint>,
674 /// The process driver, for a `transport: process` session that named one. `None` on every
675 /// other transport.
676 pub(crate) process_driver: Option<Arc<StagedProcessDriver>>,
677 /// Per-artifact narrowing for the tool/driver dispatch path, keyed by artifact name and
678 /// lowered at staging time from the operator's own manifest entries. Holds an entry only
679 /// for artifacts that actually declared a `capabilities:` block — a name absent from this
680 /// map (the overwhelmingly common case) runs on the full capsule ceiling, exactly as
681 /// before per-artifact narrowing existed.
682 pub(crate) artifact_grants: HashMap<String, ToolCapabilityGrant>,
683 pub(crate) hook_components: Vec<StagedHookArtifact>,
684 pub(crate) allowlisted_tools: HashSet<String>,
685 pub(crate) capability_policy: CapabilityPolicy,
686 pub(crate) otel_endpoint: Option<String>,
687 pub(crate) eval_config_json: Option<String>,
688 pub(crate) case_id: Option<String>,
689 pub(crate) dataset_id: Option<String>,
690 /// Resolved lifecycle config (manifest + override already applied).
691 pub(crate) lifecycle: LifecycleConfig,
692 /// How much of each turn's driver request this session's trace keeps (resolved from the
693 /// manifest's `trace:` block, or [`murmur_artifact::TraceCapture::default`] when it has
694 /// none).
695 pub(crate) trace_capture: murmur_artifact::TraceCapture,
696 /// What bounds this workdir's session directories, from the manifest's `trace.retain` block.
697 /// `None` — the overwhelmingly common case — prunes nothing, ever.
698 pub(crate) trace_retain: Option<murmur_artifact::TraceRetainConfig>,
699 /// Address the HTTP server is bound to (copied from StageRequest::bind_addr).
700 pub(crate) bind_addr: String,
701 /// Internal port from the manifest (copied from StageRequest::internal_port).
702 pub(crate) internal_port: Option<u16>,
703 /// This session's containment, decided once by `stage_session` from [`Self::host_probe`],
704 /// `StageRequest::declared_containment_floor`, `workdir_exec` and where the runtime directory
705 /// lies. The floor refusal, [`Self::scope_report`], [`Self::temp_mapping`], `MURMUR.md`'s exec
706 /// facts and the tier `launch_session`'s `sandbox::ShellEnforcement::resolve` installs all
707 /// read it, so the class and tier the session reports are the ones its commands run on.
708 pub(crate) containment: crate::containment::ContainmentDecision,
709 /// The complete effective grant set this session was staged with — the same report
710 /// `mur run --explain-scope --json` prints, built by `containment::scope_report_for_tier`
711 /// from this session's own policy and [`Self::containment`]. Written verbatim to
712 /// `trace.jsonl`'s `session_start` as `effective_grants`, and the source of the
713 /// `containment_declared`/`containment_achieved`/`workdir_exec` fields alongside it, so the
714 /// summary and the full report cannot disagree.
715 ///
716 /// The session's class here, `scope_report.achieved_containment`, is copied from
717 /// [`Self::containment`] when the report is built, so the two cannot drift.
718 pub(crate) scope_report: crate::containment::ScopeReport,
719 /// The single host reading this session was staged against, taken at the top of
720 /// `runtime::stage_session`, which [`Self::containment`] was decided from. `launch_session`
721 /// passes both to `sandbox::ShellEnforcement::resolve` rather than taking a new reading, so the
722 /// mechanisms installed at launch are the ones [`Self::scope_report`] already claims.
723 pub(crate) host_probe: crate::sandbox::HostProbe,
724 /// Whether a literal `/tmp` reaches the session's temporary folder, decided once by
725 /// `stage_session` from [`Self::host_probe`]: [`Self::scope_report`]'s `temp_dir` says it and
726 /// `launch_session` hands it to the enforcement that installs it.
727 pub(crate) temp_mapping: crate::temp_dir::TempMapping,
728 /// The declared read-only surface, lowered and validated by `stage_session` — a malformed
729 /// entry refuses the launch there rather than surfacing at the first call. Empty for every
730 /// capsule that declared no `capabilities.filesystem.read_only`.
731 pub(crate) protected_paths: crate::protected_paths::ProtectedPaths,
732 /// What each staged tool's own `input_schema` declared about where its inputs go, lowered by
733 /// `stage_session` from the manifests it wrote into `<runtime dir>/tools/`. Empty for a capsule
734 /// that declared no `read_only`, which reads no schema at all, and for one whose tools
735 /// annotated nothing.
736 pub(crate) tool_annotations: crate::tool_annotations::ToolAnnotationMap,
737 /// The declared read-only file surface, with its root already checked against this session's
738 /// accessible workdir by `stage_session` — a root that resolves outside it refuses the launch
739 /// rather than being served. `None` means no resource plane.
740 pub(crate) exports_files: Option<murmur_artifact::FileExport>,
741 /// The declared peer-handoff surface, with its root already checked against this session's
742 /// accessible workdir and its `max_ttl` already checked against `lifecycle.after_task` by
743 /// `stage_session`. `None` means no peer plane: nothing mints, and `/resources/peer/` answers
744 /// `no_peer_plane`.
745 pub(crate) exports_peer_files: Option<murmur_artifact::PeerFilesExport>,
746 /// Whether the door serves a task a peer capsule's runtime sends, read from the manifest's
747 /// own `exports.peer_tasks` by [`murmur_artifact::Exports::accepts_peer_tasks`]. `false`
748 /// answers every peer-origin request on the door with `403 peer_not_accepted`.
749 pub(crate) accepts_peer_tasks: bool,
750 /// Registry used to resolve this session's artifacts, retained so a pull can
751 /// resolve additional artifacts at runtime after staging has completed.
752 pub(crate) registry: Arc<dyn Registry>,
753 /// Copied from [`StageRequest::source_supply`]: where a runtime pull the stores cannot answer
754 /// is fetched from. `None` unless the capsule declared `capabilities.install.from: [stores,
755 /// sources]` and the launcher supplied one.
756 pub(crate) source_supply: Option<Arc<dyn crate::artifact_supply::SourceSupply>>,
757 /// Keeps this session's epoch ticker running: without it no epoch deadline set on any
758 /// store built from `engine` can ever fire. Held here (rather than detached) so ticking
759 /// stops when the session's `StagedSession` drops. `launch_session` moves the other
760 /// fields out one by one, which leaves this one in place until the session returns.
761 pub(crate) _epoch_ticker: EpochTicker,
762 /// The credential this session presents when it asks `mur-roost` to spawn a capsule.
763 ///
764 /// `None` for every session that cannot delegate, which is nearly all of them. Set by the
765 /// caller between `stage_session` and `launch_session` rather than carried in
766 /// [`StageRequest`]: the credential names a session id, and `stage_session` is what mints one,
767 /// so it cannot exist before staging returns.
768 pub(crate) spawn_credential: Option<SpawnCredential>,
769 /// Copied from [`StageRequest::spawn_grant`]. Read by `launch_session`, which presents it at
770 /// `POST /register` and never again — an approval covers one launch.
771 pub(crate) spawn_grant: Option<SpawnApproval>,
772 /// The handle the spawning capsule's launcher injected, read once per process at staging.
773 ///
774 /// `None` for every capsule nobody delegated. Its `session_id` and `delegation_id` become
775 /// `session_start.spawned_by` and `session_start.delegation_id`; its `report_to` — present
776 /// only for a parent that wants a completion — is what `launch_session`'s reporting guard
777 /// posts to. Read here rather than at each of those two points so one process reads
778 /// `MURMUR_SPAWNER` once and both readers see the same value.
779 pub(crate) spawner: Option<crate::delegation::SpawnerHandle>,
780 /// Copied from [`StageRequest::formation_id`]. Hooks see it as `MURMUR_FORMATION_ID` and the
781 /// delegation plane hands it to every child this session spawns.
782 pub(crate) formation_id: Option<FormationId>,
783 /// Copied from [`StageRequest::formation_member`].
784 pub(crate) formation_member: Option<Arc<FormationMember>>,
785 /// This member's formation lifeline, attached after staging by
786 /// [`StagedSession::attach_formation_lifeline`]. A process property rather than a staged
787 /// one, so it never travels on [`StageRequest`].
788 #[cfg(unix)]
789 pub(crate) formation_lifeline: Option<crate::lifeline::FormationLifeline>,
790 /// This delegated child's spawner lifeline, attached after staging by
791 /// [`StagedSession::attach_spawner_lifeline`]. A process property on the same terms as
792 /// [`Self::formation_lifeline`].
793 #[cfg(unix)]
794 pub(crate) spawner_lifeline: Option<crate::lifeline::SpawnerLifeline>,
795 /// [`StageRequest::session_deadline`], carried to the launch that enforces it.
796 pub(crate) session_deadline: Option<crate::session_status::SessionDeadline>,
797}
798
799impl StagedSession {
800 /// Writes `bytes` as the task input, `task.md` in the runtime directory, which the capsule's
801 /// commands can read and not change. A launch with one finds nothing to import from the
802 /// workdir.
803 pub fn write_task_input(&self, bytes: &[u8]) -> Result<(), crate::workdir_fs::WorkdirError> {
804 self.runtime_root.write_file(
805 crate::runtime::TASK_INPUT_FILE,
806 bytes,
807 crate::workdir_fs::DEFAULT_FILE_MODE,
808 )
809 }
810
811 /// Writes `bytes` as `rel` in the accessible workdir before launch, the folder the capsule's
812 /// tools see as their working directory. A link at `rel` is refused rather than written
813 /// through, and missing parent directories are created.
814 pub fn write_workdir_file(
815 &self,
816 rel: &str,
817 bytes: &[u8],
818 ) -> Result<(), crate::workdir_fs::WorkdirError> {
819 self.accessible_root
820 .write_file(rel, bytes, crate::workdir_fs::DEFAULT_FILE_MODE)
821 }
822
823 /// Hands this session the credential it will present to the spawning daemon.
824 pub fn set_spawn_credential(&mut self, credential: SpawnCredential) {
825 self.spawn_credential = Some(credential);
826 }
827
828 /// The credential this session presents, if it was granted one.
829 pub fn spawn_credential(&self) -> Option<&SpawnCredential> {
830 self.spawn_credential.as_ref()
831 }
832
833 /// The grant set this session was staged with, as written to the trace.
834 pub fn scope_report(&self) -> &crate::containment::ScopeReport {
835 &self.scope_report
836 }
837
838 /// Every door token this session minted, operator first, then the declared credentials by
839 /// name. Empty for a capsule that declares no `network.authentication`.
840 pub fn door_tokens(&self) -> Vec<(String, crate::door_auth::DoorToken)> {
841 self.door_auth
842 .as_ref()
843 .map(|auth| auth.tokens().to_vec())
844 .unwrap_or_default()
845 }
846
847 /// The formation this session is a member of, or `None` for a session nobody placed in one.
848 pub fn formation_id(&self) -> Option<&FormationId> {
849 self.formation_id.as_ref()
850 }
851
852 /// Hand this session the lifeline its process was launched with. Honoured only by
853 /// [`crate::launch_session_handling_sigterm`] on an agent capsule: every other launch of a
854 /// session holding one is refused with [`crate::RuntimeError::FormationLifelineUnreadable`]
855 /// before anything is instantiated.
856 #[cfg(unix)]
857 pub fn attach_formation_lifeline(&mut self, lifeline: crate::lifeline::FormationLifeline) {
858 self.formation_lifeline = Some(lifeline);
859 }
860
861 /// Hand this session the spawner lifeline its process was launched with. Honoured only by
862 /// [`crate::launch_session_handling_sigterm`], for an agent and a script capsule alike: every
863 /// other launch of a session holding one is refused with
864 /// [`crate::RuntimeError::SpawnerLifelineUnreadable`] before anything is instantiated.
865 #[cfg(unix)]
866 pub fn attach_spawner_lifeline(&mut self, lifeline: crate::lifeline::SpawnerLifeline) {
867 self.spawner_lifeline = Some(lifeline);
868 }
869}
870
871#[derive(Debug, Clone, PartialEq, Eq)]
872pub struct LaunchResult {
873 pub session_id: String,
874 /// The session's work folder, [`StagedSession::accessible_workdir`].
875 pub workdir: PathBuf,
876 /// The session's records directory, where its trace is, kept after the session.
877 pub records_dir: PathBuf,
878 /// How a launch that reports no error ended.
879 pub ending: LaunchEnding,
880}
881
882/// How a launch that reports no error ended. A launch whose task did not complete for any other
883/// reason is a [`crate::RuntimeError::TaskDidNotComplete`], not an ending.
884#[derive(Debug, Clone, Copy, PartialEq, Eq)]
885pub enum LaunchEnding {
886 /// Every task the launch ran completed, or it ran none: `session_end.exit_status` `ok`.
887 Completed,
888 /// The session's formation ended while it was running a task, and the wind-down that followed
889 /// cancelled that task: `session_end.exit_status` `formation_ended`.
890 FormationEnded,
891 /// The session reached its `mur run --deadline`, and the wind-down that followed cancelled
892 /// the task it was running, or it reached it before any task ran:
893 /// `session_end.exit_status` `deadline_reached`.
894 DeadlineReached,
895}
896
897impl LaunchEnding {
898 /// The `exit_status` `session_end` and `on-session-end` carry for this ending.
899 pub fn as_str(self) -> &'static str {
900 match self {
901 Self::Completed => "ok",
902 Self::FormationEnded => "formation_ended",
903 Self::DeadlineReached => "deadline_reached",
904 }
905 }
906}
907
908pub fn capability_policy_from_runtime_manifest(
909 manifest: &murmur_artifact::RuntimeManifest,
910) -> CapabilityPolicy {
911 let caps = manifest.capabilities.as_ref();
912 let network_allow = caps
913 .and_then(|c| c.network.as_ref())
914 .map(|network| network.allow.clone())
915 .unwrap_or_default();
916 // Absent `capabilities.peer_fetch` block means denied, on the same terms — and separately
917 // from `network.allow`, which this deliberately does not read.
918 let peer_fetch_allow = caps
919 .and_then(|c| c.peer_fetch.as_ref())
920 .map(|peer_fetch| peer_fetch.allow.clone())
921 .unwrap_or_default();
922
923 // Absent `capabilities.network` block, or absent key within it, both mean denied.
924 let unix_sockets_allowed = caps
925 .and_then(|c| c.network.as_ref())
926 .is_some_and(|network| network.unix_sockets);
927
928 // Absent `capabilities.plan` block, or `submit: false` within it, both mean denied.
929 let plan_submit = caps
930 .and_then(|c| c.plan.as_ref())
931 .is_some_and(|plan| plan.submit);
932
933 // Absent `capabilities.install` block means both lists are empty, which is deny.
934 let install = caps.and_then(|c| c.install.as_ref());
935 let install_skill = install.map(|i| i.skill.clone()).unwrap_or_default();
936 let install_tool = install.map(|i| i.tool.clone()).unwrap_or_default();
937 let install_from_sources = install.is_some_and(|i| i.from.sources());
938
939 // Recorded, never applied: see `CapabilityPolicy::state_declared`.
940 let state_declared = caps.is_some_and(|c| c.state.is_some());
941 let conversation_declared = caps.is_some_and(|c| c.conversation.is_some());
942
943 let filesystem_scope = caps
944 .and_then(|c| c.filesystem.as_ref())
945 .and_then(|filesystem| filesystem.scope.clone());
946 // Absent `capabilities.filesystem` block, or absent key within it, both mean denied — the same
947 // shape as `unix_sockets_allowed` above, for the same reason: the widening has to be declared.
948 let workdir_exec_allowed = caps
949 .and_then(|c| c.filesystem.as_ref())
950 .is_some_and(|filesystem| filesystem.workdir_exec);
951 // Absent block, absent key and an empty list are one declaration: nothing is protected.
952 let read_only_paths = caps
953 .and_then(|c| c.filesystem.as_ref())
954 .map(|filesystem| filesystem.read_only.clone())
955 .unwrap_or_default();
956
957 let shell_allow = caps
958 .and_then(|c| c.shell.as_ref())
959 .map(|shell| shell.allow.clone())
960 .unwrap_or_default();
961 let shell_strip_env = caps
962 .and_then(|c| c.shell.as_ref())
963 .and_then(|shell| shell.strip_env.clone())
964 .unwrap_or_default();
965 let shell_baseline_env = caps
966 .and_then(|c| c.shell.as_ref())
967 .and_then(|shell| shell.baseline_env.clone())
968 .unwrap_or_default();
969 let shell_interpreter_runtime = caps
970 .and_then(|c| c.shell.as_ref())
971 .map(|shell| shell.interpreter_runtime.clone())
972 .unwrap_or_default();
973 let shell_staged_runtime = caps
974 .and_then(|c| c.shell.as_ref())
975 .map(|shell| shell.staged_runtime.clone())
976 .unwrap_or_default();
977
978 let spawn_allow = caps
979 .and_then(|c| c.spawn.as_ref())
980 .map(|spawn| spawn.allow.clone())
981 .unwrap_or_default();
982
983 let env_allow = caps
984 .and_then(|c| c.env.as_ref())
985 .map(|env| env.allow.clone())
986 .unwrap_or_default();
987
988 let limits = ExecutionLimits::resolve(caps.and_then(|c| c.limits.as_ref()));
989 let declared_deadline_seconds = caps
990 .and_then(|c| c.limits.as_ref())
991 .and_then(|l| l.deadline_seconds);
992
993 let resources = crate::resources::resolve(caps.and_then(|c| c.resources.as_ref()));
994
995 CapabilityPolicy {
996 network_allow,
997 peer_fetch_allow,
998 unix_sockets_allowed,
999 filesystem_scope,
1000 workdir_exec_allowed,
1001 read_only_paths,
1002 shell_allow,
1003 spawn_allow,
1004 shell_strip_env,
1005 shell_baseline_env,
1006 shell_interpreter_runtime,
1007 shell_staged_runtime,
1008 shell_derived_interpreter_runtime: Vec::new(),
1009 shell_trust_store: Vec::new(),
1010 env_allow,
1011 limits,
1012 resources,
1013 containment_floor: caps.and_then(|c| c.containment).unwrap_or_default(),
1014 state_declared,
1015 conversation_declared,
1016 plan_submit,
1017 install_skill,
1018 install_tool,
1019 install_from_sources,
1020 declared_deadline_seconds,
1021 }
1022}
1023
1024#[cfg(test)]
1025mod tests {
1026 use super::*;
1027
1028 fn policy_for(manifest_yaml: &str) -> CapabilityPolicy {
1029 let manifest = murmur_artifact::RuntimeManifest::from_yaml_str(manifest_yaml)
1030 .expect("manifest fixture must parse");
1031 capability_policy_from_runtime_manifest(&manifest)
1032 }
1033
1034 /// `read_only` reaches the policy in declaration order, trimmed, with empty entries already
1035 /// dropped by the manifest parser — and a capsule that declares none gets an empty list, which
1036 /// is what makes the dispatch check free for it.
1037 #[test]
1038 fn read_only_paths_are_carried_through_and_default_to_empty() {
1039 let declared = policy_for(
1040 r#"
1041name: cap
1042version: 0.1.0
1043capabilities:
1044 filesystem:
1045 read_only:
1046 - tests
1047 - " bench/fixtures "
1048 - " "
1049"#,
1050 );
1051 assert_eq!(declared.read_only_paths, vec!["tests", "bench/fixtures"]);
1052
1053 let silent = policy_for(
1054 "name: cap
1055version: 0.1.0
1056",
1057 );
1058 assert!(silent.read_only_paths.is_empty());
1059
1060 let no_key = policy_for(
1061 r#"
1062name: cap
1063version: 0.1.0
1064capabilities:
1065 filesystem:
1066 scope: work
1067"#,
1068 );
1069 assert!(no_key.read_only_paths.is_empty());
1070 }
1071
1072 /// A manifest that sets no deadline gives hook calls the hook-specific default while the
1073 /// capsule's own guests keep the capsule-wide one — one manifest, two budgets.
1074 #[test]
1075 fn a_silent_manifest_gives_hook_calls_the_lower_deadline() {
1076 let policy = policy_for(
1077 r#"
1078name: cap
1079version: 0.1.0
1080artifacts: []
1081"#,
1082 );
1083
1084 assert_eq!(
1085 policy.limits.deadline_seconds,
1086 crate::limits::DEFAULT_DEADLINE_SECONDS
1087 );
1088 assert_eq!(
1089 policy.hook_limits().deadline_seconds,
1090 crate::limits::HOOK_DEFAULT_DEADLINE_SECONDS
1091 );
1092 // The default policy a hand-built fixture gets must agree with the silent-manifest one.
1093 assert_eq!(
1094 CapabilityPolicy::default().hook_limits().deadline_seconds,
1095 crate::limits::HOOK_DEFAULT_DEADLINE_SECONDS
1096 );
1097 }
1098
1099 /// An explicit `deadline_seconds` applies to every guest, hooks included — the hook
1100 /// default only fills a silence.
1101 #[test]
1102 fn an_explicit_deadline_applies_to_hook_calls_too() {
1103 let policy = policy_for(
1104 r#"
1105name: cap
1106version: 0.1.0
1107artifacts: []
1108capabilities:
1109 limits:
1110 deadline_seconds: 120
1111"#,
1112 );
1113
1114 assert_eq!(policy.declared_deadline_seconds, Some(120));
1115 assert_eq!(policy.limits.deadline_seconds, 120);
1116 assert_eq!(policy.hook_limits().deadline_seconds, 120);
1117 }
1118
1119 /// `capabilities.shell.staged_runtime` lowers onto the policy verbatim — binary, source path
1120 /// and pin all survive, since the pin is what a human later compares across two hosts.
1121 #[test]
1122 fn staged_runtime_grants_lower_onto_the_policy() {
1123 let policy = policy_for(
1124 r#"
1125name: cap
1126version: 0.1.0
1127artifacts: []
1128capabilities:
1129 containment: sealed
1130 shell:
1131 allow:
1132 - python3
1133 staged_runtime:
1134 - binary: python3
1135 source_path: /opt/testbed/conda/envs/django__django
1136 pin: conda-4.10.3/python-3.9.19
1137"#,
1138 );
1139
1140 assert_eq!(
1141 policy.shell_staged_runtime,
1142 vec![murmur_artifact::StagedRuntimeGrant {
1143 binary: "python3".to_string(),
1144 source_path: "/opt/testbed/conda/envs/django__django".to_string(),
1145 pin: "conda-4.10.3/python-3.9.19".to_string(),
1146 }]
1147 );
1148 // The two mechanisms stay separate all the way down: lowering one must not populate the
1149 // other.
1150 assert!(policy.shell_interpreter_runtime.is_empty());
1151 // Derived at staging, never lowered from the manifest.
1152 assert!(policy.shell_derived_interpreter_runtime.is_empty());
1153 assert!(policy.shell_trust_store.is_empty());
1154 }
1155
1156 /// A shell block that declares no staged runtime lowers to an empty list, not to a default
1157 /// grant of any kind.
1158 #[test]
1159 fn absent_staged_runtime_lowers_to_empty() {
1160 let policy = policy_for(
1161 r#"
1162name: cap
1163version: 0.1.0
1164artifacts: []
1165capabilities:
1166 shell:
1167 allow:
1168 - bash
1169"#,
1170 );
1171
1172 assert!(policy.shell_staged_runtime.is_empty());
1173 }
1174
1175 /// The default every existing manifest gets. Three separate "nothing was declared" shapes —
1176 /// no `capabilities:` at all, no `filesystem:` within it, and a `filesystem:` that only sets
1177 /// `scope` — must all lower to denied, because each one is a real manifest in the wild and any
1178 /// of them silently keeping workdir `Execute` would reopen the rename-to-an-allowed-basename
1179 /// bypass this default exists to close.
1180 #[test]
1181 fn workdir_exec_denied_for_every_shape_of_undeclared() {
1182 for manifest_yaml in [
1183 r#"
1184name: cap
1185version: 0.1.0
1186artifacts: []
1187"#,
1188 r#"
1189name: cap
1190version: 0.1.0
1191artifacts: []
1192capabilities:
1193 network:
1194 allow:
1195 - https://api.anthropic.com
1196"#,
1197 r#"
1198name: cap
1199version: 0.1.0
1200artifacts: []
1201capabilities:
1202 filesystem:
1203 scope: workdir
1204"#,
1205 ] {
1206 let policy = policy_for(manifest_yaml);
1207 assert!(
1208 !policy.workdir_exec_allowed,
1209 "an undeclared workdir_exec must lower to denied: {manifest_yaml}"
1210 );
1211 }
1212 }
1213
1214 #[test]
1215 fn workdir_exec_denied_when_declared_false() {
1216 let policy = policy_for(
1217 r#"
1218name: cap
1219version: 0.1.0
1220artifacts: []
1221capabilities:
1222 filesystem:
1223 workdir_exec: false
1224"#,
1225 );
1226
1227 assert!(!policy.workdir_exec_allowed);
1228 }
1229
1230 #[test]
1231 fn workdir_exec_allowed_when_declared_true() {
1232 let policy = policy_for(
1233 r#"
1234name: cap
1235version: 0.1.0
1236artifacts: []
1237capabilities:
1238 filesystem:
1239 scope: workdir
1240 workdir_exec: true
1241"#,
1242 );
1243
1244 assert!(policy.workdir_exec_allowed);
1245 // The two filesystem keys stay independent: lowering one must not disturb the other.
1246 assert_eq!(policy.filesystem_scope.as_deref(), Some("workdir"));
1247 }
1248
1249 /// A manifest with no `capabilities:` block at all denies unix sockets — the field must not
1250 /// pick up any implicit widening from the "nothing declared" path.
1251 #[test]
1252 fn unix_sockets_denied_when_capabilities_block_is_absent() {
1253 let policy = policy_for(
1254 r#"
1255name: cap
1256version: 0.1.0
1257artifacts: []
1258"#,
1259 );
1260
1261 assert!(!policy.unix_sockets_allowed);
1262 }
1263
1264 /// A declared `network:` block that never mentions the key still denies.
1265 #[test]
1266 fn unix_sockets_denied_when_network_block_omits_the_key() {
1267 let policy = policy_for(
1268 r#"
1269name: cap
1270version: 0.1.0
1271artifacts: []
1272capabilities:
1273 network:
1274 allow:
1275 - https://api.anthropic.com
1276"#,
1277 );
1278
1279 assert!(!policy.unix_sockets_allowed);
1280 assert_eq!(
1281 policy.network_allow,
1282 vec!["https://api.anthropic.com".to_string()]
1283 );
1284 }
1285
1286 #[test]
1287 fn unix_sockets_denied_when_declared_false() {
1288 let policy = policy_for(
1289 r#"
1290name: cap
1291version: 0.1.0
1292artifacts: []
1293capabilities:
1294 network:
1295 unix_sockets: false
1296"#,
1297 );
1298
1299 assert!(!policy.unix_sockets_allowed);
1300 }
1301
1302 #[test]
1303 fn unix_sockets_allowed_when_declared_true() {
1304 let policy = policy_for(
1305 r#"
1306name: cap
1307version: 0.1.0
1308artifacts: []
1309capabilities:
1310 network:
1311 unix_sockets: true
1312"#,
1313 );
1314
1315 assert!(policy.unix_sockets_allowed);
1316 }
1317
1318 /// The opt-in is independent of the IP allowlist in both directions: a non-empty `allow`
1319 /// does not imply unix sockets, and `unix_sockets: true` does not imply any IP destination.
1320 #[test]
1321 fn unix_sockets_opt_in_is_independent_of_the_ip_allowlist() {
1322 let policy = policy_for(
1323 r#"
1324name: cap
1325version: 0.1.0
1326artifacts: []
1327capabilities:
1328 network:
1329 unix_sockets: true
1330 allow:
1331 - https://api.anthropic.com
1332"#,
1333 );
1334
1335 assert!(policy.unix_sockets_allowed);
1336 assert_eq!(
1337 policy.network_allow,
1338 vec!["https://api.anthropic.com".to_string()]
1339 );
1340 }
1341
1342 /// `CapabilityPolicy::default()` is what the sandbox tests and every `..Default::default()`
1343 /// call site start from; it must deny, not inherit whatever a future field ordering implies.
1344 #[test]
1345 fn default_capability_policy_denies_unix_sockets() {
1346 assert!(!CapabilityPolicy::default().unix_sockets_allowed);
1347 }
1348}