feat(control): complete external operator lifecycle, batch and feature controls

[Juliusolsson05]

Continue the approved external operator plan after the first release in #793 / #794. The first release has discovery, navigation, single-agent prompt/read, common UI controls, durable history and global external Codex setup; remaining operations still require repeated UI navigation or lack an explicit outcome.

Acceptance criteria, implemented in reviewed stages:

• Complete remaining lifecycle/navigation: exact-target provider switch, reload/resume/duplicate, interrupt, history/prompt lookup and rewind; explicit view and placement operations. Reuse domain owners and preserve confirmations, draft/MCP-domain continuity, identity remapping and stale-target rejection.
• Add batch reads/prompts with per-agent cursors/results and stable child request keys so partial retries cannot repeat delivered messages.
• Add broader planned controls: templates, supported settings, external workflow ownership/list/run/read, usage/worktrees and named surface navigation. Keep existing UI-only interactions documented.
• Preserve isolated SDK/feature adapters, multi-window ownership, external-only MCP/skill setup and durable accepted-versus-completed semantics.
• Maintain live tool descriptions, operator skill and implementation plan with actual coverage and limitations. Validate meaningful contracts and appropriate real external/app integration.
• Incorporate deduplicated concrete feedback linked to feat(control): expose external MCP through an isolated capability SDK #793 as it arrives; track distinct user-facing bugs separately. Do not manufacture bug issues for ordinary implementation mistakes.

Refs #793. User explicitly authorized merge of completed feature #812 after two independent Agent Code orchestration reviews, valid-finding resolution and passing final-revision CI. Do not merge a plan-only PR.

Completed

Merged #812 as dd6a8f9cd021ee4804859e692134140955429b4a, preserving the feature’s plan-first commit history. Final reviewed head: 2b0058d759a5cb447861d5f0f6d68db7066ed32f.

Both independent Agent Code orchestration reviewers cleared the final revision after the two valid P2 findings were resolved. Both CI gates passed on that exact head: https://github.com/Juliusolsson05/agent-code/actions/runs/33990704501 . The feature worktree remains clean and retained; the original worktree’s unrelated changes remain intact. No app restart or pane-layout changes were performed.

The agreed extension is delivered. Remaining external evidence/limits remain explicit: #797 actual two-monitor activation and #800 occupied/busy native committed-text trials; no complete native composer text or universal provider queue API; bounded best-effort Claude/Codex user-prompt search rather than exhaustive archive/assistant/OpenCode search (#96/#739/#773); bounded status waits rather than durable subscriptions; documented UI routes for dangerous/uncommon operations. A5/A6 and Remote fixes remain independently owned PRs. Merging does not update the currently running app; it needs a rebuild/restart to load the new tools/skill.

[Juliusolsson05]

External operator trial reports are now available for this implementation:

• feat(control): resolve visible agent labels and titles within a window #796: visible label/title lookup within a window.
• bug(control): window activation times out during external operator handoffs #797: macOS window activation acknowledgement failures.
• bug(control): layout snapshots expose stale focused session after tiled lane changes #798: stale or ambiguous effective focus in layout snapshots.
• docs(control): clarify focus versus lane replacement in operator workflows #799: focus versus lane-placement guidance.
• feat(control): expose native terminal draft uncertainty before prompt entry #800: native terminal draft uncertainty.
• feat(control): expose running app identity for computer-use attachment #801: attaching computer use to the already-running app.
• Resume discovery evidence: Resume + session search are inconsistent — sometimes show IDs, sometimes cwd names, sometimes prompts; needs to be reasoned through and redone #96 (comment) (existing Resume + session search are inconsistent — sometimes show IDs, sometimes cwd names, sometimes prompts; needs to be reasoned through and redone #96 / feat(resume): smart resume — rank and preview sessions by where the user actually spent time #739).

Each report distinguishes observations from uncertain causes and operator mistakes. These are follow-ups to #793 for consideration in #795, with distinct bug issues kept separate. A direct follow-up prompt to B7 was rejected before writing because the running Codex process was not ready; this comment keeps the references available without interrupting its work.