[Fix] Treat a clarification question as a valid end of turn in Slack hooks

[daniel-lxs]

Problem

When an agent asks the user a question via send_chat_reply purpose clarification and ends its turn, the stop hook blocks (current_turn_nonterminal_reply — only closeout counted as terminal). The harness then injects up to 3 closeout reminders, and the model posts redundant echo messages right after its own question: "Paused here for now — I'm waiting on your answer to the question above."

Investigation notes (from the OpenRoomote debugging session; ported here per repo targeting):

• This is the only live echo path: the request_user_input branches in both hooks are dead code (nothing writes that tool into the satisfaction state), and the structured question tool blocks the turn in-flight so the stop hook never runs while it waits.
• A second echo source was caught live during e2e verification: after a closeout, a todowrite bookkeeping call staled the terminal reply (the exemption list only knew the legacy update_plan name), forcing a redundant "closed this turn" follow-up.

Fix

Principle: the hooks exist to prevent turns from ending silently. A clarification question is a visible handoff exactly like a closeout — the turn ends waiting on the user's answer — so it now counts as terminal, with the same staleness rule as closeout (non-Slack work after the question invalidates it). ack and progress still block; automation no-turn closeout enforcement is untouched.

• recordChatReplySatisfaction stamps terminal state for clarification replies
• Stop hook accepts clarification in terminal-satisfaction checks (incl. legacy states written by pre-deploy workers)
• Silence hook mirrors the same policy; OpenCode todowrite/todoread count as closeout bookkeeping and no longer stale a terminal reply
• send_chat_reply description updated: a clarification ends the turn when the next step depends on the user's answer — don't follow it with a separate "waiting on your answer" message

Verification

• Hook-script tests exercise the real generated scripts: clarification at turn end → allow (terminal_reply_satisfied); non-Slack work after the question → block (current_turn_terminal_reply_stale); ack/progress still block; todowrite/todoread don't stale a closeout
• Verified end-to-end on a live planner task via the mock Slack harness: question turn ends cleanly with the terminal state stamped and no echo for 3+ minutes (previously within one minute), and the answer round-trip completes with a single closeout
• apps/worker run-task + satisfaction tests: 228 passed; tsc --noEmit and prettier clean

🤖 Generated with Claude Code

[roomote-roomote-v1]

2 issues outstanding (1 medium, 1 low). No new issues in the latest update (test-expectation sync only). See task

• Automation no-turn tasks can now stop on a clarification — apps/worker/src/run-task/slack-stop-hook-script.ts:252-275. The no-inbound-turn branch of getCurrentTurnTerminalSatisfaction now runs through isTerminalReplyPurpose, which accepts clarification. For a requiresTerminalCloseoutWithoutTurn automation task with no currentTurnMessageTs, a send_chat_reply purpose clarification is now treated as terminal, so the Stop hook allows completion where it previously blocked with current_turn_nonterminal_reply. This diverges from the silence hook's hasNoTurnAutomationTerminalCloseout (still closeout-only) and from the PR's stated goal that "automation no-turn closeout enforcement is untouched" — the automation could end after only asking a question, with no self-contained outcome. Consider gating clarification-as-terminal on the presence of a current turn (or excluding it when requiresTerminalCloseoutWithoutTurn is set).
• todowrite/todoread exemption misses namespaced variants — apps/worker/src/run-task/slack-silence-hook-script.ts:196-202. isCloseoutBookkeepingTool matches the todo tools with endsWith('todowrite')/endsWith('todoread') only, while update_plan in the same function also handles /update_plan and .update_plan namespaced forms. If the todo tools are ever surfaced under an MCP/namespaced prefix they would not be exempted and could stale a terminal reply, reintroducing the redundant follow-up this PR fixes. Low risk today since OpenCode's todo tools are unprefixed built-ins.

[github-actions]

All contributors have signed the CLA.
_{Posted by the CLA Assistant Lite bot.}

[daniel-lxs]

I have read the CLA Document and I hereby sign the CLA