Skip to content

Show agent readiness and effective capabilities before delegation #27

Description

@jusso-dev

Context

Adapt the pre-delegation readiness model from Buzz PR #2957 to Muster. Operators should not need to infer whether an agent is healthy, authenticated, appropriately scoped, and safe to delegate to from several disconnected screens.

Proposed behaviour

Add an authorised readiness summary to agent detail and delegation flows using existing Muster evidence:

  • configured runtime and provider/model identity;
  • gateway health, authentication, process/run lifecycle, and last heartbeat;
  • reported input/output capabilities;
  • available commands and tool-source names;
  • tool risk classes;
  • requested permission mode versus observed effective mode;
  • last verification time, freshness, and known limitations.

Use explicit states: reported, unavailable, and unknown. Missing or malformed evidence stays unknown. Stopped/failed lifecycle, stale heartbeat, closed observer, or evidence from an older process cannot produce a fresh ready state.

Security and data rules

  • Enforce organisation scope and capability checks on the server, including every delegation endpoint.
  • Expose only an allowlisted projection. Never surface credentials, environment values, executable paths, raw configuration, prompts, tool arguments/results, or external payloads.
  • Treat runtime reports and external content as untrusted evidence, not policy.
  • PostgreSQL stores authoritative agent configuration, capability assignments, readiness snapshots needed for decisions, and append-only audit/timeline events.
  • A green readiness state never bypasses approval records or capability checks for dangerous actions.

UI direction

Keep Buzz-like simplicity. Default view shows one clear state—Ready, Needs attention, or Unknown—plus the most important reason and verification age. Put detailed capabilities, permissions, and tools behind progressive disclosure. Avoid a dense always-open matrix.

Acceptance criteria

  • Agent detail and delegation UI show a clear, freshness-aware readiness state.
  • Requested and effective permission modes appear together; divergence is visible.
  • Unknown, unavailable, and reported evidence remain distinct.
  • Stale or previous-process evidence cannot yield Ready.
  • Non-authorised users cannot access readiness or capability details.
  • No sensitive runtime/config/tool data crosses the allowlist boundary.
  • Delegation still performs server-side capability checks and creates required approval/audit records.
  • Unit tests cover evidence reduction, staleness, process identity, permission divergence, and redaction.
  • Browser tests cover ready, needs-attention, unknown, and unauthorised views using synthetic data.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions