fix(cli): a running server notices a CLI the installer just added

A running Codeman remembers a missing CLI for up to 5 minutes, so a status
poll never spawns a login shell, and its Run button reads that same memory.
So "Run fails, add the agent, Run again" kept saying the CLI was not
installed, right after the app had said it was ready.

install.sh now touches `cli-installed` in the data dir once a CLI has landed,
and resolve() leaves the backoff when that file's mtime changes. It is one
stat, inside the backoff branch, and buys one run of the chain, so a resolved
CLI never pays for it and a CLI that is still missing goes back to waiting.
Re-walking PATH during the backoff would have found hand installs too, but it
would put a synchronous stat of every PATH entry on the request path, where
one dead network mount freezes the server (#516), and execute candidates that
were already refused.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/utils/cli-executable-resolver.ts

@@ -18,6 +18,15 @@
  * cache, a missing CLI re-ran the whole chain and stalled the event loop for up
  * to 5s on every request, forever.
  *
+ * One thing ends a backoff early: `install.sh` touches `CLI_INSTALLED_MARKER` in
+ * the data dir when it has installed a CLI, because Run reads this same answer, so
+ * "Run fails, add the agent, Run again" would otherwise fail twice. The check is
+ * one `stat` of that file, inside the backoff branch, so a resolved CLI never pays
+ * for it. It is deliberately not a re-walk of PATH during the backoff: that would
+ * put a synchronous stat of every PATH entry on the request path, where one dead
+ * network mount freezes the server (#516), and execute candidates that were
+ * already refused. A CLI installed some other way waits out the backoff as before.
+ *
  * Test hermeticity: under vitest (`process.env.VITEST`) the production host
  * short-circuits — IO primitives that were not injected become inert stubs, so
  * a suite can never scan the machine's PATH or spawn login shells (the same
@@ -34,6 +43,7 @@ import { execFileSync } from 'node:child_process';
 import { accessSync, constants, statSync } from 'node:fs';
 import { basename, delimiter, dirname, isAbsolute, join } from 'node:path';
 import { EXEC_TIMEOUT_MS } from '../config/exec-timeout.js';
+import { dataPath } from '../config/instance.js';
 import { loginShellArgs, resolveLocalShell } from './shell-resolver.js';
 
 const SAFE_BINARY_NAME = /^[a-z0-9][a-z0-9._-]*$/i;
@@ -42,6 +52,12 @@ const LOGIN_SHELL_END_MARKER = '__CODEMAN_CLI_RESOLVE_END__';
 /** Maximum rendered length of each bounded diagnostic field, excluding its label. */
 const DIAGNOSTIC_FIELD_MAX_LENGTH = 1024;
 
+/**
+ * The file `install.sh` touches in the data dir once it has installed a CLI (`mark_cli_installed`).
+ * Its mtime is all that is read; test/install-sh-invariants.test.ts pins the name on both sides.
+ */
+export const CLI_INSTALLED_MARKER = 'cli-installed';
+
 /** First retry window after a full-chain resolution miss. */
 const RESOLVE_RETRY_BASE_MS = 60_000;
 /**
@@ -79,6 +95,8 @@ export interface CliResolverHost {
   findOnProcessPath(binary: string): string | null;
   findInLoginShell(binary: string): string | null;
   exists(path: string): boolean;
+  /** A number that changes when the installer has installed a CLI; 0 when it never did. */
+  installStamp?(): number;
 }
 
 export interface CandidateValidation<T> {
@@ -113,6 +131,8 @@ export interface ProductionCliResolverHostOptions {
   shellArgs?: string[];
   runCommand?: CliResolverCommandRunner;
   isExecutableFile?: (path: string) => boolean;
+  /** Where the installer's marker is. Defaults to `CLI_INSTALLED_MARKER` in the data dir. */
+  installMarkerPath?: string;
   /**
    * Test-only escape hatch: keep the REAL IO primitives even under vitest.
    * For tests that exercise `isExecutableRegularFile` against their own temp
@@ -202,6 +222,14 @@ export function createProductionCliResolverHost(options: ProductionCliResolverHo
       return candidate && isExecutableFile(candidate) ? candidate : null;
     },
     exists: isExecutableFile,
+    installStamp: () => {
+      if (inert) return 0;
+      try {
+        return statSync(options.installMarkerPath ?? dataPath(CLI_INSTALLED_MARKER)).mtimeMs;
+      } catch {
+        return 0;
+      }
+    },
   };
 }
 
@@ -260,6 +288,8 @@ export function createCliExecutableResolver<T = undefined>(
   let lastFailureAt = 0;
   /** The invalidation generation the cached state above belongs to. */
   let generation = binaryGenerations.get(options.binary) ?? 0;
+  /** The installer's stamp at the last run of the chain: only a CHANGE is news. */
+  let installStampSeen = 0;
   const accept = (path: string | null, source: CliResolutionSource): CliResolution<T> | null => {
     if (!path || !isAbsolute(path) || !host.exists(path)) return null;
     const validation = options.validateCandidate?.(path) ?? ({ accepted: true } as CandidateValidation<T>);
@@ -285,8 +315,13 @@ export function createCliExecutableResolver<T = undefined>(
       // Negative cache: a miss is remembered and the chain — whose login-shell
       // tail is a synchronous 5s-bounded spawn — is not re-run until the
       // backoff elapses. Without this, every status poll and Run click against
-      // a missing CLI froze the event loop for the full probe, forever.
-      if (failures > 0 && now() - lastFailureAt < cliResolveRetryDelayMs(failures)) return null;
+      // a missing CLI froze the event loop for the full probe, forever. A changed
+      // installer stamp is the one early way out (see the @fileoverview), and buys
+      // one run of the chain.
+      const installStamp = host.installStamp?.() ?? 0;
+      const installedSince = installStamp !== installStampSeen;
+      if (failures > 0 && !installedSince && now() - lastFailureAt < cliResolveRetryDelayMs(failures)) return null;
+      installStampSeen = installStamp;
 
       cached = accept(host.findOnProcessPath(options.binary), 'process-path');
       if (!cached) {