diff --git a/src/cli/agent-context.ts b/src/cli/agent-context.ts
index f307040b3a4..224eec63b5a 100644
--- a/src/cli/agent-context.ts
+++ b/src/cli/agent-context.ts
@@ -1,9 +1,6 @@
 import type { CommandSpec } from './args'
 import { effectiveAllowedFlags } from './args'
 
-// Why: serialize the live spec table so agent discovery cannot drift from the
-// command surface it describes.
-
 const SCHEMA_VERSION = 1
 
 export type AgentContextCommand = {
@@ -29,7 +26,7 @@ export type AgentContextSchema = {
 
 export function buildAgentContext(specs: CommandSpec[]): AgentContextSchema {
   const commands = specs
-    // Why: hidden specs dispatch but stay off every discovery surface, including this one.
+    // Hidden commands dispatch but stay off discovery surfaces.
     .filter((spec) => spec.hidden !== true)
     .map((spec) => ({
       command: spec.path.join(' '),
@@ -38,8 +35,7 @@ export function buildAgentContext(specs: CommandSpec[]): AgentContextSchema {
       argumentMode: spec.argumentMode ?? 'parsed',
       summary: spec.summary,
       usage: spec.usage,
-      // Why: the effective accepted set (globals + conditional --page), not just
-      // allowedFlags — otherwise agents treat --json/--help as unsupported.
+      // Include globals and conditional flags so discovery matches validation.
       flags: effectiveAllowedFlags(spec),
       positionalArgs: spec.positionalArgs ?? [],
       ...(spec.variadicPositional ? { variadicPositional: true } : {}),
@@ -47,7 +43,6 @@ export function buildAgentContext(specs: CommandSpec[]): AgentContextSchema {
       examples: spec.examples ?? [],
       notes: spec.notes ?? []
     }))
-    // Why: deterministic ordering so the JSON diffs cleanly across runs.
     .sort((a, b) => a.command.localeCompare(b.command))
   return {
     schemaVersion: SCHEMA_VERSION,
@@ -57,8 +52,6 @@ export function buildAgentContext(specs: CommandSpec[]): AgentContextSchema {
 }
 
 export function formatAgentContextSummary(schema: AgentContextSchema): string {
-  // Why: keep the default (human) output bounded — the full surface is large, so
-  // point the reader at --json rather than dumping every command.
   return [
     `${schema.commandCount} commands (schema v${schema.schemaVersion}).`,
     'Run `orca agent-context --json` for the full machine-readable command schema.'
diff --git a/src/cli/agent-launch-request.ts b/src/cli/agent-launch-request.ts
index e2fed0f9d3c..a5151b3e441 100644
--- a/src/cli/agent-launch-request.ts
+++ b/src/cli/agent-launch-request.ts
@@ -1,8 +1,4 @@
-/**
- * How the CLI starts an agent: through `agent.launch`, the host's one launch point that the desktop
- * and phone use too. `--surface chat|terminal` picks a chat or a terminal for this launch; without
- * it the user's Chat UI setting does. Either way the receipt says which ran and why.
- */
+// agent.launch shares desktop/mobile policy; absent --surface uses the host's Chat UI setting.
 
 import { randomUUID } from 'node:crypto'
 import type { AgentLaunchPrompt, AgentLaunchResult } from '../shared/agent-launch-intent'
@@ -31,7 +27,7 @@ import { withCliAgentLaunchRecovery } from './agent-launch-recovery'
 
 type Flags = Map<string, string | boolean>
 
-/** `create` is the `worktree.create` request the CLI already builds; the host strips agent fields. */
+/** The host strips agent fields from the worktree.create payload. */
 export type CliAgentLaunchTarget =
   | { kind: 'existing'; worktree: string }
   | { kind: 'create-worktree'; create: Record<string, unknown> }
@@ -63,8 +59,6 @@ export function parseCliAgent(value: string): TuiAgent {
   return value
 }
 
-/** The agent, prompt, model and effort a CLI start asks for; refuses a model or effort the launch
- *  could not apply, by the same rule as `orchestration worker-start`. */
 export function readCliAgentLaunch(
   flags: Flags,
   agent: TuiAgent,
@@ -94,7 +88,6 @@ export function readCliAgentLaunch(
     agent,
     ...(surface ? { surface } : {}),
     ...(operationId ? { operationId } : {}),
-    // A blank prompt is no prompt: the agent starts with nothing sent.
     ...(typeof prompt === 'string' && prompt.trim()
       ? { prompt: { text: prompt, delivery: 'submit' as const } }
       : {}),
diff --git a/src/cli/agent-launch-test-fixtures.ts b/src/cli/agent-launch-test-fixtures.ts
index faa6539554d..07a53cf4f8a 100644
--- a/src/cli/agent-launch-test-fixtures.ts
+++ b/src/cli/agent-launch-test-fixtures.ts
@@ -1,12 +1,10 @@
 import { AGENT_LAUNCH_RUNTIME_CAPABILITIES } from '../shared/agent-launch-runtime-capability'
 import { okFixture } from './test-fixtures'
 
-/** A host that runs `agent.launch`, as `status.get` reports it. */
 export function agentLaunchHostFixture() {
   return okFixture('req_status', { capabilities: [...AGENT_LAUNCH_RUNTIME_CAPABILITIES] })
 }
 
-/** What `agent.launch` answers when the user's Chat UI setting opened a chat. */
 export function agentLaunchFixture(worktreeId: string, warning?: string) {
   return okFixture('req_launch', {
     outcome: { kind: 'structured', sessionId: 'codex_1', handle: 'agent_1' },
diff --git a/src/cli/agent-session-search-format.ts b/src/cli/agent-session-search-format.ts
index ebefb6ff7af..d1953c6de34 100644
--- a/src/cli/agent-session-search-format.ts
+++ b/src/cli/agent-session-search-format.ts
@@ -16,12 +16,7 @@ export type PrintableSessionSearchResponse = Exclude<
   { kind: 'malformed-cursor' }
 >
 
-/**
- * Transcript text reaches the terminal verbatim, so an OSC 52 or cursor sequence
- * inside a tool log would otherwise run on the reader's terminal. The `[[` `]]`
- * match marks are left as the engine wrote them: this CLI emits no ANSI anywhere,
- * so a colour scheme invented here would be the only one in the surface.
- */
+// Strip terminal controls from untrusted transcripts, including OSC 52 clipboard writes.
 export function terminalSafe(value: string): string {
   return stripAnsiEscapeSequences(value).replace(TERMINAL_CONTROL_CHARACTER_PATTERN, '')
 }
@@ -85,8 +80,7 @@ function formatUnavailable(
   if (reason === 'unsupported-agent') {
     return 'This host does not support history search for the selected agent. Update Orca on that host or select another agent.'
   }
-  // The CLI scopes with --path, never with an identity, so this only reaches a
-  // caller that built a request by hand.
+  // Only hand-built requests can select an unknown identity; the CLI scopes by path.
   if (reason === 'scope-unknown') {
     return 'This host does not know the workspace or project that search was scoped to.'
   }
diff --git a/src/cli/args.ts b/src/cli/args.ts
index 6ac8dd52ce2..6c2304a83ba 100644
--- a/src/cli/args.ts
+++ b/src/cli/args.ts
@@ -38,7 +38,7 @@ function setFlagValue(
   flags.set(name, value)
 }
 
-/** The most specific spec whose path prefixes `path`, so a group never shadows a leaf. */
+// Prefer the longest path so command groups cannot shadow leaf commands.
 function specForPathPrefix(
   specs: readonly CommandSpec[],
   path: readonly string[]
@@ -75,9 +75,7 @@ export function parseArgs(
     }
 
     const assignment = token.slice(2)
-    // Why: `--flag=value` is the only unambiguous way to pass a value that
-    // itself starts with `--` (e.g. `--text=--help`); the space-separated form
-    // treats a `--`-leading next token as a new flag, so it can't express one.
+    // --flag=value permits values starting with -- without treating them as another flag.
     const equalsIndex = assignment.indexOf('=')
     if (equalsIndex !== -1) {
       flagEntries.push([assignment.slice(0, equalsIndex), assignment.slice(equalsIndex + 1)])
@@ -89,7 +87,7 @@ export function parseArgs(
       flagEntries.push([flag, true])
       continue
     }
-    // Why: a pre-command flag must not consume a registry-resolvable command path.
+    // Pre-command flags must not consume a registered command path.
     if (commandPath.length === 0 && i + 1 === commandIndex) {
       flagEntries.push([flag, true])
       continue
@@ -176,7 +174,6 @@ export function supportsBrowserPageFlag(commandPath: string[]): boolean {
   ].includes(joined)
 }
 
-// Why: validation and agent discovery must expose the same effective flag set.
 export function effectiveAllowedFlags(spec: CommandSpec): string[] {
   if (spec.argumentMode === 'passthrough') {
     return []
@@ -208,14 +205,12 @@ export function isCommandGroup(specs: CommandSpec[], commandPath: string[]): boo
 export function normalizeCommandPositionals(specs: CommandSpec[], parsed: ParsedArgs): ParsedArgs {
   for (const spec of specs) {
     const positionalArgs = spec.positionalArgs ?? []
-    // Why: aliased paths still need canonicalization when there are no positionals.
+    // Alias-only paths still need canonicalization without positional arguments.
     if (positionalArgs.length === 0 && !spec.aliases) {
       continue
     }
-    // Why: canonicalize aliases before validation and dispatch so both use one key.
     for (const base of specPaths(spec)) {
-      // Why: `< 0` (not `<= 0`) so an exact base match with zero positionals
-      // still canonicalizes an aliased path; upper bound guards over-consumption.
+      // Zero positionals still permit an exact alias match.
       const positionalCount = parsed.commandPath.length - base.length
       if (
         positionalCount < 0 ||
@@ -228,8 +223,7 @@ export function normalizeCommandPositionals(specs: CommandSpec[], parsed: Parsed
       }
       const flags = new Map(parsed.flags)
       const values = parsed.commandPath.slice(base.length)
-      // Why: validation runs inside main's error-reporting path, so normalization
-      // records ambiguity instead of throwing before CLI errors can be formatted.
+      // Record ambiguity here; validation reports it through the CLI error formatter.
       const providedPositionals = values.map(
         (_, index) => positionalArgs[Math.min(index, positionalArgs.length - 1)]
       )
diff --git a/src/cli/automation-destination.ts b/src/cli/automation-destination.ts
index 0428a5b334e..ac5ef81b553 100644
--- a/src/cli/automation-destination.ts
+++ b/src/cli/automation-destination.ts
@@ -1,14 +1,4 @@
-/**
- * Where a CLI create — or an edit that moves the record's project/workspace —
- * wants the automation to land.
- *
- * The CLI cannot project a host itself, so it asks the same authority that will
- * perform the write to resolve the selector it was given and to name its
- * current SSH registrations. That captured incarnation rides along on the
- * request, and the authority re-checks it inside the write, so a target that
- * disappears in between fails the write instead of silently repointing the
- * record at a dead or re-registered host.
- */
+// Capture the authority's SSH generation so it can reject removed or re-registered targets at write time.
 
 import {
   AUTOMATION_OWNER_CONFLICT_CODES,
@@ -21,7 +11,7 @@ import type { Repo } from '../shared/repo-types'
 import type { RuntimeClient } from './runtime-client'
 import { fetchSshTargetSummaries } from './host-selector-alternatives'
 
-/** The project/workspace selectors the same request carries; both empty means the host cannot change. */
+/** Empty selectors mean the execution host cannot change. */
 export type AutomationDestinationTarget = { repo?: string; workspace?: string }
 
 async function resolveTargetRepo(
@@ -40,12 +30,7 @@ async function resolveTargetRepo(
     .repo
 }
 
-/**
- * `undefined` means the authority registers the target but assigns no
- * generation, which is a host with nothing to fence — not a host to fail
- * against. A target the authority does not register at all is positive
- * evidence that the destination is gone.
- */
+// A registered target without a generation needs no fence; an unregistered target is gone.
 async function sshTargetGeneration(
   client: RuntimeClient,
   targetId: string
diff --git a/src/cli/automation-format.test.ts b/src/cli/automation-format.test.ts
index 875307930df..e5b550aab56 100644
--- a/src/cli/automation-format.test.ts
+++ b/src/cli/automation-format.test.ts
@@ -31,10 +31,6 @@ function automation(overrides: Partial<Automation> = {}): Automation {
   }
 }
 
-// `orca automations run` sends users to `orca automations show` to find out where
-// an automation lives. The stored `executionTarget*` fields survive their host's
-// removal untouched, so printing only those renders a dead automation byte-for-byte
-// like a healthy one and the named recovery step cannot be completed.
 describe('formatAutomationShow reports the host the authority projects', () => {
   it('distinguishes a live SSH host from a replaced one by its incarnation', () => {
     const live = formatAutomationShow({
@@ -58,7 +54,7 @@ describe('formatAutomationShow reports the host the authority projects', () => {
     })
 
     expect(output).toContain('host: orphan (no host can run this automation)')
-    // The stored pin stays visible; it is the record's history, not its current owner.
+    // Keep the stored target as history, separately from current host ownership.
     expect(output).toContain('target: ssh:box-1')
   })
 
@@ -71,8 +67,6 @@ describe('formatAutomationShow reports the host the authority projects', () => {
     expect(output).toContain('host: self')
   })
 
-  // An older authority projects no owner; inventing one from the stored fields would
-  // report a removed host as healthy.
   it('omits the host line when the authority reports no owner', () => {
     const output = formatAutomationShow({ automation: automation() })
 
diff --git a/src/cli/automation-format.ts b/src/cli/automation-format.ts
index d7d355a736c..ec81da2d055 100644
--- a/src/cli/automation-format.ts
+++ b/src/cli/automation-format.ts
@@ -1,12 +1,4 @@
-/**
- * Human-readable automation output for the CLI.
- *
- * The host lines come from the authority's own projection, never from the
- * stored `executionTarget*` fields: a target that was removed — or removed and
- * re-registered under the same id — leaves those fields untouched, so printing
- * them alone would render a dead host identically to a healthy one and leave
- * `orca automations show` unable to complete the recovery it is named in.
- */
+// Use authority-projected owners: stored target fields survive host removal and re-registration.
 
 import type { Automation, AutomationRun } from '../shared/automations-types'
 import type { AutomationListItem } from '../shared/automation-list-scope'
@@ -95,7 +87,6 @@ export function formatAutomationShow(result: AutomationShowPayload): string {
     `baseBranch: ${automation.baseBranch ?? 'null'}`,
     `reuseSession: ${automation.reuseSession}`,
     `target: ${automation.executionTargetType}:${automation.executionTargetId}`,
-    // Omitted rather than guessed: a host that reports no owner has none to report.
     ...(result.owner ? [`host: ${formatOwnerSelector(result.owner.selector)}`] : []),
     `prompt: ${automation.prompt}`
   ].join('\n')
diff --git a/src/cli/automation-owner-conflict-recovery.test.ts b/src/cli/automation-owner-conflict-recovery.test.ts
index 738632945f8..379ff1367ba 100644
--- a/src/cli/automation-owner-conflict-recovery.test.ts
+++ b/src/cli/automation-owner-conflict-recovery.test.ts
@@ -25,8 +25,6 @@ describe('automation owner conflicts read as CLI outcomes', () => {
     expect(output).toContain('Next step: Run the command again')
   })
 
-  // A removed SSH host fails identically forever, so proposing a retry would send
-  // the user in a loop instead of to the two things that actually resolve it.
   it('never proposes a retry for a removed target', () => {
     const output = formatCliError(
       conflict(AUTOMATION_OWNER_CONFLICT_CODES.targetRemoved, 'Host removed.')
@@ -58,9 +56,6 @@ describe('automation owner conflicts read as CLI outcomes', () => {
     expect(output).toBe('Something else: runtime_error')
   })
 
-  // The trailing token is the convention precisely because a hop can flatten the
-  // error class. Classifying on `.code` alone reads only the hop that preserved it,
-  // and the user loses the recovery on every other one.
   it('recovers the conflict from the message when a hop flattened the code', () => {
     const flattened = new RuntimeRpcFailureError({
       id: 'request-1',
@@ -100,7 +95,6 @@ describe('automation owner conflicts read as CLI outcomes', () => {
     expect(payload.error.data.nextSteps[0]).toContain('no longer registered')
   })
 
-  // A local throw never reaches the envelope, so the token is all the CLI has.
   it('classifies a locally thrown conflict the same way', () => {
     const output = formatCliError(
       new Error(
diff --git a/src/cli/automation-owner-conflict-recovery.ts b/src/cli/automation-owner-conflict-recovery.ts
index 8a900fcb17c..d785a39977f 100644
--- a/src/cli/automation-owner-conflict-recovery.ts
+++ b/src/cli/automation-owner-conflict-recovery.ts
@@ -1,11 +1,4 @@
-/**
- * What a user should do about each automation owner conflict, phrased for the
- * CLI rather than for a window that can re-render itself.
- *
- * `automation_target_removed` deliberately offers no retry: the SSH host the
- * record is pinned to is gone, so every retry fails identically until the user
- * re-adds that host or deletes the automation.
- */
+// Removed SSH targets cannot recover through retry; users must re-add the host or delete the automation.
 
 import {
   AUTOMATION_OWNER_CONFLICT_CODES,
diff --git a/src/cli/automation-runs-paging.test.ts b/src/cli/automation-runs-paging.test.ts
index 1fd88730dcb..048c7558391 100644
--- a/src/cli/automation-runs-paging.test.ts
+++ b/src/cli/automation-runs-paging.test.ts
@@ -19,7 +19,6 @@ function run(index: number, snapshotBytes = 0): AutomationRun {
   } as unknown as AutomationRun
 }
 
-/** A paired host: pages when asked, and refuses any reply a remote connection cannot carry. */
 function pairedHost(runs: AutomationRun[], options: { pages: boolean } = { pages: true }) {
   return vi.fn(async (params: { automationId?: string; limit?: number; cursor?: string }) => {
     const result =
@@ -35,7 +34,6 @@ function pairedHost(runs: AutomationRun[], options: { pages: boolean } = { pages
 
 describe('readAllAutomationRuns', () => {
   it('reads a history larger than one remote reply, in order and without gaps', async () => {
-    // 430 runs of ~23 KB is ~10 MB, which a single unpaged reply could not carry.
     const runs = Array.from({ length: 430 }, (_, index) => run(index, 23_000))
     const host = pairedHost(runs)
     const result = await readAllAutomationRuns(host, undefined)
diff --git a/src/cli/automation-runs-paging.ts b/src/cli/automation-runs-paging.ts
index 25f31033798..264a6c5e266 100644
--- a/src/cli/automation-runs-paging.ts
+++ b/src/cli/automation-runs-paging.ts
@@ -2,7 +2,7 @@ import type { AutomationRun } from '../shared/automations-types'
 import type { RuntimeRpcSuccess } from '../shared/runtime-rpc-envelope'
 import { RuntimeClientError, type RuntimeClient } from './runtime-client'
 
-// Why 25: a snapshot can reach ~200 KB, so 25 runs stay under a remote reply's 4 MB in practice.
+// Keep typical snapshot pages below the remote reply limit; retry smaller when needed.
 export const AUTOMATION_RUNS_PAGE_SIZE = 25
 const REPLY_TOO_LARGE = 'reply_too_large'
 
@@ -10,7 +10,6 @@ type RunsReply = { runs: AutomationRun[]; nextCursor?: string | null }
 
 type RunsQuery = { automationId?: string; limit: number; cursor?: string }
 
-/** Sends one `automation.runs` request. */
 export type AutomationRunsRpc = (
   params: Partial<RunsQuery>
 ) => Promise<RuntimeRpcSuccess<RunsReply>>
@@ -44,7 +43,6 @@ async function readShrinkingPage(
   }
 }
 
-/** One page, halving the page size while the remote host refuses the reply as too large. */
 export async function readAutomationRunsPage(
   rpc: AutomationRunsRpc,
   query: RunsQuery
@@ -52,10 +50,7 @@ export async function readAutomationRunsPage(
   return (await readShrinkingPage(rpc, query)).page
 }
 
-/**
- * Every run, read page by page so no single reply has to carry the whole history (#28254).
- * A host older than paging ignores `limit` and returns everything with no cursor, ending the loop.
- */
+/** Older hosts ignore paging and return all runs without a cursor, ending the loop. */
 export async function readAllAutomationRuns(
   rpc: AutomationRunsRpc,
   automationId: string | undefined
@@ -65,7 +60,7 @@ export async function readAllAutomationRuns(
   let cursor: string | undefined
   let limit = AUTOMATION_RUNS_PAGE_SIZE
   for (;;) {
-    // Why carry the limit: the next page is likely as heavy as the one that had to shrink.
+    // Later pages may need the same reduced limit.
     const read = await readShrinkingPage(rpc, { automationId, limit, cursor })
     const page = read.page
     limit = read.limit
diff --git a/src/cli/browser-empty-values.test.ts b/src/cli/browser-empty-values.test.ts
index 23b110d46cd..5928f8c0200 100644
--- a/src/cli/browser-empty-values.test.ts
+++ b/src/cli/browser-empty-values.test.ts
@@ -37,7 +37,6 @@ vi.mock('./runtime-client', () => {
 import { main } from './index'
 import { okFixture, queueFixtures } from './test-fixtures'
 
-// Why: StorageKeyValue requires a non-empty key but accepts any string value, empty included.
 describe('orca cli storage set preserves an empty value', () => {
   beforeEach(() => {
     callMock.mockReset()
@@ -82,7 +81,7 @@ describe('orca cli storage set preserves an empty value', () => {
     })
   })
 
-  // Why: PowerShell 5.1 drops empty native-exe args, so `--value=` is the Windows escape hatch.
+  // PowerShell 5.1 drops empty native arguments; equals form preserves them.
   it('passes an empty --value= (equals form) through to browser.storage.local.set', async () => {
     queueFixtures(callMock, okFixture('req_storage_local_set', { success: true }))
     vi.spyOn(console, 'log').mockImplementation(() => {})
@@ -133,8 +132,6 @@ describe('orca cli storage set preserves an empty value', () => {
   })
 })
 
-// Why: CookieSet.value and SetCredentials.pass accept any string, empty included,
-// while name and user require a non-empty one.
 describe('orca cli cookie set and set credentials preserve an empty value', () => {
   beforeEach(() => {
     callMock.mockReset()
diff --git a/src/cli/browser-handler-groups.ts b/src/cli/browser-handler-groups.ts
index 87cf8850380..8d55d88af1b 100644
--- a/src/cli/browser-handler-groups.ts
+++ b/src/cli/browser-handler-groups.ts
@@ -1,7 +1,5 @@
 import type { HandlerGroup } from './handler-group-manifest'
 
-// Why split out: the browser command surface is a third of the CLI's groups and
-// changes as a unit, so it keeps handler-group-manifest.ts readable at a glance.
 export const BROWSER_HANDLER_GROUPS: readonly HandlerGroup[] = [
   {
     name: 'browser-identity',
diff --git a/src/cli/caller-host-worktree-selection.ts b/src/cli/caller-host-worktree-selection.ts
index 931e6d0555e..b2edfb0b2e1 100644
--- a/src/cli/caller-host-worktree-selection.ts
+++ b/src/cli/caller-host-worktree-selection.ts
@@ -14,7 +14,7 @@ type CatalogRow = Pick<RuntimeWorktreeRecord, 'id' | 'path' | 'hostId'> & {
   identity?: { key: string }
 }
 
-/** The execution host this CLI process runs on: the SSH target its relay stamped, else this machine. */
+/** SSH relays stamp the caller host; otherwise use this machine. */
 export function getCallerExecutionHostId(
   env: Readonly<Record<string, string | undefined>> = process.env
 ): ExecutionHostId {
@@ -22,9 +22,7 @@ export function getCallerExecutionHostId(
   return host?.kind === 'ssh' ? toSshExecutionHostId(host.targetId) : LOCAL_EXECUTION_HOST_ID
 }
 
-// Why: the catalog lists SSH hosts' worktrees too and two hosts can hold the same path, so an SSH
-// caller may only take its own host's rows and a local caller none of an SSH host's. `runtime:*`
-// rows in this runtime's catalog are checkouts it holds itself, and unstamped rows predate stamping.
+// Paths may exist on multiple hosts; runtime:* and unstamped rows belong to this runtime.
 function isOnCallerHost(row: CatalogRow, callerHostId: ExecutionHostId): boolean {
   const rowHostId = row.hostId ?? LOCAL_EXECUTION_HOST_ID
   return parseExecutionHostId(callerHostId)?.kind === 'ssh'
@@ -32,17 +30,15 @@ function isOnCallerHost(row: CatalogRow, callerHostId: ExecutionHostId): boolean
     : parseExecutionHostId(rowHostId)?.kind !== 'ssh'
 }
 
-// Why: `id:` keeps repo inference and the runtime's scoped fast path; only a bare id that another
-// host's row shares needs the exact identity, or the runtime would refuse it as ambiguous.
+// Prefer id: for repo inference; shared IDs need an exact host-qualified identity.
 function toHostScopedSelector(row: CatalogRow, catalog: readonly CatalogRow[]): string {
   const sharedId = catalog.some((other) => other !== row && other.id === row.id)
   return sharedId && row.identity?.key ? `identity:${row.identity.key}` : `id:${row.id}`
 }
 
-/** A host-scoped selector plus the worktree id it names, which an `identity:` selector hides. */
 export type HostScopedWorktreeSelection = { selector: string; worktreeId: string }
 
-/** The caller host's innermost worktree enclosing `cwd`; undefined when none does. */
+/** Prefer the innermost enclosing worktree on the caller's host. */
 export function selectEnclosingWorktreeSelectionOnHost(
   catalog: readonly CatalogRow[],
   cwd: string,
@@ -67,7 +63,6 @@ export function selectEnclosingWorktreeSelectionOnHost(
     : undefined
 }
 
-/** The caller host's innermost worktree enclosing `cwd`, as a selector; undefined when none does. */
 export function selectEnclosingWorktreeOnHost(
   catalog: readonly CatalogRow[],
   cwd: string,
@@ -76,7 +71,6 @@ export function selectEnclosingWorktreeOnHost(
   return selectEnclosingWorktreeSelectionOnHost(catalog, cwd, hostId)?.selector
 }
 
-/** Scopes a pane's forwarded bare worktree id to the caller's host; undefined when it has none. */
 export function selectWorktreeIdOnHost(
   catalog: readonly CatalogRow[],
   worktreeId: string,
@@ -86,7 +80,6 @@ export function selectWorktreeIdOnHost(
   return match ? toHostScopedSelector(match, catalog) : undefined
 }
 
-/** Scopes a pane's forwarded bare `ORCA_WORKTREE_ID` to the caller's host before it is used. */
 export async function resolveCallerPaneWorktreeSelector(
   worktreeId: string,
   client: RuntimeClient
diff --git a/src/cli/cli-error.ts b/src/cli/cli-error.ts
index 15491801247..d4bb81ed14d 100644
--- a/src/cli/cli-error.ts
+++ b/src/cli/cli-error.ts
@@ -13,7 +13,7 @@ export type CliErrorContext = {
   commandPath?: readonly string[]
   /** The `--worktree` value this invocation sent; the runtime's error never echoes it. */
   worktreeSelector?: string
-  /** The `--repo` value this invocation sent, for the same reason. */
+  /** The runtime error omits the original --repo selector. */
   repoSelector?: string
   /** True when --environment, --pairing-code, --host runtime:, or their env vars chose a server. */
   remoteServerSelected?: boolean
@@ -38,7 +38,7 @@ function contextRecovery(
       remoteServerSelected: context.remoteServerSelected
     })
   }
-  // Why: a paired server's handle sent to this machine reads as a dead terminal, not a missing flag.
+  // A remote handle sent locally looks stale rather than missing a flag.
   if (
     code === 'terminal_handle_stale' &&
     context.commandPath?.[0] === 'terminal' &&
@@ -78,8 +78,7 @@ export function formatCliError(error: unknown, context: CliErrorContext = {}): s
     }
     return `${message}\nOrca is not running. Run 'orca open' first.`
   }
-  // Why: error-specific recovery must win over the generic computer fallback.
-  // Classified from the whole error, not just `.code`: a hop that flattens the class leaves only the token.
+  // Match before generic recovery; flattened errors may retain only the conflict token.
   const conflict = automationOwnerConflictRecovery(matchAutomationOwnerConflict(error))
   if (conflict) {
     return formatMessageWithNextSteps(stripAutomationOwnerConflictCode(message), conflict.nextSteps)
@@ -167,7 +166,6 @@ export function reportCliError(error: unknown, json: boolean, context: CliErrorC
   }
 }
 
-/** Machine-readable half of the same recovery the human message carries. */
 function withAutomationOwnerConflictRecovery(response: RuntimeRpcFailure): RuntimeRpcFailure {
   const code = matchAutomationOwnerConflict(response)
   const conflict = automationOwnerConflictRecovery(code)
@@ -178,7 +176,7 @@ function withAutomationOwnerConflictRecovery(response: RuntimeRpcFailure): Runti
     ...response,
     error: {
       ...response.error,
-      // Restores the classification a flattening hop dropped, so --json consumers read the conflict, not the transport.
+      // Restore conflict classification lost during transport flattening.
       code,
       message: stripAutomationOwnerConflictCode(response.error.message),
       data: response.error.data ?? conflict
@@ -193,7 +191,7 @@ function formatMessageWithNextSteps(message: string, nextSteps: readonly string[
   return `${message}\n${nextSteps.map((step) => `Next step: ${step}`).join('\n')}`
 }
 
-/** Why merge: a mutation error already carries its request id, and `??` dropped the selector grammar. */
+// Preserve both mutation request IDs and selector recovery hints.
 function mergeContextRecovery(
   data: unknown,
   selector: ReturnType<typeof contextRecovery>
@@ -224,7 +222,7 @@ function nextStepsFromData(data: unknown): string[] {
   return []
 }
 
-/** Read here, not from the error's data: the servers that send this predate the hint (#31777). */
+// Older servers do not include this recovery hint in error data.
 function computerSidecarExitRecovery(error: unknown, context: CliErrorContext) {
   if (
     context.commandPath?.[0] !== 'computer' ||
@@ -243,7 +241,7 @@ function localCliErrorData(error: unknown, context: CliErrorContext): unknown {
     return sidecarExit
   }
   const selector = contextRecovery(errorCode(error), context)
-  // Why: error-specific recovery must win over the generic computer fallback.
+  // Error-specific recovery takes precedence over the generic fallback.
   if (error instanceof RuntimeClientError && error.data !== undefined) {
     return mergeContextRecovery(error.data, selector)
   }
diff --git a/src/cli/codex-command-classification.ts b/src/cli/codex-command-classification.ts
index b68efdcb60d..3c07ceb5424 100644
--- a/src/cli/codex-command-classification.ts
+++ b/src/cli/codex-command-classification.ts
@@ -231,8 +231,7 @@ function isNonInteractiveCodexSubcommand(tokens: string[]): boolean {
 
   const normalizedSubcommand = subcommand.value.toLowerCase()
   if (normalizedSubcommand === 'login') {
-    // Why: bare `codex login` displays an auth flow; only explicit status/help
-    // or stdin-fed token modes are safe to leave in a background PTY.
+    // Bare codex login opens an auth flow; only status/help or stdin-token modes are background-safe.
     const loginStartIndex = subcommand.index + 1
     const loginSubcommand = findCodexSubcommand(
       tokens,
@@ -250,8 +249,7 @@ function isNonInteractiveCodexSubcommand(tokens: string[]): boolean {
     )
   }
   if (normalizedSubcommand === 'cloud') {
-    // Why: bare `codex cloud` opens the interactive cloud browser, while its
-    // named child commands are plain one-shot commands.
+    // Bare codex cloud is interactive; named subcommands are one-shot commands.
     const cloudStartIndex = subcommand.index + 1
     const cloudSubcommand = findCodexSubcommand(
       tokens,
diff --git a/src/cli/command-scoped-flag-help.ts b/src/cli/command-scoped-flag-help.ts
index b3e85334382..e9a09bf61ca 100644
--- a/src/cli/command-scoped-flag-help.ts
+++ b/src/cli/command-scoped-flag-help.ts
@@ -1,8 +1,7 @@
-// Why: the shared --focus line describes terminal create's terminal session.
+// File focus differs from terminal create's shared --focus help.
 const FILE_OPEN_FOCUS_HELP =
   "--focus                Bring the user to the file (switches Orca's window to its worktree)"
 
-/** Per-command flag help, kept out of the shared help chain it would crowd. */
 const COMMAND_SCOPED_FLAG_HELP: Record<string, Record<string, string>> = {
   'reference add': {
     url: '--url <url-or-jira-key> Full review/issue URL or Jira key; use positionals or repeated --url',
diff --git a/src/cli/command-spec.ts b/src/cli/command-spec.ts
index 3df33960e07..8d328f69a29 100644
--- a/src/cli/command-spec.ts
+++ b/src/cli/command-spec.ts
@@ -1,23 +1,20 @@
 export type CommandSpec = {
   path: string[]
-  // Why: conventional alternate verbs should resolve without duplicating specs or handlers.
   aliases?: string[][]
   argumentMode?: 'parsed' | 'passthrough'
-  // Why: typo recovery must never steer a benign mistake into destructive state changes.
+  // Prevent typo recovery from suggesting destructive actions for unrelated commands.
   destructive?: boolean
   hidden?: boolean
   summary: string
   usage: string
   allowedFlags: string[]
-  // Why: repeatability is per-command vocabulary. `--agent` repeats for `search`
-  // and is single-valued for `worktree create`, which one global set cannot say.
+  // Flag repeatability varies by command, even for the same flag name.
   repeatableFlags?: string[]
   positionalArgs?: string[]
   variadicPositional?: boolean
   examples?: string[]
   notes?: string[]
-  // Why: `--from`/`--terminal` names either the acting caller or a target, and only the spec can
-  // say which. An agent session refuses a caller flag naming anyone else before dispatch.
+  // Caller flags must match the acting agent session; target flags may name another session.
   identityFlagRoles?: Partial<Record<IdentityFlag, 'caller' | 'target'>>
 }
 
diff --git a/src/cli/command-suggestion.test.ts b/src/cli/command-suggestion.test.ts
index 2f8a5ce71ca..8f2b6056d8b 100644
--- a/src/cli/command-suggestion.test.ts
+++ b/src/cli/command-suggestion.test.ts
@@ -28,8 +28,6 @@ const specs: CommandSpec[] = [
     allowedFlags: []
   },
   {
-    // A destructive command outside the delete-family, to prove the guard keys
-    // off the spec flag rather than a hardcoded verb list.
     path: ['emulator', 'kill'],
     destructive: true,
     summary: 'Kill the emulator',
@@ -64,8 +62,6 @@ describe('suggestCommands', () => {
   })
 
   it('never suggests a destructive command for a benign non-destructive typo', () => {
-    // `worktree move` sits distance 2 from `worktree remove`; without the
-    // guard it would sole-suggest an irreversible delete on blind retry. #6303
     const result = suggestCommands(specs, ['worktree', 'move'])
     expect(result).not.toContain('worktree remove')
     expect(result).not.toContain('worktree rm')
@@ -83,10 +79,7 @@ describe('suggestCommands', () => {
   })
 
   it('guards destructive commands outside the delete-family via the spec flag', () => {
-    // `emulator ball` is a benign token, distance 2 from the flagged `emulator
-    // kill` — close enough to otherwise rank, so the guard must exclude it.
     expect(suggestCommands(specs, ['emulator', 'ball'])).not.toContain('emulator kill')
-    // A genuine near-miss of the destructive verb still recovers.
     expect(suggestCommands(specs, ['emulator', 'kil'])).toContain('emulator kill')
   })
 
diff --git a/src/cli/command-suggestion.ts b/src/cli/command-suggestion.ts
index 47c78d74119..192bfe8b552 100644
--- a/src/cli/command-suggestion.ts
+++ b/src/cli/command-suggestion.ts
@@ -1,21 +1,17 @@
 import { specPaths, type CommandSpec } from './command-spec'
 import { levenshtein } from '../shared/edit-distance'
 
-// Why: rank the live registry so typo recovery cannot drift from accepted paths.
-
 const SUGGESTION_THRESHOLD = 3
 const MAX_SUGGESTIONS = 3
 
-// Why: a close typo of a destructive verb (`remov`→`remove`) still signals that
-// intent, but `move` (distance 2 from `remove`) does not — keep this at 1 so
-// genuine recovery works while unrelated verbs stay locked out. #6303
+// Distance 1 recovers remov→remove without suggesting deletion for move.
 const DESTRUCTIVE_INTENT_THRESHOLD = 1
 
 function finalToken(path: string[]): string {
   return path.at(-1) ?? ''
 }
 
-// Why: declared aliases connect equivalent destructive verbs without unlocking unrelated actions.
+// Aliases join destructive verb families without admitting unrelated verbs.
 function destructiveVerbFamilies(specs: CommandSpec[]): Map<string, Set<string>> {
   const families = new Map<string, Set<string>>()
   for (const spec of specs) {
@@ -48,7 +44,6 @@ export type CommandErrorData = {
   nextSteps: string[]
 }
 
-// Why: one bounded near-match ranking keeps command and flag recovery consistent.
 function rankByDistance(scored: { label: string; distance: number }[]): string[] {
   return scored
     .filter((entry) => entry.distance > 0 && entry.distance <= SUGGESTION_THRESHOLD)
@@ -57,7 +52,7 @@ function rankByDistance(scored: { label: string; distance: number }[]): string[]
     .map((entry) => entry.label)
 }
 
-// Why: same-depth matching avoids suggesting parent groups or unrelated commands.
+// Match depth to avoid suggesting parent groups or unrelated commands.
 export function suggestCommands(specs: CommandSpec[], commandPath: string[]): string[] {
   const input = commandPath.join(' ')
   const families = destructiveVerbFamilies(specs)
@@ -107,10 +102,7 @@ export type FlagErrorData = {
   nextSteps: string[]
 }
 
-// Why: edit distance cannot recover a rename. `orchestration check` is the one verb
-// that identifies its caller with `--terminal` while every sibling uses `--from`, so
-// the near-miss ranking answered `--json`/`--run` and left the caller stuck (#16904).
-// A synonym only fires where the typed flag is rejected and its partner is accepted.
+// Recover renamed flags only when the typed flag is rejected and its synonym is accepted.
 const FLAG_SYNONYMS: Readonly<Record<string, string>> = { from: 'terminal' }
 
 function suggestFlags(flag: string, validFlags: string[]): string[] {
@@ -127,7 +119,6 @@ function suggestFlags(flag: string, validFlags: string[]): string[] {
     : ranked
 }
 
-// Why: include the accepted set so agents can recover without another help call.
 export function unknownFlagData(flag: string, validFlags: string[]): FlagErrorData {
   const sortedValid = [...validFlags].sort((a, b) => a.localeCompare(b))
   const suggestions = suggestFlags(flag, sortedValid)
diff --git a/src/cli/computer-format.ts b/src/cli/computer-format.ts
index 3ec34c9cfbc..a3b4eef6e55 100644
--- a/src/cli/computer-format.ts
+++ b/src/cli/computer-format.ts
@@ -95,8 +95,7 @@ export function prepareComputerCliJsonResult<TResult>(
       }
     } as RuntimeRpcSuccess<TResult>
   } catch {
-    // Why: temp-file export is an ergonomics optimization; keep inline screenshot
-    // data when disk, permissions, or path validation would otherwise fail --json.
+    // Preserve inline screenshot data if the optional temp-file export fails.
     return response
   }
 }
@@ -124,8 +123,7 @@ function cleanupComputerScreenshots(outputDir: string): void {
   const now = Date.now()
   const markerPath = join(outputDir, COMPUTER_SCREENSHOT_CLEANUP_MARKER)
   try {
-    // Why: agents can call computer-use CLI commands in loops; a marker keeps
-    // temp cleanup from becoming a synchronous directory scan per screenshot.
+    // Throttle synchronous cleanup scans across repeated screenshots.
     if (statSync(markerPath).mtimeMs > now - COMPUTER_SCREENSHOT_CLEANUP_INTERVAL_MS) {
       return
     }
@@ -144,13 +142,13 @@ function cleanupComputerScreenshots(outputDir: string): void {
         rmSync(path, { force: true })
       }
     } catch {
-      // Best-effort cleanup only; formatting should not fail because a temp file raced.
+      // A raced temp file must not fail screenshot formatting.
     }
   }
   try {
     writeFileSync(markerPath, `${now}\n`, { mode: 0o600 })
   } catch {
-    // Best-effort marker only; stale cleanup state should not hide a screenshot.
+    // A failed cleanup marker must not hide the screenshot.
   }
 }
 
diff --git a/src/cli/dispatch.ts b/src/cli/dispatch.ts
index 7ad882b1e11..b6fb322420b 100644
--- a/src/cli/dispatch.ts
+++ b/src/cli/dispatch.ts
@@ -12,8 +12,7 @@ export type HandlerContext = {
 
 export type CommandHandler = (ctx: HandlerContext) => Promise<void>
 
-// Why: routing only needs key→group, so every CLI invocation can skip the
-// transitive module graph of the 24 groups it does not dispatch into.
+// Load only the selected handler group to avoid importing unrelated command dependencies.
 function buildRoutes(groups: readonly HandlerGroup[]): Map<string, HandlerGroup> {
   const table = new Map<string, HandlerGroup>()
   for (const group of groups) {
@@ -32,7 +31,6 @@ function buildRoutes(groups: readonly HandlerGroup[]): Map<string, HandlerGroup>
 
 const ROUTES = buildRoutes(HANDLER_GROUPS)
 
-// Why: exposes only the canonical command keys, not the handler internals.
 export const HANDLER_COMMAND_KEYS: ReadonlySet<string> = new Set(ROUTES.keys())
 
 export async function dispatch(commandPath: string[], ctx: HandlerContext): Promise<void> {
@@ -42,8 +40,7 @@ export async function dispatch(commandPath: string[], ctx: HandlerContext): Prom
     throw new RuntimeClientError('invalid_argument', `Unknown command: ${key}`)
   }
   const handler = (await group.load())[key]
-  // Why: the manifest key list is verified against the real exports in CI, so a
-  // miss here means the group changed without the manifest — fail loudly.
+  // Missing handlers indicate drift from the CI-verified manifest.
   if (!handler) {
     throw new RuntimeClientError(
       'invalid_argument',
diff --git a/src/cli/execution-host-flag.test.ts b/src/cli/execution-host-flag.test.ts
index 4cb9d9b9a36..21d6551c954 100644
--- a/src/cli/execution-host-flag.test.ts
+++ b/src/cli/execution-host-flag.test.ts
@@ -54,8 +54,6 @@ describe('parseHostFlag', () => {
     expect(() => parseHostFlag(flags({ host: true }))).toThrow('Missing value for --host')
   })
 
-  // A value that claims a kind but does not parse is malformed, not a name someone could have
-  // meant — only an unprefixed value can be a host alias.
   it.each(['runtime:', 'ssh:', 'ssh:a|b', '   '])('rejects the unparseable host id %s', (value) => {
     expect(() => parseHostFlag(flags({ host: value }))).toThrow(`Invalid --host value: ${value}`)
   })
@@ -70,8 +68,6 @@ describe('parseHostFlag', () => {
     })
   })
 
-  // Why: `orca host list` prints the alias, so the alias is what gets typed. Rejecting it sent
-  // people hunting for a machine-generated id the listing showed in a different column.
   it('carries an unprefixed host name through as an alias', () => {
     expect(parseHostFlag(flags({ host: 'trader-local' }))).toEqual({
       kind: 'alias',
@@ -111,8 +107,6 @@ describe('resolveHostFlagEnvironmentId', () => {
     ).rejects.toThrow('no paired Orca server is named or has id env-missing')
   })
 
-  // Why: the name is what a person or agent actually knows; requiring the raw uuid made the
-  // obvious spelling fail against a server that is sitting right there.
   it('accepts the environment name as well as its id', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'gpu')])
 
@@ -213,8 +207,6 @@ describe('resolveHostFlagEnvironmentId', () => {
   })
 })
 
-// Why: the reported failure. `orca host list` shows `trader-local`, so `--host trader-local` is
-// what gets typed, and only the machine-generated `ssh:ssh-1783599077486-fyasux` was accepted.
 describe('unprefixed --host aliases', () => {
   function clientWithTargets(targets: { id: string; label: string }[]): RuntimeClient {
     // oxlint-disable-next-line typescript/consistent-type-assertions -- SAFETY: SSH target resolution reads only client.call.
@@ -232,7 +224,6 @@ describe('unprefixed --host aliases', () => {
     )
   })
 
-  // The SSH axis lives on the connected runtime, so a miss here is not yet a failure.
   it('leaves an alias that names no paired server to the SSH axis', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'gpu')])
 
@@ -260,8 +251,6 @@ describe('unprefixed --host aliases', () => {
     ).rejects.toThrow('Unknown host in --host trader-local')
   })
 
-  // Why: a paired server and an SSH target are different machines. Picking either would run the
-  // command somewhere the caller never named.
   it('refuses an alias that names a machine on both axes', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'awin')])
     listSshTargetsMock.mockResolvedValue([{ id: 'ssh-1-a', label: 'awin' }])
@@ -279,9 +268,6 @@ describe('unprefixed --host aliases', () => {
     })
   })
 
-  // Why: findSshTargetByName returns undefined for a duplicated label, which reads the same as
-  // "no SSH target by that name" — so this case used to look conflict-free and resolve to the
-  // paired server, the one direction that still guessed.
   it('refuses an alias naming a server and two SSH targets sharing a label', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'awin')])
     listSshTargetsMock.mockResolvedValue([
@@ -329,7 +315,6 @@ describe('unprefixed --host aliases', () => {
     })
   })
 
-  // Why: `orca host list` prints `this machine` as the local row's name, with a space.
   it('resolves the local row name to --host local', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'gpu')])
     await expect(
@@ -375,7 +360,7 @@ describe('unprefixed --host aliases', () => {
     })
   })
 
-  // Why: cmd.exe does not strip single quotes, so `--host 'gpu'` arrives with them.
+  // cmd.exe passes wrapping single quotes through literally.
   it('ignores one pair of wrapping quotes around a host name', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'gpu')])
     listSshTargetsMock.mockResolvedValue([])
@@ -385,8 +370,6 @@ describe('unprefixed --host aliases', () => {
     expect(parseHostFlag(flags({ host: '"local"' }))).toEqual({ kind: 'local', id: 'local' })
   })
 
-  // Why: with --environment the command runs on that server, so a bare name is a machine it
-  // reaches; a locally paired server sharing the name, even the routed one, is not a host there.
   it('leaves an alias to the routed server SSH targets when --environment picks the server', async () => {
     listEnvironmentsMock.mockReturnValue([
       environment('env-1', 'openclaw'),
@@ -482,8 +465,7 @@ describe('resolveEnvironmentSelectorForStore', () => {
     expect(listSshTargetsMock).not.toHaveBeenCalled()
   })
 
-  // Why: the store matches names exactly, so a loosely matched name has to leave here as the id;
-  // an exact one stays as typed, which `orca status` shows as the target.
+  // The store matches names exactly; canonicalize loose matches to IDs.
   it('hands back the id for a name typed in another case or quoted', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'Build')])
 
@@ -495,8 +477,6 @@ describe('resolveEnvironmentSelectorForStore', () => {
     )
   })
 
-  // Why: the inverse of the --host case, and the one the report actually hit — a name that is
-  // an SSH target dead-ended with a bare "unknown environment".
   it('names the SSH target when the selector is one', async () => {
     listEnvironmentsMock.mockReturnValue([environment('env-1', 'awin')])
     listSshTargetsMock.mockResolvedValue([{ id: 'ssh-1-a', label: 'openclaw' }])
diff --git a/src/cli/execution-host-flag.ts b/src/cli/execution-host-flag.ts
index ba4cd93287d..57351723ec7 100644
--- a/src/cli/execution-host-flag.ts
+++ b/src/cli/execution-host-flag.ts
@@ -23,21 +23,16 @@ import type { RuntimeClient } from './runtime-client'
 import { RuntimeClientError } from './runtime/types'
 
 export type HostFlagRoutingSelection = {
-  // Why: SSH targets live in the running Orca host, not on disk, so enumerating them needs a
-  // client. Injected as a thunk so the lookup only happens on the error path we are explaining.
+  // Enumerating SSH targets needs the running host; defer lookup until error recovery.
   listSshTargets: () => Promise<SshTargetSummary[]>
-  // Why: unlike the suggestion listing, this throws when the targets cannot be read. A bare name
-  // may only route to a paired server once no SSH target is known to share it.
+  // Fail on unreadable targets: routing a bare name requires ruling out SSH collisions.
   verifySshTargets: () => Promise<SshTargetSummary[]>
   pairingCode: string | null
-  // Why: an ambient ORCA_ENVIRONMENT counts as a selection too, so carry the label to
-  // name the real source in the conflict message.
+  // Preserve the selection source, including ORCA_ENVIRONMENT, for conflict messages.
   environmentSelector: { value: string; label: string } | null
 }
 
-// Why: `orca host list` prints the alias a caller actually knows (`trader-local`), so that is what
-// they type — but which machine it names needs the host listings, which only the async resolvers
-// below can read. Carry the alias unresolved rather than rejecting the obvious spelling.
+// Host labels need async catalog lookup before they can resolve to a machine.
 export type HostFlagAlias = { kind: 'alias'; alias: string }
 export type HostFlagSelection = ParsedExecutionHost | HostFlagAlias
 
@@ -59,8 +54,7 @@ export function parseHostFlag(flags: Map<string, string | boolean>): HostFlagSel
     return parsed
   }
   const alias = value.trim()
-  // A value that claims a kind but does not parse is malformed, not an alias: `ssh:` names no
-  // target, and a `|` would rebind the id downstream. Only an unprefixed name can be a label.
+  // Prefixed values must parse as IDs, never aliases; malformed IDs could rebind downstream.
   if (!alias || alias.startsWith('ssh:') || alias.startsWith('runtime:')) {
     throw new RuntimeClientError(
       'invalid_argument',
@@ -70,10 +64,7 @@ export function parseHostFlag(flags: Map<string, string | boolean>): HostFlagSel
   return { kind: 'alias', alias }
 }
 
-// Why: `runtime:<environment-id>` ids are minted by this machine's pairing store, so they
-// only name a machine relative to it. Filtering or mutating the locally connected runtime
-// with one answers for the wrong host, and an id that matches nothing is indistinguishable
-// from a real host with no rows — so resolve the id here and send the command to it.
+// Pairing IDs are local to this machine; route to the paired host before applying its filters.
 export async function resolveHostFlagEnvironmentId(
   flags: Map<string, string | boolean>,
   selection: HostFlagRoutingSelection
@@ -93,25 +84,18 @@ export async function resolveHostFlagEnvironmentId(
   }))
   const wanted = host.kind === 'runtime' ? host.environmentId : host.alias
   const flagSpelling = host.kind === 'runtime' ? `--host ${host.id}` : `--host ${host.alias}`
-  // Why: with --environment or --pairing-code the command already runs on a chosen server, so a
-  // bare name is this machine or an SSH target that server reaches, never a second server.
+  // An explicit server selection scopes bare aliases to that server's local/SSH hosts.
   if (host.kind === 'alias' && (selection.environmentSelector || selection.pairingCode)) {
     return null
   }
-  // Why: --environment has always taken a name or an id, and the name is what people and agents
-  // actually know. Requiring the raw uuid here made the obvious spelling fail; accept either and
-  // canonicalize to the id so stored host ids still compare correctly downstream.
+  // Accept environment names but canonicalize them to IDs for stored-host comparisons.
   const environment = findEnvironmentByName(environments, wanted)
   if (!environment) {
     assertEnvironmentNameUnambiguous(environments, wanted, flagSpelling)
-    // A bare alias has axes left to try: this machine, and SSH targets on the connected runtime,
-    // which resolveHostFlagTarget answers once a client exists. Missing here is not a failure.
+    // A missing server alias may still name the local machine or one of its SSH targets.
     if (host.kind === 'alias') {
       return null
     }
-    // Why: `runtime:<id>` is a host id that also appears in stored rows, so it resolves by id
-    // only — unlike --environment, which also accepts a name. Say so, and hand back the ids an
-    // agent can retry with instead of making it scrape the sentence.
     const sshTargets = await selection.listSshTargets()
     throw new RuntimeClientError(
       'invalid_argument',
@@ -163,9 +147,7 @@ export async function resolveHostFlagEnvironmentId(
   return environment.id
 }
 
-// Why: a paired server and an SSH target are different machines, and one alias can name both.
-// Picking either would send the command to a machine the caller did not choose — the failure the
-// prefixed spellings exist to prevent — so hand back both flags and let them say which.
+// An alias naming both a paired server and an SSH target must not choose either silently.
 async function assertAliasNamesOneHost(
   alias: string,
   environment: EnvironmentSummary,
@@ -186,9 +168,7 @@ async function assertAliasNamesOneHost(
       }
     )
   }
-  // Why: findSshTargetByName collapses a duplicated label to undefined, which reads here as "no
-  // SSH target by that name" — so two colliding targets would have looked like no conflict at all
-  // and resolved to the server. Duplicates are more ambiguity, not less.
+  // Duplicate SSH labels are ambiguous, not evidence that no target matches.
   const matched = findSshTargetByName(sshTargets, alias)
   const conflicts = matched ? [matched] : ambiguousSshTargets(sshTargets, alias)
   if (conflicts.length === 0) {
@@ -212,9 +192,7 @@ async function assertAliasNamesOneHost(
   )
 }
 
-// Why: a runtime stamps its own setups `local` when they were made on the box and
-// `runtime:<id>` when a paired client made them. Once --host routed the command to that
-// runtime both spellings mean the same machine, so a host filter has to accept both.
+// After routing, local and runtime:<id> setups both belong to the selected server.
 export function hostFilterMatchesHostId(
   filter: ParsedExecutionHost,
   candidateHostId: string | null | undefined
@@ -226,10 +204,7 @@ export function hostFilterMatchesHostId(
   return filter.kind === 'runtime' && candidate === LOCAL_EXECUTION_HOST_ID
 }
 
-// Why: `ssh:` reaches a machine the connected runtime owns, so it can only be checked against
-// that runtime — unlike `runtime:`, which is resolved from this machine's pairing store before a
-// client exists. Callers that act on a --host value run this so an unknown target fails loudly
-// instead of quietly filtering to nothing.
+// Validate SSH IDs against the connected runtime, not this machine's pairing store.
 export async function resolveHostFlagTarget(
   flags: Map<string, string | boolean>,
   client: RuntimeClient
@@ -249,8 +224,7 @@ export async function resolveHostFlagTarget(
   if (host.kind === 'alias' && isLocalHostListName(host.alias)) {
     return resolveLocalHostListName(host.alias, client)
   }
-  // A bare alias reaches here only after resolveHostFlagEnvironmentId found no paired server by
-  // that name, so SSH is the one axis left and an unknown name is a real failure.
+  // Server lookup already missed; SSH is the remaining alias scope.
   const wanted = host.kind === 'ssh' ? host.targetId : host.alias
   const flagSpelling = host.kind === 'ssh' ? `--host ${host.id}` : `--host ${host.alias}`
   const targetId = await resolveSshHostTargetId(
@@ -289,9 +263,7 @@ async function resolveLocalHostListName(
   return { kind: 'local', id: LOCAL_EXECUTION_HOST_ID }
 }
 
-// Why: the store refuses an ambiguous environment name rather than guessing which server was
-// meant. Resolving one here would put the guess back, in the flag whose whole purpose is to stop
-// a command reaching a machine the caller did not choose.
+// Preserve the store's refusal to guess between duplicate environment names.
 function assertEnvironmentNameUnambiguous(
   environments: readonly { id: string; name: string }[],
   name: string,
@@ -313,11 +285,7 @@ function assertEnvironmentNameUnambiguous(
   )
 }
 
-// Why: the inverse of the --host case. `--environment openclaw` failed with a bare "Unknown
-// environment", when openclaw is very often an SSH target — a different axis, not a typo. The
-// store's own error cannot carry the hint (translateStoreError forwards code and message only,
-// dropping data), so resolve the selector here where the payload survives. Returns what the store
-// resolves to the same server: the store matches names exactly, so a loose match leaves as the id.
+// Resolve here so cross-host suggestions survive store errors that discard structured data.
 export async function resolveEnvironmentSelectorForStore(
   selector: string,
   listSshTargetsForSuggestion: () => Promise<SshTargetSummary[]>
diff --git a/src/cli/flag-help-text.ts b/src/cli/flag-help-text.ts
index ae5745998ff..cefbfe52e36 100644
--- a/src/cli/flag-help-text.ts
+++ b/src/cli/flag-help-text.ts
@@ -1,4 +1,3 @@
-/** One-line flag descriptions shared by every command's help output. */
 export const FLAG_HELP_TEXT: Record<string, string> = {
   'external-worktree-visibility':
     '--external-worktree-visibility show|hide|inherit  Override or inherit non-Orca worktree visibility',
@@ -113,7 +112,6 @@ export const FLAG_HELP_TEXT: Record<string, string> = {
   'value-stdin': '--value-stdin         Read set-value payload from stdin',
   'window-id': '--window-id <id>      Target a window id from list-windows',
   'window-index': '--window-index <n>   Target a window index from list-windows',
-  // Browser automation flags
   element: '--element <ref>        Element ref from snapshot (e.g. e3)',
   url: '--url <url>            URL to navigate to',
   value: '--value <text>         Value to fill or select',
diff --git a/src/cli/flags.ts b/src/cli/flags.ts
index 358217d3aa6..30fc0a72628 100644
--- a/src/cli/flags.ts
+++ b/src/cli/flags.ts
@@ -32,11 +32,7 @@ export function getOptionalStringFlag(
   return typeof value === 'string' && value.length > 0 ? value : undefined
 }
 
-/**
- * A valued flag whose value the shell (or a missing variable) ate parses as `true`. Dropping it
- * silently mints a fresh mutation identity and can deliver a prompt twice (#15180), so every
- * valued-flag accessor refuses the damaged shape by name.
- */
+// A missing valued flag must fail: silently regenerating mutation IDs can submit prompts twice.
 export function rejectValuelessFlag(value: string | boolean | undefined, name: string): void {
   if (value === true) {
     throw new RuntimeClientError(
@@ -46,11 +42,7 @@ export function rejectValuelessFlag(value: string | boolean | undefined, name: s
   }
 }
 
-/**
- * A JSON-valued flag, rejected up front when a native argv boundary stripped its quotes so the
- * error names the shell instead of the user's value (#16706). The value itself is still parsed
- * downstream; this only catches the damaged shape.
- */
+/** Detect shell-stripped JSON quotes before downstream parsing so the error names the argv boundary. */
 export function getOptionalJsonFlag(
   flags: Map<string, string | boolean>,
   name: string
diff --git a/src/cli/format-recovery.test.ts b/src/cli/format-recovery.test.ts
index a448f0f074c..c78e62a9bcd 100644
--- a/src/cli/format-recovery.test.ts
+++ b/src/cli/format-recovery.test.ts
@@ -120,7 +120,7 @@ describe('worktree selector recovery', () => {
         error: {
           code: 'selector_not_found',
           message: 'selector_not_found',
-          // The mutation-recovery layer already attached its request id.
+          // Preserve the request ID already attached by mutation recovery.
           data: { orchestrationRequestId: 'req_abc', nextSteps: ['Run request-show first.'] }
         },
         _meta: { runtimeId: 'runtime_local' }
@@ -218,7 +218,7 @@ describe('CLI error recovery', () => {
   })
 
   it('tells an older server to reprovision when its computer sidecar is missing', () => {
-    // An older orcad answers with the generic provider steps, which never name the server install.
+    // Older servers return generic steps without naming the server installation.
     const error = new RuntimeRpcFailureError({
       id: 'req_sidecar_exit',
       ok: false,
diff --git a/src/cli/format.ts b/src/cli/format.ts
index 6cb79f3e985..f5ebdf11c40 100644
--- a/src/cli/format.ts
+++ b/src/cli/format.ts
@@ -89,8 +89,7 @@ export type HostListEntry = {
   connectionStatus?: string
 }
 
-// Why: the selector column is the point of this command — the name alone is what callers already
-// had, and passing it on the wrong axis is the mistake this output exists to prevent.
+// Explicit host selectors prevent callers from using a display name on the wrong host scope.
 export function formatHostList(result: { hosts: HostListEntry[] }): string {
   const kindLabel: Record<HostListEntry['kind'], string> = {
     local: 'local',