Document and test Fable model support

AGENTS.md

@@ -30,6 +30,10 @@ Codex remains the orchestrator.
   work to Sonnet.
 - Let Claude Code use the user's configured default model unless the user
   explicitly asks for another model.
+- Use Fable only when the user explicitly asks for it; do not make it the
+  advisor, review, rescue, or delegation default.
+- Keep explicit model values as pass-through CLI arguments. Do not add a local
+  allowlist that would reject full model names or future Claude Code aliases.
 - Pass xhigh effort for Claude advisor, review, adversarial-review, rescue, and
   background work unless the user explicitly asks for another effort.
 - Use Sonnet only for junior-agent delegation governed by the

README.md

@@ -60,9 +60,14 @@ The stable command form is `$claude`. If your Codex UI exposes the skill as
   active, stale, or finished.
 - `$claude status`, `$claude result`, and `$claude cancel` manage Claude jobs.
 
+Pass `--model fable`, `--model opus`, `--model sonnet`, or a full model name to
+select a model explicitly. The companion forwards the value to Claude Code
+unchanged. Omit `--model` to use your configured default.
+
 Longer jobs can run in the background:
 
 ```text
+$claude advise --background --model fable challenge this architecture decision
 $claude advise --background should this VAD tuning loop collect N=5 now?
 $claude do --background --model sonnet map the auth module and return file:line citations
 $claude rescue --background --model opus investigate the flaky integration test
@@ -107,6 +112,9 @@ syntax errors.
 - Claude Code installed and authenticated on the same machine.
 - Node.js 18.18 or newer.
 
+Using `--model fable` also requires a Claude Code build and account that expose
+the `fable` alias. Run `$claude setup` after upgrading Claude Code.
+
 ## Local Development
 
 From a local checkout:
@@ -129,7 +137,9 @@ Use `$claude` in a Codex thread:
 
 ```text
 $claude setup
+$claude advise --background --model fable challenge this architecture decision
 $claude advise --background should this plan use a background worker?
+$claude do --background --model fable inspect the proposed design and cite risks
 $claude do --background --model sonnet map this package and cite file:line call sites
 $claude do --background --model opus debug this cross-module failure with a prepared task
 $claude rescue --background --model opus investigate the flaky integration test
@@ -179,6 +189,10 @@ Review and adversarial review are read-only. `advise` and `rescue` are also
 read-only unless you pass `--write`. Write-capable Claude work is recorded as a
 separate job type.
 
+Fable is an explicit opt-in, not a new default. Use `--model fable` only when
+the user asks for it. The same read-only, prepared-task, MCP, web, and
+explicit-write boundaries continue to apply.
+
 `$claude do --model sonnet` is for prepared junior-agent work. Before using it,
 Codex should apply `tasks-for-sonnet` and turn the request into a bounded task:
 role, absolute paths, word cap, What Must Be True, Known Constraints, Mechanical
@@ -296,6 +310,9 @@ through the installed skill. Sonnet is used only for this small routing test.
 
 - The plugin depends on the installed Claude Code CLI contract. Run
   `$claude setup` after upgrading Claude Code.
+- Model aliases and full model names are resolved by the installed Claude Code
+  CLI. Fable availability therefore depends on the installed version and
+  account; the companion does not translate or locally allowlist model names.
 - Background mode is optional. If the companion cannot verify `claude --bg`,
   `claude agents`, `claude logs`, `claude attach`, and `claude stop`, it
   degrades to foreground-only behavior.

plugins/claude-code-advisor/scripts/claude-companion.mjs

@@ -136,7 +136,7 @@ function isMeaningfulLogLine(line) {
   if (/^warning: the 'NO_COLOR' env is ignored/i.test(text) || /^at\s+/.test(text) || /\binternal:/.test(text)) {
     return false;
   }
-  if (/opus|sonnet|haiku|claude max/i.test(text)) {
+  if (/fable|opus|sonnet|haiku|claude max/i.test(text)) {
     return false;
   }
   if (/thinking with .* effort/i.test(text)) {
@@ -798,16 +798,18 @@ function printUsage() {
     [
       "Usage:",
       "  claude-companion setup [--json]",
-      "  claude-companion advise [--background] [--write] [--max-turns <n>] [--effort <level>] [--allow-mcp] [--allow-web] [--no-background-fallback] [prompt]",
+      "  claude-companion advise [--background] [--write] [--model <model>] [--max-turns <n>] [--effort <level>] [--allow-mcp] [--allow-web] [--no-background-fallback] [prompt]",
       "  claude-companion do [--background] [--write] [--model <model>] [--max-turns <n>] [--effort <level>] [--allow-mcp] [--allow-web] [prompt]",
       "  claude-companion rescue [--background] [--write] [--resume] [--model <model>] [--max-turns <n>] [--effort <level>] [--allow-mcp] [--allow-web] [--no-background-fallback] [prompt]",
-      "  claude-companion review [--base <ref>] [--max-turns <n>] [--effort <level>] [--json]",
-      "  claude-companion adversarial-review [--base <ref>] [--max-turns <n>] [--effort <level>] [focus] [--json]",
+      "  claude-companion review [--base <ref>] [--model <model>] [--max-turns <n>] [--effort <level>] [--json]",
+      "  claude-companion adversarial-review [--base <ref>] [--model <model>] [--max-turns <n>] [--effort <level>] [focus] [--json]",
       "  claude-companion monitor [job-id] [--interval-ms <ms>] [--max-checks <n>] [--stale-after-ms <ms>] [--json]",
       "  claude-companion status [job-id] [--watch] [--json]",
       "  claude-companion result [job-id] [--json]",
       "  claude-companion cancel [job-id] [--json]",
-      "  claude-companion resume-candidate [--json]"
+      "  claude-companion resume-candidate [--json]",
+      "",
+      "Models are passed through to Claude Code. Examples: fable, opus, sonnet, or a full model name."
     ].join("\n") + "\n"
   );
 }

plugins/claude-code-advisor/skills/claude/SKILL.md

@@ -3,10 +3,10 @@ name: claude
 description: >
   Use when the user asks Codex to consult Claude Code, run a Claude advisor
   pass, get a Claude adversarial review, check a plan or diff with Claude,
-  ask Claude to do a prepared coding, exploration, verifier, scout, or rescue
-  task, inspect Claude advisor job status, fetch a Claude result, or cancel a
-  Claude advisor job. This skill routes through the bundled claude-companion.mjs
-  runtime; it does not call the Claude CLI directly.
+  ask Claude Fable, Opus, or Sonnet to do a prepared coding, exploration,
+  verifier, scout, or rescue task, inspect Claude advisor job status, fetch a
+  Claude result, or cancel a Claude advisor job. This skill routes through the
+  bundled claude-companion.mjs runtime; it does not call the Claude CLI directly.
 ---
 
 # Claude Code Advisor
@@ -21,18 +21,23 @@ Canonical forms:
 
 ```text
 $claude setup
-$claude advise [--max-turns <n>] <question>
-$claude do [--background] [--write] [--model sonnet|opus] [--max-turns <n>] <prepared task>
-$claude rescue [--background] [--write] [--resume] [--model sonnet|opus] [--max-turns <n>] <task>
-$claude review [--base <ref>]
-$claude adversarial-review [--base <ref>] [focus]
+$claude advise [--background] [--model <model>] [--max-turns <n>] <question>
+$claude do [--background] [--write] [--model <model>] [--max-turns <n>] <prepared task>
+$claude rescue [--background] [--write] [--resume] [--model <model>] [--max-turns <n>] <task>
+$claude review [--base <ref>] [--model <model>]
+$claude adversarial-review [--base <ref>] [--model <model>] [focus]
 $claude monitor [job-id]
 $claude status [job-id]
 $claude result [job-id]
 $claude cancel [job-id]
 $claude resume-candidate
 ```
 
+The companion passes explicit model values through to Claude Code unchanged.
+Common aliases include `fable`, `opus`, and `sonnet`; full model names accepted
+by the installed Claude Code CLI also work. Omit `--model` to use the user's
+configured default.
+
 If Codex passes slash-style text through to this skill, normalize it before
 routing:
 
@@ -75,9 +80,12 @@ commands. The guaranteed Codex surface is the `$claude` skill mention.
   default: `Read,Glob,Grep`. `advise` may use web tools. For `do` and `rescue`,
   enable `WebFetch` or `WebSearch` only when the task needs web access and
   `--allow-web` is explicit.
-- Do not pass `--model sonnet` for advice, review, adversarial-review, rescue,
-  or monitor work unless the user explicitly asks for Sonnet. Let Claude Code
-  use the user's configured default model.
+- Do not pass an explicit `--model` for advice, review,
+  adversarial-review, rescue, or delegation work unless the user explicitly
+  asks for that model. Let Claude Code use the user's configured default model.
+- Pass explicit model values through unchanged. Do not translate `fable` or
+  maintain a local model allowlist; the installed Claude Code CLI is the source
+  of truth for supported aliases and full model names.
 - Pass `--effort xhigh` for advice, review, adversarial-review, rescue, and
   background jobs unless the user explicitly asks for another effort.
 - Sonnet is reserved for explicit junior-agent delegation governed by the
@@ -101,7 +109,8 @@ commands. The guaranteed Codex surface is the `$claude` skill mention.
 - `setup`: run the companion setup command and show the result.
 - `advise`: use for architecture questions, second opinions, and checker work.
   Use `--background` for substantive prompts, large context, or anything likely
-  to need more than one short answer.
+  to need more than one short answer. Honor an explicit model request such as
+  `--model fable`; otherwise omit `--model`.
 - `do`: use only for a specific prepared task. This is the preferred route when
   the user asks Claude to do coding, exploration, scout, verifier, reviewer, or
   synthesis work. For Sonnet, first apply `tasks-for-sonnet`; then route the
@@ -125,6 +134,7 @@ For long-running work, prefer:
 
 ```bash
 node "<plugin root>/scripts/claude-companion.mjs" advise --background "<question>"
+node "<plugin root>/scripts/claude-companion.mjs" advise --background --model fable "<question>"
 node "<plugin root>/scripts/claude-companion.mjs" do --background --model sonnet "<prepared task>"
 node "<plugin root>/scripts/claude-companion.mjs" do --background --model opus "<prepared task>"
 node "<plugin root>/scripts/claude-companion.mjs" rescue --background --model opus "<task>"
@@ -202,6 +212,23 @@ work around them when copying commands:
 --mcp-config '{"mcpServers":{}}' --strict-mcp-config --no-chrome
 ```
 
+## Explicit Fable Tasks
+
+Use `--model fable` only when the user explicitly asks for Fable. The companion
+forwards the alias unchanged and Claude Code resolves it according to the
+installed CLI and account. Keep the same read-only, task-preparation, MCP, web,
+and explicit-write boundaries used by the selected route.
+
+```text
+$claude advise --background --model fable <question>
+$claude do --background --model fable <prepared task>
+```
+
+```bash
+node "<plugin root>/scripts/claude-companion.mjs" advise --background --model fable --effort xhigh "<question>"
+node "<plugin root>/scripts/claude-companion.mjs" do --background --model fable --effort xhigh "<prepared task>"
+```
+
 ## Prepared Opus Tasks
 
 Use `$claude do --model opus` when the user asks Claude to do a complex task
@@ -243,6 +270,7 @@ Examples:
 node "<plugin root>/scripts/claude-companion.mjs" setup --json
 node "<plugin root>/scripts/claude-companion.mjs" adversarial-review --base main --json
 node "<plugin root>/scripts/claude-companion.mjs" advise --background --effort xhigh "<question>"
+node "<plugin root>/scripts/claude-companion.mjs" advise --background --model fable --effort xhigh "<question>"
 node "<plugin root>/scripts/claude-companion.mjs" do --background --model sonnet --effort xhigh "<prepared task>"
 node "<plugin root>/scripts/claude-companion.mjs" do --background --model opus --effort xhigh "<prepared task>"
 node "<plugin root>/scripts/claude-companion.mjs" monitor <job-id> --interval-ms 30000 --stale-after-ms 120000

tests/commands.test.mjs

@@ -265,6 +265,7 @@ if (args[0] === "logs") {
   console.log("at refresh (internal:util/colors:18:31)");
   console.log("at loadAssertionError (node:assert:28:96)");
   console.log("▝▜█████▛▘Opus 4.7 with xhigh effort");
+  console.log("▝▜█████▛▘Fable 5 with xhigh effort");
   console.log("▘▘ ▝▝  ~/Documents/GitHub/claude-plugin-codex");
   console.log("❯ Return exactly PASS.");
   console.log("✳Hyperspacing…");

tests/runtime.test.mjs

@@ -79,6 +79,28 @@ test("buildClaudeArgs passes explicit effort", () => {
   assert.deepEqual(args.slice(args.indexOf("--effort"), args.indexOf("--effort") + 2), ["--effort", "xhigh"]);
 });
 
+test("argument builders pass Fable through for foreground and background work", () => {
+  const foreground = buildClaudeArgs({
+    mode: "advise",
+    prompt: "check this",
+    model: "fable"
+  });
+  const background = buildBackgroundArgs({
+    prompt: "check this",
+    name: "codex-advice",
+    model: "fable"
+  });
+
+  assert.deepEqual(foreground.slice(foreground.indexOf("--model"), foreground.indexOf("--model") + 2), [
+    "--model",
+    "fable"
+  ]);
+  assert.deepEqual(background.slice(background.indexOf("--model"), background.indexOf("--model") + 2), [
+    "--model",
+    "fable"
+  ]);
+});
+
 test("buildClaudeArgs keeps local read-only tasks off web tools by default", () => {
   const args = buildClaudeArgs({
     mode: "do",

tests/validate-plugin.mjs

@@ -26,12 +26,13 @@ assert.match(skill, /claude-companion\.mjs/);
 assert.match(skill, /\/claude:rescue/);
 assert.match(skill, /\/claude:do/);
 assert.match(skill, /tasks-for-sonnet/);
+assert.match(skill, /\$claude do --background --model fable/);
 assert.match(skill, /\$claude do --model opus/);
 assert.match(skill, /complex\/high-judgment/);
 assert.match(skill, /What Must Be True/);
 assert.match(skill, /Mechanical Verification/);
 assert.match(skill, /\$claude monitor/);
-assert.match(skill, /Do not pass `--model sonnet`/);
+assert.match(skill, /Do not pass an explicit `--model`/);
 assert.match(skill, /`--effort xhigh`/);
 assert.match(skill, /--stale-after-ms 120000/);
 assert.match(skill, /--mcp-config/);
@@ -50,6 +51,7 @@ assert.match(readme, /ancestor directories/);
 assert.match(readme, /Unable to load skill contents/);
 assert.match(readme, /\$claude do/);
 assert.match(readme, /tasks-for-sonnet/);
+assert.match(readme, /--model fable/);
 assert.match(readme, /\$claude do --model opus/);
 assert.match(readme, /\$claude advise --model sonnet/);
 assert.match(e2e, /--model sonnet/);