05 / Daily operations

Attention & discoveries

Spend your attention where it helps the fleet move.

Find who needs you

On the galaxy, Tab selects the next agent needing you. Uppercase C opens the attention list across projects. Home also prioritizes projects with input or review waiting.

Permission questions appear as incoming transmissions with the requested command highlighted. Use a displayed number key to answer, or c to open comms and inspect the conversation. Read the request before approving it. When the galaxy is closed, desktop notifications can report questions and completed turns.

Inspect a finding

Uppercase A opens anomalies across the galaxy; lowercase a inside a project filters to that world. Findings include detected Git conflicts and failing CI, plus reported bugs, unclear behavior, documentation gaps and improvement ideas.

Each finding keeps its source and evidence. Reported findings are labelled as not independently verified. Select one and press t to prepare an investigation task. Read and edit the brief before sending; its request for a read-only investigation does not replace the agent's permission controls.

Manage the list

Automatic findings resolve when a fresh survey confirms recovery. An agent finishing an investigation is not proof that the original issue is fixed.

Report from a terminal

You or an agent can record a discovery without starting the daemon:

empire anomaly /path/to/project 'Confusing login flow' \
  'The callback discards the destination; see src/auth.rs.' 'agent report'

Use concrete evidence and a useful source label. Keep secrets and personal data out of report text if you plan to share it.

Optional report severity

With EMPIRE_ADVISOR_SEVERITY=1, PROVIDER_API_KEY and an explicit PROVIDER_BASE_URL, a compatible typed-decision service can score reports to order the attention list. Report titles, evidence and project basenames leave your machine. This is off by default; failures and uncertain answers preserve the normal ordering. The service must implement POST /v1/advice, and PROVIDER_DEFAULT_MODEL selects its model. This is not a generic chat API.

Overlapping missions

EMPIRE_ADVISOR_OVERLAP=1 adds a dim ≈ badge when working ships on the same world appear to share a task. Without advisor configuration this compares task words and file paths locally. With a configured advisor it sends project basenames, species and task snippets (up to 300 characters per ship). Failures fall back to the local comparison. This does not stop ships or replace the red badge for actual file conflicts.

Optional approval advice

With advisor configuration and EMPIRE_ADVISOR_TRIAGE=1, approval prompts show an advisory risk label. The configured service receives the question, command details, context and choices, with common secret patterns removed. Recognized destructive commands and credential requests receive a local “needs you” label instead. You still decide every approval; these labels are not a security boundary and never change an agent’s native permissions.