Write the 0.2.93 changelog and release notes

Parallel sub-agents were close to unusable: the watchdog meant to catch
a hung run was aborting healthy fan-outs, approvals asked once per agent
and printed twice, and the panel showing what each agent was doing had
collapsed to a line of counts. The notes say what changed for the
operator rather than which branch restored it.

Author: TheGreatAxios

CHANGELOG.md

@@ -19,6 +19,66 @@ Format loosely follows [Keep a Changelog](https://keepachangelog.com/). Versions
 - Live OTEL collector verify (Phoenix or equivalent) against the merged sink.
 - Dogfood session migrate: new session under `~/.corbits/projects`, one legacy `.agent-state` migrate, write under state root still asks.
 
+## [0.2.93] - 2026-08-08
+
+Running several sub-agents at once was close to unusable. The watchdog meant to
+catch a hung run was killing healthy ones, approvals piled up one at a time and
+printed twice, and the panel showing what each agent was doing had collapsed to
+a line of counts. This release fixes that path end to end.
+
+### Sub-agents were being killed mid-run
+
+- **The stall watchdog aborted healthy parallel fan-outs.** Any silence counted
+  as a hang, and in a fan-out the first sub-agent to finish flipped the run back
+  to awaiting-a-response while the rest still worked. Since the parent emits
+  nothing while children run, the whole run read as silent: a no-response notice
+  at ninety seconds, then an abort of everything in flight at fifteen minutes.
+  The watchdog now consults the outstanding tool calls it was already tracking.
+  (CL-5641)
+- **The clock also ran while you read an approval.** Blocked-on-the-operator is
+  now part of the turn state rather than something only the painter derived, so
+  the watchdog and the phase line read one source. Gates still queued behind
+  another are covered, not only the one on screen. (CL-5642)
+
+### Approvals
+
+- **A grant no longer has to be given once per agent.** Minting one now settles
+  every queued request it already covers. The reconciliation lives in the
+  permission layer behind a single idempotent settle, so it holds for any
+  surface and cannot double-resolve or strand a request. Session teardown denies
+  whatever is still queued instead of abandoning it. (CL-4995)
+- **Every approval wrote two transcript rows.** A screen of approvals read as
+  twice as many requests as had happened. There is now exactly one row per
+  decision — including denials, timeouts, and aborts, which previously wrote
+  nothing, so a refused permission was indistinguishable from a hang. (CL-5644)
+- **Project-scoped grants never matched a sub-agent.** A grant was stamped with
+  the session root while a sub-agent asks from its own git worktree, and the two
+  were compared as plain strings — so the agents generating the approvals could
+  never benefit from an earlier answer. Both sides now resolve through the
+  worktree registry that already governs path containment, by exact match rather
+  than prefix. (CL-5662)
+
+### Watching the work
+
+- **The live agents panel is back above the prompt** — one row per running
+  sub-agent with elapsed time, current tool, and whether it has gone quiet.
+  Rows hold position instead of reordering on each event, the zone shrinks a row
+  at a time under a short terminal rather than disappearing, and when a fan-out
+  exceeds the space the quietest agent stays visible. Rows are measured in
+  terminal columns, so a wide-character description cannot overflow the zone.
+  (CL-5646)
+
+### Turns that never ended
+
+- **Goal mode stuck at "working" indefinitely.** Two events registered the same
+  tool call under different identities — one by id, one by name — so a call was
+  counted twice and cleared once, and the turn waited forever on work that had
+  finished. Goal mode was worst affected because it continues on its own and
+  never emits the fallback that settles a stalled turn. (CL-5645)
+- `run.json` records its turn count at every turn boundary rather than only at
+  session start and end, so a resumed session reports what it actually did.
+  (CL-5570, CL-5534)
+
 ## [0.2.92] - 2026-08-07
 
 A second sweep over the OpenTUI cutover, plus a rebuilt model surface. Two of

docs/release-notes-0.2.93.md

@@ -0,0 +1,62 @@
+## Parallel sub-agents actually work now
+
+Running several sub-agents at once was close to unusable. The watchdog meant to
+catch a hung run was killing healthy ones, approvals piled up one at a time and
+printed twice, and the panel that used to show what each agent was doing had
+collapsed to a single line of counts. This release fixes that path end to end.
+
+### Your agents were being killed mid-run
+
+The stall watchdog treated any silence as a hang. In a parallel fan-out the
+first sub-agent to finish flipped the run back to "awaiting a response", and
+because the parent emits nothing while children work, the whole run looked
+silent — so after ninety seconds it announced no response, and after fifteen
+minutes it aborted everything still in flight. The watchdog now knows that work
+is outstanding and stays quiet while it is.
+
+The same watchdog also ran while you were reading an approval. Being blocked on
+an operator is now part of the run's state rather than something only the
+painter knew, so a prompt you take your time over cannot be torn down
+underneath you. That applies to approvals still queued behind another, not only
+the one on screen.
+
+### Approvals
+
+**Answer once, not once per agent.** Granting a permission now clears every
+queued request that grant already covers, instead of asking again for each
+sub-agent that wanted the same thing. The reconciliation lives in the
+permission layer, so it holds for any surface, not just the terminal.
+
+**Every approval printed twice.** A screen of approvals read as twice as many
+requests as had actually happened. Now there is exactly one row per decision —
+including denials, timeouts, and aborts, which previously left no trace at all.
+An agent that died of a refused permission looked identical to one that hung.
+
+**Grants you had already given were being ignored.** A project-scoped grant was
+recorded against the session root, while a sub-agent asks from inside its own
+git worktree, and the two were compared as plain strings. No project grant ever
+matched a sub-agent — so the agents generating the approvals were the ones that
+could never benefit from a previous answer. Worktrees are now resolved through
+the same registry that governs path containment, matched exactly rather than by
+prefix, so a directory that merely starts with the same characters cannot pass
+for one.
+
+### Watching the work
+
+The live agents panel is back above the prompt: one row per running sub-agent
+with elapsed time, current tool, and whether it has gone quiet. It holds
+position as agents work rather than reordering on every event, shrinks a row at
+a time when the terminal is short instead of vanishing, and when more agents are
+running than fit, the quietest one stays visible — that is the one worth seeing.
+
+### Goal mode no longer sticks at "working"
+
+Two events could register the same tool call under different identities, one by
+id and one by name, so a call was counted twice and only cleared once. The turn
+then waited forever on a call that had already finished. Worst in goal mode,
+which keeps continuing on its own and never emitted the fallback that settles a
+stuck turn.
+
+Relatedly, `run.json` now records its turn count at every turn boundary instead
+of only at the start and end of a session, so a resumed session reports what it
+actually did.