blafourcade · Oct 10, 2026 · Oct 10, 2026 · Oct 10, 2026 · Oct 10, 2026
@@ -341,7 +341,7 @@ Full catalog → [`CATALOG.md`](docs/CATALOG.md).
 | 🍳 **Recipes** | Bundled how-to sheets: [start a project](plugins/aidd-context/skills/12-cook/assets/recipes/start-a-project.md), [ship a feature](plugins/aidd-context/skills/12-cook/assets/recipes/ship-a-feature.md), [MCP installations](plugins/aidd-context/skills/12-cook/assets/recipes/mcp-installation.md), [token optimization](plugins/aidd-context/skills/12-cook/assets/recipes/token-optimization.md). Project recipes created by cook live in `aidd_docs/recipes/`. |
 | 🏛️ **[Architecture](docs/ARCHITECTURE.md)** | How the framework composes: plugins, skills, hooks, agents. |
 | 🧩 **[Create a plugin](docs/CREATE_PLUGIN.md)** | Build and publish your own. |
-| 🛒 **[Marketplace](docs/MARKETPLACE.md)** | Install scopes, versioning, LLM tiers. |
+| 🛒 **[Marketplace](docs/MARKETPLACE.md)** | Install scopes and versioning. |
 | ❓ **[FAQ & Troubleshooting](docs/FAQ.md)** · **[Glossary](docs/GLOSSARY.md)** | Common questions, fixes, and terms. |
 
 ## 🔒 Trust and safety

@@ -1,6 +1,6 @@
 # Marketplace, scopes & versioning
 
-Reference for how the `aidd-framework` marketplace is registered, scoped, and versioned, plus the LLM tier mapping used by skills.
+Reference for how the `aidd-framework` marketplace is registered, scoped, and versioned.
 
 ## 🛒 How marketplaces work
 
@@ -38,13 +38,3 @@ Set scope at install time via the `/plugin` UI, or edit `enabledPlugins` directl
 - Each plugin and the root marketplace version independently via `release-please` (tags `<plugin>-vX.Y.Z`, root `vX.Y.Z`). Tooling → [`deployment.md`](../aidd_docs/memory/deployment.md).
 - Pull updates inside Claude Code: `/plugin marketplace update aidd-framework`.
 - Full history → [`CHANGELOG.md`](../CHANGELOG.md).
-
-## 🧠 LLM tier reference
-
-Some skills target a model **tier** for a needed capability. The framework is authored against Claude; on another AI tool, map each tier to its nearest model.
-
-| Tier | Best for | Claude | Other tools (examples) |
-| ---- | -------- | ------ | ---------------------- |
-| **T1 Fast** | Mechanical, deterministic tasks, templates, git ops | Haiku 4.5 | GPT-5.5 mini, Gemini Flash, Grok fast |
-| **T2 Balanced** | Implementation, validation, code generation | Sonnet 4.6 | GPT-5.5, Gemini Pro |
-| **T3 Thinking** | Deep reasoning, synthesis, planning, onboarding | Opus 4.8 | GPT-5.5 (thinking), Gemini Pro thinking |
@@ -12,7 +12,7 @@ Write one canonical agent from intent and render it per confirmed host tool, or
 
 | #   | Action          | Role                                           | Input        |
 | --- | --------------- | ---------------------------------------------- | ------------ |
-| 01  | `capture-agent` | Gather the role, propose names, pick the model | user request |
+| 01  | `capture-agent` | Gather the role and propose names              | user request |
 | 02  | `write-agent`   | Render the agent per tool and write            | the role     |
 | 03  | `validate`      | Check each agent file                          | the files    |
 

@@ -8,12 +8,12 @@ A free-form description of the agent's purpose, tools, and instructions.
 
 ## Output
 
-In-context: the role and its system prompt, the chosen name, the model, a quality score, and the write mode (host with confirmed tools, or plugin source).
+In-context: the role and its system prompt, the chosen name, an optional user-specified model, a quality score, and the write mode (host with confirmed tools, or plugin source).
 
 ## Process
 
 1. **Gate.** Run the asset-access precheck ([tool-paths.md](../references/tool-paths.md)).
-2. **Clarify.** Ask about the purpose, tools, inputs, instructions, and preferred model. If anything stays vague, ask again before moving on.
+2. **Clarify.** Ask about the purpose, tools, inputs, and instructions. If anything stays vague, ask again before moving on.
 3. **Draft.** Write the canonical role ([agent-authoring.md](../references/agent-authoring.md)): a frontmatter intent + a system-prompt body. Include only the optional and orchestration sections it needs.
 4. **Name.** Propose three short, catchy names that fit the purpose, and have the user pick one.
 5. **Score.** Rate the agent 1-10 on clarity and completeness. If it scores under 8, revise and score again.
@@ -23,5 +23,5 @@ In-context: the role and its system prompt, the chosen name, the model, a qualit
 
 ## Test
 
-- The role, name, model, and score are stated and confirmed in writing.
+- The role, name, and score are stated and confirmed in writing. A model is included only when the user specified one.
 - The score is at least 8 before writing.
@@ -4,7 +4,7 @@ Render the canonical agent per confirmed tool and write it.
 
 ## Input
 
-From 01: the role, the chosen name, the model, and the write mode.
+From 01: the role, the chosen name, an optional user-specified model, and the write mode.
 
 ## Output
 

@@ -1,7 +1,6 @@
 ---
 name: <agent-name>                         # required
 description: <what it does + when to use>  # required, third person
-model: <e.g. opus>                         # optional, per tool-paths.md
 ---
 
 # Role

@@ -16,14 +16,14 @@ Agents are supported on all five tools.
 
 ## Frontmatter per tool
 
-The canonical agent carries `name`, `description`, `model`. Emit those a row accepts, drop the rest. Optional fields (also listed) only if the user asked. Never invent a value.
+The canonical agent carries `name` and `description`. Emit those a row accepts, drop the rest. Optional fields, including `model`, only if the user asked. Never invent a value.
 
 | Tool           | Accepts                                                      |
 | -------------- | ----------------------------------------------------------- |
-| Claude Code    | `name`, `description`, `model`, optional `color`, `tools`   |
-| Cursor         | `name`, `description`, `model`, optional `readonly`, `is_background` |
-| OpenCode       | `name`, `description`, `model`, optional `temperature`, `permission` |
-| GitHub Copilot | `name`, `description`, `model`, optional `tools`            |
+| Claude Code    | `name`, `description`, optional `model`, `color`, `tools`   |
+| Cursor         | `name`, `description`, optional `model`, `readonly`, `is_background` |
+| OpenCode       | `name`, `description`, optional `model`, `temperature`, `permission` |
+| GitHub Copilot | `name`, `description`, optional `model`, `tools`            |
 | Codex CLI      | `name`, `description` (drops `model`)                       |
 
 ## Codex TOML conversion

@@ -1,7 +1,6 @@
 ---
 name: checker
 description: Judges finished work against its validator and the real need, leaving nothing unchecked. Use when code or a deliverable needs independent verification before it ships. Never edits the work, never implements the fix.
-model: opus
 ---
 
 # Role

@@ -1,7 +1,6 @@
 ---
 name: executor
 description: Turns a dispatched task into working, validated code that fits the project. Use when an approved scope must become code. Never plans, never judges its own work.
-model: sonnet
 ---
 
 # Role

@@ -2,7 +2,6 @@
 name: 03-assert
 description: Assert the work behaves by iterating the project's coding assertions until they pass, plus optional architecture and frontend facets. Use to validate an implementation. Not for reviewing or writing tests.
 argument-hint: work | scope
-model: sonnet
 ---
 
 # Skill: assert

@@ -2,7 +2,6 @@
 name: 04-audit
 description: Audit a codebase read-only across seven quality pillars into one ranked report. Use when the user wants to assess, health-check, or audit a codebase or one pillar. Not for fixing findings, reviewing a change, or checking a feature works.
 argument-hint: scope | pillar
-model: opus
 ---
 
 # Skill: audit

@@ -2,7 +2,6 @@
 name: 05-review
 description: Review a diff read-only on three axes, code, behavior versus the plan, and relevancy, into one verdict report. Use before shipping a change. Not for fixing findings or auditing a codebase.
 argument-hint: diff | plan
-model: opus
 ---
 
 # Skill: review

@@ -2,7 +2,6 @@
 name: 06-test
 description: Write and iterate developer-side tests until they pass, or validate a user journey end to end in the browser during implementation. Use when the user wants to add coverage, find what's untested, or walk a flow while building it. Do NOT use for independent acceptance QA or reviewer evidence, auditing test health, or debugging a failure.
 argument-hint: scope | journey
-model: sonnet
 ---
 
 # Skill: test

@@ -2,7 +2,6 @@
 name: 08-debug
 description: Reproduce and fix a known bug, or find an unknown root cause by hypothesis validation. Use when the user wants to fix a bug, find why something breaks, or reopen a stuck investigation. Not for building a feature or reviewing a diff.
 argument-hint: bug | symptom
-model: opus
 ---
 
 # Skill: debug

@@ -0,0 +1,63 @@
+const assert = require("node:assert/strict");
+const fs = require("node:fs");
+const os = require("node:os");
+const path = require("node:path");
+const { spawnSync } = require("node:child_process");
+const test = require("node:test");
+
+const ROOT = path.resolve(__dirname, "../..");
+
+test("bundled skills and agents leave model selection to the host", () => {
+  const plugins = path.join(ROOT, "plugins");
+  const files = fs.readdirSync(plugins, { recursive: true }).filter((file) =>
+    /(?:\/SKILL\.md$|\/agents\/[^/]+\.md$|\/assets\/agent-template\.md$)/.test(file.replaceAll(path.sep, "/"))
+  );
+  assert.ok(files.length > 0);
+  const overrides = files.filter((file) => {
+    const content = fs.readFileSync(path.join(plugins, file), "utf8");
+    const frontmatter = content.match(/^---\r?\n([\s\S]*?)\r?\n---/);
+    return frontmatter && /^model\s*:/m.test(frontmatter[1]);
+  });
+  assert.deepEqual(overrides, []);
+});
+
+for (const model of [undefined, "user-selected-model"]) {
+  test(`skill eval ${model ? "honors an explicit model" : "does not override the host model"}`, (t) => {
+    const tmp = fs.mkdtempSync(path.join(os.tmpdir(), "model-selection-"));
+    t.after(() => fs.rmSync(tmp, { recursive: true, force: true }));
+    const argsFile = path.join(tmp, "args.json");
+    const stub = path.join(tmp, "claude.cjs");
+    fs.writeFileSync(stub, [
+      'const fs = require("node:fs");',
+      'fs.writeFileSync(process.env.MODEL_SELECTION_ARGS, JSON.stringify(process.argv.slice(2)));',
+      'fs.writeFileSync("Makefile-shadow-report.md", "stub report");',
+    ].join("\n"));
+    const preload = path.join(tmp, "preload.cjs");
+    fs.writeFileSync(preload, [
+      'const childProcess = require("node:child_process");',
+      'const spawn = childProcess.spawn;',
+      `childProcess.spawn = (command, args, options) => command === "claude"`,
+      `  ? spawn(process.execPath, [${JSON.stringify(stub)}, ...args], options)`,
+      '  : spawn(command, args, options);',
+      'require("node:module").syncBuiltinESMExports();',
+    ].join("\n"));
+    const result = spawnSync(process.execPath, [
+      "--require", preload,
+      path.join(ROOT, "scripts/skill-eval.mjs"),
+      "--case=filename rule keeps a dotless name",
+      ...(model ? [`--model=${model}`] : []),
+    ], {
+      encoding: "utf8",
+      env: { ...process.env, MODEL_SELECTION_ARGS: argsFile },
+    });
+    assert.equal(result.status, 0, result.stderr + result.stdout);
+    const args = JSON.parse(fs.readFileSync(argsFile, "utf8"));
+    const at = args.indexOf("--model");
+    if (model) {
+      assert.notEqual(at, -1);
+      assert.equal(args[at + 1], model);
+    } else {
+      assert.equal(at, -1);
+    }
+  });
+}
@@ -18,7 +18,7 @@ const CASES = JSON.parse(readFileSync(join(HERE, "skill-eval", "cases.json"), "u
 const args = process.argv.slice(2);
 const JUDGE = args.includes("--judge");
 const KEEP = args.includes("--keep");
-const MODEL = (args.find((a) => a.startsWith("--model=")) || "--model=sonnet").split("=")[1];
+const MODEL = args.find((a) => a.startsWith("--model="))?.slice("--model=".length);
 const JOBS = Number((args.find((a) => a.startsWith("--jobs=")) || "--jobs=4").split("=")[1]);
 const filter = args.find((a) => !a.startsWith("--"));
 const REPEAT = Number((args.find((a) => a.startsWith("--repeat=")) || "--repeat=1").split("=")[1]);
@@ -40,7 +40,7 @@ function runClaude(prompt, cwd) {
     // settings (hooks, output modes) so results are reproducible across machines.
     const child = spawn(
       "claude",
-      ["-p", prompt, "--model", MODEL, "--setting-sources", "project,local", "--add-dir", cwd, "--dangerously-skip-permissions"],
+      ["-p", prompt, ...(MODEL ? ["--model", MODEL] : []), "--setting-sources", "project,local", "--add-dir", cwd, "--dangerously-skip-permissions"],
       { cwd, timeout: 600000 },
     );
     let out = "";
@@ -176,7 +176,7 @@ async function runAll(list, jobs) {
   return out;
 }
 
-console.log(`Running ${cases.length} case(s) on ${MODEL}, ${JOBS} at a time${JUDGE ? ", with --judge" : ""}.\n`);
+console.log(`Running ${cases.length} case(s) on ${MODEL || "the host default model"}, ${JOBS} at a time${JUDGE ? ", with --judge" : ""}.\n`);
 const results = await runAll(cases, JOBS);
 const failed = results.filter(({ checks }) => checks.some((k) => !k.ok)).length;
 

@@ -16,6 +16,8 @@ node scripts/skill-eval.mjs --keep          # keep temp dirs to inspect
 Local and opt-in. Needs an authenticated `claude` CLI and spends tokens, so it
 is not a CI gate.
 
+No model override is passed unless you supply `--model=<model-id>`.
+
 ## How it works
 
 Each case runs in a throwaway temp project. The skill under test is copied into