Self-hosted, Self-Improving, Federated Everything App — dev management, agent orchestration, persistent mind, digital twin, second brain, and creative pipeline from a single dashboard over Tailscale to port the user into the OS
JavaScript ai-agentsdashboarddeveloper-toolsdigital-twinquantified-selfsecond-brainself-hostedtailscale
/**
* The projection that turns a generated autonomous-job task back into the
* top-level fields `cos.addTask` maps onto `metadata`.
*
* `generateTaskFromJob` (`services/autonomousJobs/skillTemplates.js`) builds the
* task the SCHEDULED fire emits straight onto `task:ready`, metadata and all.
* Every other caller queues it through the non-raw `addTask` path instead, which
* only reads top-level keys — so each of them has to flatten the same bag.
*
* That flattening is an allowlist, which is exactly why it lives here once: a
* key added to the generator and missed by one caller silently vanishes from
* that lane, and the task still queues, so nothing fails loudly. (`noChangeSuccess`
* had to be added twice for this reason.) Callers layer their OWN posture on top
* of the spread — `approvalRequired`, `context`, provider pins, quota-burn
* provenance — because that is where they legitimately differ.
*
* Pure: `server/lib/` rather than beside the generator, because it needs none of
* the generator's app / skill-template / data-input I/O and both callers'
* suites double that module.
*/
/** @param {object} generated - a `generateTaskFromJob` result */
export function generatedJobTaskFields(generated) {
const meta = generated?.metadata || {};
return {
description: generated?.description,
// Preserve workflow identity through manual/quota projection. Keep any
// checkpoint id so the creation boundary rejects it instead of erasing it.
...(meta.auditDepth != null || meta.auditWorkflow != null || meta.deepAuditId != null ? {
metadata: { auditDepth: meta.auditDepth, auditWorkflow: meta.auditWorkflow, deepAuditId: meta.deepAuditId }
} : {}),
priority: generated?.priority,
prompt: meta.prompt,
// App-scoped jobs carry the target app so `prepareAgentWorkspace` resolves
// the agent's workspace to the app's repoPath instead of the PortOS root.
app: meta.app,
useWorktree: meta.useWorktree,
openPR: meta.openPR,
simplify: meta.simplify,
// A marked audit may complete successfully with a verified empty branch.
noChangeSuccess: meta.noChangeSuccess,
// The report-shaped posture: the deliverable is an action performed DURING
// the run, or a scratch checkout nothing may land from. `FILE_ISSUES_DELIVERY_SETTINGS`
// stamps `noCodeOutput` on every issues-only job, and a job converted from a
// legacy quota-burn step carries both (#6381) — missing here, the burn lane
// queued them as ordinary code work and told the agent to push.
noCodeOutput: meta.noCodeOutput,
discardWorktree: meta.discardWorktree,
worktreeChangesExpected: meta.worktreeChangesExpected,
// Read by the `job:spawned` listener, which records the execution and
// re-registers the saved schedule.
autonomousJob: meta.autonomousJob,
jobId: meta.jobId,
// Per-job AI overrides: `addTask` maps these onto `metadata.provider` /
// `.model` / `.effort`, which the spawner resolves against.
provider: meta.provider,
model: meta.model,
effort: meta.effort,
};
}
/**
* A scheduled fire hands its generated task straight to `task:ready` without
* writing it to the task store (id `<jobId>-<base36 ts>`), so the status writes
* at spawn and finalization legitimately find no row. Manual/quota runs queue
* through `addTask` and get a store-minted id, so a missing one IS a deletion.
*/
function isEphemeralJobTask(task) {
const jobId = task?.metadata?.jobId;
return task?.metadata?.autonomousJob === true && !!jobId && typeof task.id === 'string' && task.id.startsWith(`${jobId}-`);
}
/** True when an `updateTask` error is the expected miss for an unpersisted scheduled job task. */
export function isExpectedUnpersistedMiss(task, error) {
return error === 'Task not found' && isEphemeralJobTask(task);
}