Struct Plane 

pub struct Plane {
    pub replaced_journal: Option<PathBuf>,
    /* private fields */
}

An open plane.

Fields

replaced_journal: Option<PathBuf>

Where an unreadable journal was moved before a fresh one was opened.

Implementations

impl Plane

pub fn open_ask( dir: &Path, model: &ModelConfig, provider: Arc<dyn ModelProvider>, wiring: AskWiring, ) -> Result<Self, PlaneError>

Opens (or creates) the Ask journal at dir/journal.redb.

Errors

Another turn holds the journal, or it cannot be opened.

pub async fn ask_start( &self, id: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Starts conversation id: the run waits for its first message.

Errors

The run cannot be admitted.

pub async fn ask_deliver( &self, id: &str, message_id: &str, payload: Value, ) -> Result<bool, PlaneError>

Delivers a person’s message to conversation id; its run answers and waits again before this returns. false: no run waits (it ended or went idle).

Errors

The message cannot be stored, or the run cannot resume.

pub async fn ask_stop(&self, run: &str, by: &str) -> Result<(), PlaneError>

Stops conversation run run at its next step boundary, on the record (by): the turn in flight ends and the run with it.

Errors

run is not a run id, or the stop cannot be recorded.

impl Plane

pub fn open( dir: &Path, model: &ModelConfig, provider: Arc<dyn ModelProvider>, ) -> Result<Self, PlaneError>

Opens (or creates) the journal at dir/journal.redb with provider registered under the configured provider name.

pub fn open_with_tools( dir: &Path, model: &ModelConfig, provider: Arc<dyn ModelProvider>, tools: Option<Arc<WorkspaceTools>>, ) -> Result<Self, PlaneError>

Plane::open with speccoder’s tools bound to a worktree.

pub fn open_with( dir: &Path, model: &ModelConfig, provider: Arc<dyn ModelProvider>, tools: Option<Arc<WorkspaceTools>>, hooks: Hooks, ) -> Result<Self, PlaneError>

Plane::open_with_tools with someone to ask inline.

pub fn open_for_decisions(dir: &Path) -> Result<Self, PlaneError>

The journal with no skills and no model: enough to record a person’s decision on a waiting call (speccoder task approve), which the attempt’s own run picks up when it resumes (agentplane’s Delivery::Buffered).

pub fn open_for_reading(dir: &Path) -> Result<Option<Self>, PlaneError>

The journal read beside its writer, with no skills and no model: the waiting calls and recorded calls as they stand. None when the project has no journal. Nothing is written — not even an expired call’s denial, which the next writer records.

Errors

The journal cannot be read (open_reader).

pub async fn expire_due(&self, now: Timestamp) -> Result<(), PlaneError>

Settles every waiting call whose deadline has passed at now: the call is denied as expired, nobody’s decision. Only on a plane without skills (Plane::open_for_decisions) — the attempt resumes in its own run, with its own worktree, the next time it runs.

pub async fn withdraw( &self, task: &Task, by: &Operator, reason: &str, ) -> Result<(), PlaneError>

Withdraws a waiting call whose attempt was set aside: its run is cancelled on the record (who, why) and the call leaves the queue.

pub async fn waiting_on( &self, outcome: &RunOutcome, ) -> Result<Option<Task>, PlaneError>

The person’s task a suspended run waits on, if it waits on one.

pub async fn task(&self, id: TaskId) -> Result<Option<Task>, PlaneError>

One task row.

pub async fn tool_calls( &self, server: &str, tool: &str, ) -> Result<RecordedCalls, PlaneError>

Every call of server/tool the journal records as dispatched, in every run, with the time its run last wrote and whether it was decided on an object (a person’s decision or grant) — and what could not be read: a run unreadable is reported, never skipped silently.

pub async fn holds_a_pre_054_decision(&self, run: RunId) -> bool

Whether run read a person’s decision recorded by speccoder 0.2.0 (agentplane 0.53 and earlier): a decision with approved, which agentplane 0.54 replaced by a verdict, so the run cannot read it and fails on it. Told from the record’s shape, never a message.

pub async fn open_tasks(&self) -> Result<Vec<Task>, PlaneError>

Every task waiting for a person, oldest first.

pub async fn decide( &self, task: TaskId, decision: &Decision, ) -> Result<Delivery, PlaneError>

Records a person’s decision on task: the run continues at once when this plane holds its skill, else when it next resumes.

pub async fn decide_at( &self, task: TaskId, decision: &Decision, expected: Option<Digest>, ) -> Result<Delivery, PlaneError>

Plane::decide, refused (nothing recorded) unless the task row is still the one whose justification digest is expected.

pub async fn decide_inline( &self, outcome: RunOutcome, approver: Option<&dyn Approver>, context: &PresentContext, ) -> Result<Inline, PlaneError>

Asks approver about each call the run waits on, records the decision and lets the run continue, until it no longer waits on a person. Returns the run as it rests, the call still waiting, if the person answered “later”, nobody is there or the call waits in its run’s batch, and the calls answered Always Allow (approved once here; their argv is added to the allow list by actions::perform(AllowAlways), which the caller runs).

pub async fn outcome( &self, run: RunId, ) -> Result<Option<RunOutcome>, PlaneError>

The run as it rests now, its output included: the journal records what happened, not a step’s value, so the value is re-derived by strict replay, which performs nothing.

pub async fn converge( &self, feature: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.feature.converge, correlated to the feature.

pub async fn shape( &self, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.shape: a person’s words made into the field’s kind of text.

pub async fn skill( &self, name: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.skill.run for an installed skill.

pub async fn monitor( &self, feature: &str, task: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.task.monitor, correlated to the task, admitted once per input: judging the same change again reads the recorded answer back instead of asking again.

pub async fn near_miss( &self, feature: &str, task: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.task.near_miss, admitted once per input like monitor: a held attempt judged again after a person’s ruling meets the same variants, not new ones.

pub async fn acceptance( &self, feature: &str, task: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.task.acceptance: acceptance tests derived from the spec.

pub async fn attempt( &self, feature: &str, task: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs one speccoder.task.attempt for a bundle, correlated to the task.

pub async fn attempt_keyed( &self, feature: &str, task: &str, input: Tainted<Value>, key: &str, ) -> Result<(RunOutcome, bool), PlaneError>

attempt admitted under key: the first time it runs; after a crash the same key finds the interrupted run — a run that had rested is re-derived by strict replay, which performs nothing, and one still in flight is resumed from its journal (a dead process’s lease is freed when the store opens), so finished model and tool calls are read back, not repeated. The bool says whether this was such a resume: what the run’s notice may claim.

pub async fn propose( &self, feature: &str, task: &str, input: Tainted<Value>, ) -> Result<RunOutcome, PlaneError>

Runs speccoder.task.propose for a bundle, correlated to the task.