Cockpit Phase 2: live per-worker progress events (depends on #51)

[robercano-ghbot]

Context

Follow-up to #51 (Cockpit Phase 1 — read-only dashboard over issues/PRs/routing/worktrees, no harness changes). Phase 1 covers everything visible from GitHub + static config; the one thing it structurally cannot show is what a worker is doing right now.

Goal

Live per-worker progress: scoped → implementing → gate-running → reviewing → done, surfaced in the Phase 1 (or successor) dashboard.

Scope

This requires the orchestrator/implementer/reviewer agents to emit a structured event at each phase transition — e.g. one JSON line appended to a local log (.claude/state/events.jsonl or similar) with: timestamp, agent role, model, task/issue id, skill/lens in use, status. This is a real, if small, change to harness runtime behavior (prompts/agent instructions gain a "log this transition" step), so it needs its own review rather than being bundled into #51.

Design questions to resolve during implementation

• Event log format and location (single append-only file vs. per-worktree).
• Retention/rotation (unattended runs over days shouldn't grow the log unbounded).
• Whether the dashboard tails the log live or just re-reads it on refresh (Phase 1 already established "no persistent server" as the delivery model — this should stay consistent unless there's a strong reason to change it).

Depends on

#51 (the dashboard needs to exist before there's anywhere to surface this).

Acceptance

• Orchestrator/implementer/reviewer agents append a progress event at each phase transition.
• The dashboard (from Cockpit Phase 1: read-only dashboard over issues/PRs/routing/worktrees (no harness changes) #51) reflects current in-flight worker state, not just GitHub-visible state.
• No change to gate enforcement or review consensus logic — this is observability only, additive to existing agent behavior.