Summary

Remove dead model/thinking/reasoning/elevated/exec infrastructure — inline directive type fields, model directive parser, thinking-level types and normalization, help text references, and the stop command registry entry with its stub handlers.

Infrastructure to remove: model directive parser (model.ts), thinking types and normalization (thinking.ts), dead InlineDirectives type fields (hasThinkDirective, hasReasoningDirective, hasElevatedDirective, hasExecDirective + associated data fields), stop command registry entry, help text for /compact, /stop, /think, /model, /context, /skill.

Why

This infrastructure is non-functional in RemoteClaw:

• Model directive (/model <id>, /models): extractModelDirective() parser still runs during parseInlineDirectives(), but the result is never acted on — persistInlineDirectives() has the model handling gutted (line 24: "Model directive handling gutted in RemoteClaw — CLI runtimes own model selection").
• Thinking/reasoning/elevated (/think, /reasoning, /elevated): Type fields (hasThinkDirective, hasReasoningDirective, hasElevatedDirective) defined on InlineDirectives but never populated to true — the upstream parsing code that would set them was gutted. get-reply-directives.ts checks these fields but they are always undefined. The thinking.ts types (ThinkLevel, ElevatedLevel, ReasoningLevel) and normalization functions (normalizeThinkLevel, normalizeElevatedLevel, normalizeReasoningLevel) are imported across ~10 files but serve no middleware purpose — CLI agents manage their own thinking/reasoning levels.
• Exec (hasExecDirective, execAsk, execHost, execNode, execSecurity): Dead type fields on InlineDirectives, never populated. No execution handler exists.
• Stop (/stop): Registry entry at commands-registry.data.ts:458-464, but its handler in commands-session.ts:23-24 is a no-op stub returning null. The real abort mechanism lives in abort.ts (a separate code path triggered by the abort infrastructure, not the command registry).
• Help text: buildHelpMessage() in status.ts:649-678 references /compact, /stop, /think <level>, /model <id>, /context, /skill <name> [input] — all non-functional.

Context

Part of the /remoteclaw command subtree refactor. All surviving RemoteClaw commands will move under /remoteclaw <subcommand> so the agent CLI owns the bare / namespace. Removing dead infrastructure first clears noise before the restructure.

Scope

Section 1: Remove stop command registry entry + stub handlers

Note: abort.ts, abort-cutoff.ts, and their test remain untouched — the real abort mechanism is separate from the command registry stub.

Section 2: Remove model directive parser

After removal, hasModelDirective / rawModelDirective / rawModelProfile still exist on InlineDirectives but will always be false/undefined. Full InlineDirectives type cleanup is Section 3.

Section 3: Remove dead InlineDirectives type fields

Remove fields from the InlineDirectives type in directive-handling.parse.ts:11-48 that are never meaningfully populated:

Model fields (populated by parser being deleted in Section 2):

• hasModelDirective (line 17)
• rawModelDirective (line 18)
• rawModelProfile (line 19)

Think/reasoning/elevated/exec fields (never populated to true — upstream parsing was gutted):

• thinkLevel (line 32), hasThinkDirective (line 33)
• reasoningLevel (line 35), hasReasoningDirective (line 36)
• elevatedLevel (line 38), rawElevatedLevel (line 39), hasElevatedDirective (line 40)
• execAsk (line 42), execHost (line 43), execNode (line 44), execSecurity (line 45), hasExecDirective (line 47)

Cascade: remove all reads of these fields from consuming files:

• get-reply-directives.ts — checks on hasModelDirective, hasThinkDirective, hasReasoningDirective, hasElevatedDirective, hasExecDirective
• get-reply-directives-apply.ts:182 — directives.hasModelDirective check
• get-reply-directives-utils.ts:11 — hasModelDirective: false in clearInlineDirectives()
• get-reply-run.ts:391 — directives.hasThinkDirective check
• directive-handling.parse.ts:124 — directives.hasModelDirective in isDirectiveOnly()

Section 4: Remove thinking types and normalization

Section 5: Remove help text references

Section 6: Remove directive-handling.persist.ts model handling comment

Files affected

Delete entirely

• src/auto-reply/model.ts (56 lines)
• src/auto-reply/model.test.ts (181 lines)
• src/auto-reply/reply/directive-handling.levels.ts (41 lines)

Modify

• src/auto-reply/thinking.ts — remove think/elevated/reasoning types and functions (~165 lines removed, ~66 lines kept)
• src/auto-reply/commands-registry.data.ts — delete stop definition
• src/auto-reply/reply/commands-session.ts — delete handleAbortTrigger/handleStopCommand stubs + export
• src/auto-reply/reply/commands-core.ts — remove stop/abort imports and array entries
• src/auto-reply/reply/directive-handling.parse.ts — remove model parser import, call, and dead type fields
• src/auto-reply/reply/directive-handling.persist.ts — remove gutted comment
• src/auto-reply/reply/directive-handling.ts — barrel re-exports may need update
• src/auto-reply/reply/get-reply-directives.ts — remove dead directive field checks and think/reasoning/elevated level resolution
• src/auto-reply/reply/get-reply-directives-apply.ts — remove hasModelDirective check
• src/auto-reply/reply/get-reply-directives-utils.ts — remove hasModelDirective from clearInlineDirectives()
• src/auto-reply/reply/get-reply-run.ts — remove think/reasoning/elevated level resolution
• src/auto-reply/reply/get-reply.ts — remove resolved*Level destructuring
• src/auto-reply/reply/get-reply-inline-actions.ts — remove resolved*Level optional fields
• src/auto-reply/status.ts — remove ThinkLevel/ElevatedLevel/ReasoningLevel imports + help text
• src/auto-reply/reply/get-reply-run.media-only.test.ts — update test fixtures
• src/auto-reply/reply/get-reply.reset-hooks-fallback.test.ts — update test fixtures
• src/agents/model-selection.ts — remove duplicate ThinkLevel + resolveThinkingDefault
• src/agents/subagent-spawn.ts — remove thinking imports and usage
• src/cron/isolated-agent/run.test-harness.ts — remove normalizeThinkLevel mock
• src/auto-reply/commands-registry.test.ts — update/remove tests referencing model/stop/think

Acceptance criteria

• grep -r "extractModelDirective" src/ returns zero matches
• grep -r "hasThinkDirective\|hasReasoningDirective\|hasElevatedDirective\|hasExecDirective" src/ returns zero matches
• grep -r "ThinkLevel\|ElevatedLevel\|ReasoningLevel" src/ returns zero matches (excluding Telegram's local TelegramReasoningLevel type in bot-message-dispatch.ts — that is a separate local type, not the dead upstream one)
• grep -r "normalizeThinkLevel\|normalizeElevatedLevel\|normalizeReasoningLevel" src/ returns zero matches
• No stop entry in commands-registry.data.ts
• No handleStopCommand/handleAbortTrigger in commands-session.ts
• buildHelpMessage() does not reference /compact, /stop, /think, /model, /context, /skill
• abort.ts and abort-cutoff.ts remain untouched (real abort mechanism)
• VerboseLevel, normalizeVerboseLevel, UsageDisplayLevel, resolveResponseUsageMode remain in thinking.ts (alive infrastructure)
• Verbose and queue directive fields remain on InlineDirectives (alive infrastructure)
• npm run build succeeds
• npm test passes