v0.13.0

[0.13.0] - 2026-07-08

This release merges upstream (tintinweb/pi-subagents) through v0.13.0 — adopting FleetView, persist_session, foreground lifecycle symmetry, forgiving model resolution, and more — while preserving every fork feature (herdr, dashboard UI, event-forwarding, recursive widget tree, c2c, scheduling, retry handles, abort-resend, wait timeout, tool-description mode, custom agents).

Added

• FleetView — Claude Code-style subagent navigator below the editor (upstream tintinweb#114) — a persistent, navigable list of main + active subagents beneath the editor, auto-shown whenever agents run. ↓/← at an empty prompt enters the list, ↑/↓ selects, Enter opens the live conversation overlay, Esc returns. Toggle via /agents → Settings → Fleet view (default on). Pure-UI, no LLM-context cost.
• persist_session / session_dir agent frontmatter (upstream tintinweb#111) — persist_session: true writes the subagent's full transcript to pi's session location (inspectable and resumable) instead of keeping it in memory only. session_dir optionally overrides the path.
• Widget mode for the above-editor widget (upstream tintinweb#118) — new widgetMode setting: background (default — hide foreground agents that already render inline as the Agent tool result), all, or off.
• Foreground agent lifecycle symmetry (upstream tintinweb#105) — onComplete now fires for foreground agents too (with resultConsumed pre-set), and foreground conversations stream to .output files.
• Forgiving model resolution + richer Agent types list (upstream) — agent model pins resolve fuzzily (date-stamp optional, provider fallback) and the /agents types list shows each agent's resolved model.
• Herdr working status — when inside a herdr-managed pane and the new herdrReportWorking setting is on (default true), the extension reports the pane as working (custom-status running subagents) for the duration that ≥1 subagent is running, and releases the status authority when the last one finishes. Without this, herdr screen-scrapes the parent pane's buffer and mis-classifies the blocked-waiting parent as idle while subagents do the real work. Refcounted across foreground/background/concurrent/resume/abort paths. No-op outside herdr. Toggle via /agents → Settings → Herdr status.
• Steer-with-input form (/steer-subagent) — new dashboard management-modal form (agent picker + message textarea) that steers a running subagent with a user-typed message. The row-action Steer button on the /subagents table can't collect text (the dashboard's UiAction only supports a yes/no confirm), so the form is the in-protocol way to send custom steer text. The agent list rebuilds on every probe so it stays current.

Fixed

• Preserve unread completed subagents across session events (upstream tintinweb#108) — clearCompleted(skipUnconsumed) now keeps results the LLM hasn't read yet across /new, /resume, and session switches; the 10-minute cleanup timer handles eventual eviction.
• Dashboard View Result row action now opens the log — it previously replied with a ui_data_list on the row-action event, but the dashboard's /subagents table only ever renders rows for its bound dataEvent (subagents:rows), so the reply was stored and never displayed (nothing happened). The action now opens the agent's streaming outputFile (always present) in the host's default viewer (xdg-open/open/start), falling back to a tmp file of the result text when no outputFile exists. This is the one reliable way to surface a subagent log from this modal — the protocol has no row-action detail-view mechanism. (The first-party SubagentDetailView inspector is not an option for this fork: it's hard-coupled to @blackbelt-technology/pi-dashboard-subagents, a competing subagent engine that conflicts with this extension.)