refactor: final Agent loader

README.md

@@ -216,6 +216,7 @@ For optimal coding assistance, we recommend using these top-tier models:
 
 ### Additional Configuration
 
+- **[Creating and Extending Agents](docs/creating-agents.md)** - Add custom agents or override bundled ones with TOML profiles
 - **[Agent Roles](docs/agent-roles.md)** - Create AI personas with specialized system prompts
 - **[Chat Summarization](docs/chat-summarization.md)** - Compress conversations to save context tokens
 - **[Ignoring Files](docs/ignoring-files.md)** - Exclude files from context using `.qodeassistignore`
@@ -531,7 +532,7 @@ If you find QodeAssist helpful, there are several ways you can support the proje
 
 1. **Report Issues**: If you encounter any bugs or have suggestions for improvements, please [open an issue](https://github.com/Palm1r/qodeassist/issues) on our GitHub repository.
 
-2. **Contribute**: Feel free to submit pull requests with bug fixes or new features.
+2. **Contribute**: Feel free to submit pull requests with bug fixes or new features. The easiest contribution is an agent preset for a provider or model you use — it's a single TOML file, no C++ required; see [Contributing your agent](docs/creating-agents.md#contributing-your-agent-to-qodeassist).
 
 3. **Spread the Word**: Star our GitHub repository and share QodeAssist with your fellow developers.
 
@@ -579,6 +580,10 @@ cmake --build .
 
 ## For Contributors
 
+### Adding an agent preset
+
+New provider/model presets are plain TOML — extend a provider base, register the file in `agents.qrc`, and the test suite validates it automatically. Step-by-step guide: [docs/creating-agents.md](docs/creating-agents.md#contributing-your-agent-to-qodeassist).
+
 ### Code Style
 
 - **QML**: Follow [QML Coding Guide](https://github.com/Furkanzmc/QML-Coding-Guide) by @Furkanzmc

docs/agent-templates-design.md

@@ -79,22 +79,31 @@ A missing/typo'd partial is a **load-time** error.
 ### 3. `extends` shares config down a hierarchy
 
 `extends` already exists (`resolveExtends` + `deepMerge` + `abstract`/`hidden`); it
-keeps doing what it does, now over the structured `[body]` too. A flat 2 levels — each
+keeps doing what it does, now over the structured `[body]` too. Each API-shape base
-provider base sets `system_prompt = """{{ agent_role("developer") }}"""` (the role text
+sets `system_prompt = """{{ agent_role() }}"""` (the role text comes from the role
-comes from the role JSON via the `agent_role` callback; see below). No shared root base:
+JSON via the `agent_role` callback; see below). No shared root base. Between the
+API-shape base and the concrete agents sits one thin abstract base **per provider**
+(provider_instance + endpoint only) — the designated extension point for user
+agents, so a custom agent is `extends` + `name` + `model`:
 
 ```
-openai_base (abstract)     → system_prompt + provider/endpoint/enable_tools + [body]
+openai_base (abstract)        → system_prompt + [body]   (API shape)
-  ├─ openai_chat           → name
+  ├─ mistral_base (abstract)  → provider, endpoint       (per-provider)
-  ├─ mistral_chat          → name, provider, endpoint
+  │   ├─ mistral_chat         → name, model
-  └─ mistral_reasoning     → + enable_thinking
+  │   └─ mistral_reasoning    → name, model + enable_thinking
-anthropic_base (abstract)  → system_prompt + provider/endpoint + [body]
+  ├─ openrouter_base (abstract) ...
-  ├─ claude_chat           → name
+  └─ openai_chat              → name, model              (own provider = no mid layer)
-  └─ claude_sonnet         → + [body] thinking / output_config
+anthropic_base (abstract)     → system_prompt + provider/endpoint + [body]
-google_base (abstract)     → system_prompt + provider + [body]
+  └─ claude_sonnet46          → name, model + [body] thinking / output_config
-  └─ gemini_chat           → endpoint (${MODEL}) + [body.generationConfig] thinkingConfig
+google_base (abstract)        → system_prompt + provider + [body]
+  └─ gemini_chat              → endpoint (${MODEL}) + [body.generationConfig] thinkingConfig
 ```
 
+Bundled agents are read-only: the loader rejects a user file that reuses a bundled
+`name`. Customisation = a user agent under a new name extending a bundled base (or a
+concrete bundled agent); the per-agent model override in settings covers the
+model-only case without any file.
+
 Notes:
 - `[body]` is shared whole when identical (the 8 OpenAI-compatible providers); a
   variant overrides only the differing field — no duplicated body.

docs/architecture.md

@@ -161,8 +161,13 @@ string values render and splice as raw JSON, literals pass through, empty render
 drop the key). `include` resolves only sandboxed partial roots. Profiles validate
 at load: a referenced partial must resolve and the assembled body must parse as
 JSON against a synthetic context — config errors surface in the agents settings
-page, never as a silent runtime drop. Full spec:
+page, never as a silent runtime drop. The loader also lints: unknown top-level /
-[`agent-templates-design.md`](agent-templates-design.md).
+`[match]` keys and same-layer duplicate names are warnings; a user file that
+reuses a bundled agent's name is rejected (bundled agents cannot be replaced —
+users extend them, or the per-provider abstract bases, under a new name);
+`abstract` and `hidden` are never inherited through `extends`. Full spec:
+[`agent-templates-design.md`](agent-templates-design.md); user-facing guide:
+[`creating-agents.md`](creating-agents.md).
 
 ---
 

docs/creating-agents.md

@@ -0,0 +1,339 @@
+# Creating and Extending Agents
+
+An *agent* is a TOML profile that tells QodeAssist which provider to call,
+which model to use, and exactly what request body to send. All bundled agents
+(Settings → QodeAssist → Agents) are built from the same files described here —
+anything a bundled agent does, a user agent can do too.
+
+## Where user agents live
+
+Drop `*.toml` files into the user agents directory:
+
+| OS | Path |
+|---|---|
+| Linux / macOS | `~/.config/QtProject/qtcreator/qodeassist/config/agents/` |
+| Windows | `%APPDATA%\QtProject\qtcreator\qodeassist\config\agents\` |
+
+QodeAssist creates the directory on startup. Files are loaded at plugin
+startup; after adding or editing a file, restart Qt Creator.
+
+Two layers are loaded:
+
+1. **Bundled** agents shipped inside the plugin — read-only.
+2. **User** agents from the directory above (marked with a `user` pill).
+
+Agent `name`s are global across both layers. A user file that reuses a
+bundled agent's `name` is rejected with an error — bundled agents cannot be
+replaced; create your own agent under a new name and `extends` what you want
+to build on. Two *user* files with the same `name` produce a warning, and
+the alphabetically later file wins.
+
+Load errors and warnings (TOML syntax, unknown keys, missing `extends`
+parents, bodies that don't render to valid JSON) are reported in Qt Creator's
+**General Messages** pane, prefixed with `[Agents]`.
+
+## Minimal example
+
+```toml
+schema_version = 1
+
+name        = "My DeepSeek Chat"
+description = "DeepSeek V3 via an OpenAI-compatible endpoint."
+
+provider_instance = "OpenAI Compatible"
+model             = "deepseek-chat"
+endpoint          = "/chat/completions"
+
+system_prompt = """{{ agent_role() }}"""
+
+[body]
+max_tokens  = 8192
+temperature = 0.7
+stream      = true
+messages    = """
+[ {% include "partials/openai_messages.jinja" %} ]
+"""
+```
+
+Shorter still — extend a bundled provider base and state only the delta:
+
+```toml
+schema_version = 1
+
+extends = "OpenAI Compatible Base Chat"
+name    = "My DeepSeek Chat"
+model   = "deepseek-chat"
+```
+
+Bundled agents themselves are read-only. To get a variant of a preset, create
+your own agent under a new name and put it where you want it in the roster:
+
+```toml
+schema_version = 1
+
+extends = "Mistral Base Chat"
+name    = "My Mistral (low temp)"
+model   = "mistral-small-latest"
+
+[body]
+temperature = 0.3
+```
+
+If all you want is a different model for a preset, you don't need a file at
+all — set the per-agent model override in the settings UI.
+
+## Key reference
+
+| Key | Required | Meaning |
+|---|---|---|
+| `schema_version` | no (default 1) | Format version; the plugin refuses files newer than it supports. |
+| `name` | yes | Unique identifier; shown in the UI, referenced by rosters and `extends`. |
+| `description` | no | Tooltip text in the Agents list. |
+| `provider_instance` | yes* | Name of a provider instance (see below). |
+| `model` | yes* | Default model; can be overridden per agent in settings. |
+| `endpoint` | yes* | Path appended to the provider instance URL. May contain `${MODEL}` (e.g. Google: `/models/${MODEL}:streamGenerateContent?alt=sse`). |
+| `system_prompt` | no | Jinja template for the system prompt (see building blocks below). FIM agents usually omit it. |
+| `tags` | no | Free-form strings shown as pills in the UI for discoverability. |
+| `enable_thinking` | no | Capability hint (UI badge). The `[body]` is the source of truth for what is sent. |
+| `enable_tools` | no | Lets the provider inject tool definitions into the request. |
+| `cache_prompt` / `cache_ttl` | no | Prompt caching (Anthropic); `cache_ttl = "1h"` selects the long TTL. |
+| `extends` | no | Name of a parent agent to inherit from. |
+| `abstract` | no | Mark as template-only: it can be extended but is never loaded as a usable agent. Not inherited. |
+| `hidden` | no | Loaded and usable, but not listed in selection UIs. Not inherited. |
+| `[match]` | no | Routing constraints (see Routing). |
+| `[body]` | yes* | The literal request body (see below). |
+
+\* required after `extends` resolution — a child inherits these from its
+parent, so it only states what differs.
+
+### Required keys checked at load
+
+A concrete (non-abstract) agent must end up with `name`,
+`provider_instance`, `model`, `endpoint`, and a non-empty `[body]`. Unknown
+keys anywhere at the top level or in `[match]` produce a warning — this
+catches typos like `enable_thinkin`.
+
+## Provider instances
+
+`provider_instance` refers to a provider configuration (URL + API key
+reference + client API). Bundled instances:
+
+`Claude`, `Codestral`, `Google AI`, `llama.cpp`,
+`LM Studio (Chat Completions)`, `LM Studio (Responses API)`, `Mistral AI`,
+`Ollama (Native)`, `Ollama (OpenAI-compatible)`, `OpenAI (Chat Completions)`,
+`OpenAI (Responses API)`, `OpenAI Compatible`, `OpenRouter`.
+
+User-defined instances live next to agents in
+`…/qodeassist/config/providers/*.toml` and follow the same
+override-by-name layering.
+
+## `extends` — inheriting from another agent
+
+A child deep-merges over its parent: scalar keys are replaced, tables (such
+as `[body]` and `[body.options]`) merge key-by-key, and **arrays are replaced
+whole** (a child that wants the parent's `tags` plus one more must restate
+the full list). Chains can be deeper than one level; cycles and missing
+parents are load errors.
+
+`abstract` and `hidden` are never inherited — extending a hidden agent
+yields a visible child unless the child says otherwise.
+
+Every provider ships an abstract base that already carries the provider
+instance, endpoint, and request body — extending one and setting `model` is
+usually all a custom agent needs:
+
+| Base | Provider / API |
+|---|---|
+| `Anthropic Base Chat` | Claude, Anthropic Messages (`/v1/messages`) |
+| `OpenAI Base Chat` | OpenAI, Chat Completions (`/chat/completions`) |
+| `OpenAI Responses Base` | OpenAI, Responses API (`/responses`) |
+| `OpenAI Compatible Base Chat` | Any OpenAI-compatible server |
+| `Google Base Chat` | Google AI, Gemini `generateContent` |
+| `Mistral Base Chat` | Mistral AI, Chat Completions |
+| `Codestral Base Chat` | Codestral, Chat Completions |
+| `Codestral FIM Base` | Mistral AI, `/v1/fim/completions` code completion |
+| `OpenRouter Base Chat` | OpenRouter, Chat Completions |
+| `LM Studio Base Chat` | LM Studio, Chat Completions |
+| `LM Studio Responses Base` | LM Studio, Responses API |
+| `llama.cpp Base Chat` | llama.cpp server, Chat Completions |
+| `Ollama Base Chat` | Ollama, native `/api/chat` |
+| `Ollama (OpenAI-compatible) Base Chat` | Ollama, OpenAI-compatible endpoint |
+| `Ollama FIM Base` | Ollama, native `/api/generate` fill-in-the-middle |
+
+Concrete agents work as parents too — `extends = "Claude Sonnet 4.6 Chat"`
+inherits everything including the model.
+
+## `[body]` — the request, literally
+
+`[body]` is the request body, written exactly like the provider's curl
+example. Per key, recursively:
+
+- **string containing jinja** (`{{` or `{%`) — rendered, and the output is
+  spliced in as raw JSON. A render that produces nothing drops the key.
+- **plain string / number / bool / table** — passed through unchanged.
+
+```toml
+[body]
+max_tokens  = 16000
+stream      = true
+thinking    = { type = "adaptive", display = "summarized" }
+messages    = """
+[ {% include "partials/anthropic_messages.jinja" %} ]
+"""
+```
+
+There are no runtime toggles: a thinking variant is a separate agent file
+that carries the thinking fields in its body.
+
+Every agent body is dry-run rendered at load against a synthetic
+conversation (text, thinking, tool calls, tool results, images), so jinja
+syntax errors, unknown callbacks, missing partials, and invalid JSON are
+reported at startup — not mid-conversation. Trailing commas emitted by loops
+are stripped automatically; don't bother with `loop.is_last` bookkeeping.
+
+### Template data (`ctx`)
+
+| Field | Content |
+|---|---|
+| `ctx.system_prompt` | Rendered system prompt (present only if the agent has one). |
+| `ctx.prefix` / `ctx.suffix` | Code around the cursor (FIM/completion sessions). |
+| `ctx.files_metadata` | Array of `{ file_path, content }` for attached files. |
+| `ctx.history` | Array of messages: `{ role, content, content_blocks, images? }`. |
+
+`content` is the message's flattened text; `content_blocks` is the
+structured form:
+
+| `type` | Fields |
+|---|---|
+| `text` | `text` |
+| `thinking` | `thinking`, `signature` |
+| `redacted_thinking` | `data` |
+| `tool_use` | `id`, `name`, `input` (JSON object) |
+| `tool_result` | `tool_use_id`, `content`, `name` |
+| `image` | `data`, `media_type`, `is_url` |
+
+### Callbacks available in `[body]`
+
+| Callback | Purpose |
+|---|---|
+| `tojson(x)` | Serialize any value as JSON (correct quoting/escaping). Use it for every interpolated value. |
+| `filter_by_type(blocks, "tool_use")` | Subset of `content_blocks` with the given type. |
+| `filter_skip_role(history, "system")` | History without messages of a role. |
+| `strip_signature_suffix(s)` | Remove a trailing `[Signature: …]` marker. |
+
+### Partials and `{% include %}`
+
+The repetitive message-array rendering lives in shared partials. Includes
+resolve against the bundled set first, then the user agent's own directory —
+so a user agent can both reuse bundled partials and ship its own next to its
+TOML (e.g. `partials/my_messages.jinja`). Paths with `..` or a leading `/`
+are rejected.
+
+Bundled partials: `partials/openai_messages.jinja`,
+`partials/openai_responses_input.jinja`, `partials/anthropic_messages.jinja`,
+`partials/google_contents.jinja`, `partials/ollama_messages.jinja` (plus the
+per-message helpers they include).
+
+## `system_prompt` — composable building blocks
+
+`system_prompt` is itself a jinja template, rendered with:
+
+| Helper | Purpose |
+|---|---|
+| `{{ agent_role() }}` | Text of the role currently selected in the chat (falls back to `developer`). |
+| `{{ agent_role("reviewer") }}` | A specific role by id (Settings → QodeAssist → Roles). |
+| `{{ read_file("${PROJECT_DIR}/STYLE.md") }}` | Inline a file. Reads are restricted to the project directory and `~/qodeassist`. |
+| `{{ file_exists(p) }}` / `{{ read_dir(p) }}` | Existence check / directory listing (same root restrictions). |
+| `{{ head_lines(s, n) }}` | First `n` lines of a string. |
+| `basename`, `dirname`, `ext`, `lower`, `upper` | Path/string helpers. |
+| `${PROJECT_DIR}`, `${HOME}` | Substituted before rendering. |
+
+Example:
+
+```toml
+system_prompt = """
+{{ agent_role() }}
+
+{% if file_exists("${PROJECT_DIR}/.qodeassist-style.md") %}
+Project conventions:
+{{ read_file("${PROJECT_DIR}/.qodeassist-style.md") }}
+{% endif %}
+"""
+```
+
+## Routing — `[match]` and rosters
+
+Each pipeline (code completion, chat, compression, quick refactor) has an
+ordered roster of agents. For the current file, the **first roster entry
+whose `[match]` accepts** wins.
+
+```toml
+[match]
+file_patterns = ["*.qml", "*.js"]
+path_patterns = ["*/tests/*"]
+project_names = ["MyProject"]
+```
+
+- Dimensions are ANDed; an empty dimension is unconstrained; an entirely
+  empty/absent `[match]` is a catch-all.
+- `file_patterns` are case-insensitive globs tested against the file name
+  and the full path; `path_patterns` against the full path only.
+- `project_names` are exact, case-sensitive project names.
+
+Typical setup: a specialized agent (e.g. `Qt CodeLlama 13B QML FIM` with
+`*.qml`) first, a catch-all agent last.
+
+## Models
+
+The TOML `model` is only the default. The settings UI can set a per-agent
+override (stored in `agent_models.json`); the resolved model is also
+substituted into `${MODEL}` in `endpoint` before sending.
+
+## Contributing your agent to QodeAssist
+
+The bundled agent set grows through contributions — if you've made an agent
+for a provider or model that others could use, please send it upstream
+instead of keeping it local. No C++ is needed:
+
+1. Develop and verify the agent locally in the user agents directory.
+2. In a fork, copy the TOML to `sources/agents/` and register the file in
+   `sources/agents/agents.qrc`.
+3. Keep it a thin delta: extend the matching provider base and set only
+   `name`, `description`, `model`, `tags` (and `[body]` keys that genuinely
+   differ). Look at `mistral_chat.toml` or `ollama_qwen25_coder_fim.toml`
+   for the expected shape.
+4. Run the tests (`QodeAssistTest`): `BundledAgentsTest` automatically
+   loads every bundled agent, resolves its `extends` chain, and dry-renders
+   its `[body]` — if your TOML passes, it works.
+5. Open a pull request.
+
+Conventions:
+
+- File name: `<provider>_<model_or_purpose>_<kind>.toml`
+  (e.g. `ollama_qwen25_coder_fim.toml`).
+- `name` is user-visible and must be unique; include the provider and model
+  (e.g. `Ollama Qwen2.5-Coder FIM`).
+- Specialized completion agents should carry a `[match]` block so routing
+  can pick them automatically (e.g. `file_patterns = ["*.qml"]`).
+- A whole new provider with an OpenAI-compatible API is also TOML-only: a
+  provider instance file in `sources/providersConfig/`, one abstract
+  `<Provider> Base Chat`, and concrete agents on top. New request/response
+  *formats* are the only thing that needs C++.
+
+## Troubleshooting
+
+- **Agent missing from the list** — check General Messages for `[Agents]
+  error:` lines; the file failed to parse, resolve, or validate.
+- **`… has the same name as a bundled agent — bundled agents cannot be
+  replaced`** — pick a different `name`; use `extends` to inherit from the
+  bundled agent instead.
+- **`Unknown key 'x' … ignored (typo?)`** — the key isn't part of the
+  schema; compare with the table above.
+- **`Agent 'X' extends unknown agent 'Y'`** — the parent's `name` (not file
+  name) must match exactly; the parent must be bundled or in the same
+  directory.
+- **`[body] failed to render to valid JSON`** — the dry run failed; the log
+  contains the rendered snippet. Usually a missing `tojson(...)` around an
+  interpolated string.
+- **Edits not picked up** — agents are loaded at startup; restart
+  Qt Creator.

sources/agents/AgentConfig.hpp

@@ -48,7 +48,6 @@ struct AgentConfig
     bool hidden = false;
 
     QString sourcePath;
-    bool overridesBundled = false;
     bool isUserSource() const { return !sourcePath.startsWith(QLatin1StringView{":/"}); }
 
     static QString validate(const AgentConfig &config);

sources/agents/AgentFactory.cpp

@@ -59,6 +59,7 @@ void AgentFactory::reload()
     Q_ASSERT(thread() == QThread::currentThread());
     clear();
 
+    QDir().mkpath(userAgentsDir());
     auto result = Agents::AgentLoader::load(agentQrcPrefix(), userAgentsDir());
     for (const QString &err : result.errors)
         LOG_MESSAGE(QString("[Agents] error: %1").arg(err));

sources/agents/AgentLoader.cpp

@@ -145,16 +145,49 @@ struct RawEntry
 {
     QJsonObject obj;
     QString filePath;
-    bool overridesBundled = false;
+    bool isUserLayer = false;
 };
 
 constexpr int kMaxExtendsDepth = 32;
 
+void lintUnknownKeys(const QJsonObject &obj, const QString &filePath, QStringList &warnings)
+{
+    static const QSet<QString> kTopLevelKeys = {
+        QStringLiteral("schema_version"), QStringLiteral("name"),
+        QStringLiteral("description"),    QStringLiteral("provider_instance"),
+        QStringLiteral("model"),          QStringLiteral("endpoint"),
+        QStringLiteral("system_prompt"),  QStringLiteral("tags"),
+        QStringLiteral("match"),          QStringLiteral("enable_thinking"),
+        QStringLiteral("enable_tools"),   QStringLiteral("cache_prompt"),
+        QStringLiteral("cache_ttl"),      QStringLiteral("body"),
+        QStringLiteral("extends"),        QStringLiteral("abstract"),
+        QStringLiteral("hidden")};
+    static const QSet<QString> kMatchKeys = {
+        QStringLiteral("file_patterns"),
+        QStringLiteral("path_patterns"),
+        QStringLiteral("project_names")};
+
+    for (auto it = obj.constBegin(); it != obj.constEnd(); ++it) {
+        if (!kTopLevelKeys.contains(it.key())) {
+            warnings.append(QStringLiteral("Unknown key '%1' in %2 — ignored (typo?)")
+                                .arg(it.key(), filePath));
+        }
+    }
+    const QJsonObject matchObj = obj.value("match").toObject();
+    for (auto it = matchObj.constBegin(); it != matchObj.constEnd(); ++it) {
+        if (!kMatchKeys.contains(it.key())) {
+            warnings.append(QStringLiteral("Unknown key 'match.%1' in %2 — ignored (typo?)")
+                                .arg(it.key(), filePath));
+        }
+    }
+}
+
 void scanDir(
     const QString &dir,
     bool isUserLayer,
     QHash<QString, RawEntry> &raw,
-    QStringList &errors)
+    QStringList &errors,
+    QStringList *warnings)
 {
     if (dir.isEmpty()) return;
     QDir d(dir);
@@ -173,11 +206,39 @@ void scanDir(
             errors.append(QStringLiteral("Agent at %1 has no 'name'").arg(fullPath));
             continue;
         }
-        const bool overrides = isUserLayer && raw.contains(name);
+        if (warnings)
-        raw.insert(name, {*objOpt, fullPath, overrides});
+            lintUnknownKeys(*objOpt, fullPath, *warnings);
+        const auto existing = raw.constFind(name);
+        if (existing != raw.constEnd() && existing->isUserLayer != isUserLayer) {
+            errors.append(
+                QStringLiteral("Agent '%1' at %2 has the same name as a bundled agent — "
+                               "bundled agents cannot be replaced; rename it and use "
+                               "'extends' to build on the bundled one")
+                    .arg(name, fullPath));
+            continue;
+        }
+        if (warnings && existing != raw.constEnd()) {
+            warnings->append(
+                QStringLiteral("Agent '%1' is defined in both %2 and %3 — %3 wins")
+                    .arg(name, existing->filePath, fullPath));
+        }
+        raw.insert(name, {*objOpt, fullPath, isUserLayer});
     }
 }
 
+QJsonObject mergeChild(const QJsonObject &parentMerged, const QJsonObject &self, const QString &name)
+{
+    QJsonObject merged = deepMerge(parentMerged, self);
+    merged["name"] = name;
+    for (const QString &key : {QStringLiteral("abstract"), QStringLiteral("hidden")}) {
+        if (self.contains(key))
+            merged[key] = self.value(key);
+        else
+            merged.remove(key);
+    }
+    return merged;
+}
+
 QJsonObject resolveExtends(
     const QString &name,
     const QHash<QString, RawEntry> &raw,
@@ -196,7 +257,7 @@ QJsonObject resolveExtends(
         return {};
     }
     if (!raw.contains(name)) {
-        errors.append(QStringLiteral("Unknown parent agent '%1'").arg(name));
+        errors.append(QStringLiteral("Unknown agent '%1'").arg(name));
         return {};
     }
     visiting.insert(name);
@@ -204,15 +265,15 @@ QJsonObject resolveExtends(
     QJsonObject self = raw.value(name).obj;
     const QString parent = self.value("extends").toString();
     if (!parent.isEmpty()) {
+        if (!raw.contains(parent)) {
+            errors.append(QStringLiteral("Agent '%1' extends unknown agent '%2' (%3)")
+                              .arg(name, parent, raw.value(name).filePath));
+            visiting.remove(name);
+            return {};
+        }
         const QJsonObject parentMerged
             = resolveExtends(parent, raw, visiting, errors, depth + 1);
-        QJsonObject merged = deepMerge(parentMerged, self);
+        self = mergeChild(parentMerged, self, name);
-        merged["name"] = name;
-        if (self.contains("abstract"))
-            merged["abstract"] = self.value("abstract");
-        else
-            merged.remove("abstract");
-        self = merged;
     }
     visiting.remove(name);
     return self;
@@ -224,7 +285,7 @@ std::optional<AgentConfig> AgentLoader::parseFile(
     const QString &path,
     const QString &qrcPrefix,
     QString *error,
-    QStringList * /*warnings*/)
+    QStringList *warnings)
 {
     auto objOpt = parseTomlFile(path, error);
     if (!objOpt) return std::nullopt;
@@ -234,12 +295,14 @@ std::optional<AgentConfig> AgentLoader::parseFile(
         if (error) *error = QStringLiteral("Agent at %1 has no 'name'").arg(path);
         return std::nullopt;
     }
+    if (warnings)
+        lintUnknownKeys(*objOpt, path, *warnings);
 
     QHash<QString, RawEntry> raw;
     QStringList scanErrors;
-    scanDir(qrcPrefix, /*isUserLayer=*/false, raw, scanErrors);
+    scanDir(qrcPrefix, /*isUserLayer=*/false, raw, scanErrors, nullptr);
-    scanDir(QFileInfo(path).absolutePath(), /*isUserLayer=*/true, raw, scanErrors);
+    scanDir(QFileInfo(path).absolutePath(), /*isUserLayer=*/true, raw, scanErrors, nullptr);
-    raw.insert(name, {*objOpt, path, raw.contains(name)});
+    raw.insert(name, {*objOpt, path, true});
 
     QSet<QString> visiting;
     QStringList resolveErrors;
@@ -270,8 +333,8 @@ AgentLoader::LoadResult AgentLoader::load(const QString &qrcPrefix, const QStrin
     LoadResult result;
     QHash<QString, RawEntry> raw;
 
-    scanDir(qrcPrefix, /*isUserLayer=*/false, raw, result.errors);
+    scanDir(qrcPrefix, /*isUserLayer=*/false, raw, result.errors, &result.warnings);
-    scanDir(userDir,   /*isUserLayer=*/true,  raw, result.errors);
+    scanDir(userDir,   /*isUserLayer=*/true,  raw, result.errors, &result.warnings);
 
     for (auto it = raw.constBegin(); it != raw.constEnd(); ++it) {
         const QString &name = it.key();
@@ -282,7 +345,6 @@ AgentLoader::LoadResult AgentLoader::load(const QString &qrcPrefix, const QStrin
 
         AgentConfig cfg = configFromMerged(merged);
         cfg.sourcePath = it.value().filePath;
-        cfg.overridesBundled = it.value().overridesBundled;
 
         if (cfg.abstract) continue;
 

sources/agents/AgentRouter.cpp

@@ -5,6 +5,9 @@
 #include "AgentRouter.hpp"
 
 #include <QFileInfo>
+#include <QHash>
+#include <QMutex>
+#include <QMutexLocker>
 #include <QRegularExpression>
 
 #include "AgentFactory.hpp"
@@ -13,16 +16,29 @@ namespace QodeAssist::AgentRouter {
 
 namespace {
 
+QRegularExpression compiledGlob(const QString &pattern)
+{
+    static QHash<QString, QRegularExpression> cache;
+    static QMutex mutex;
+    QMutexLocker lock(&mutex);
+    const auto it = cache.constFind(pattern);
+    if (it != cache.constEnd())
+        return *it;
+    const QRegularExpression re(
+        QRegularExpression::anchoredPattern(
+            QRegularExpression::wildcardToRegularExpression(
+                pattern, QRegularExpression::NonPathWildcardConversion)),
+        QRegularExpression::CaseInsensitiveOption);
+    cache.insert(pattern, re);
+    return re;
+}
+
 bool matchesAnyGlob(const QStringList &patterns, const QString &subject)
 {
     if (subject.isEmpty())
         return false;
     for (const QString &pat : patterns) {
-        const QRegularExpression re(
+        const QRegularExpression re = compiledGlob(pat);
-            QRegularExpression::anchoredPattern(
-                QRegularExpression::wildcardToRegularExpression(
-                    pat, QRegularExpression::NonPathWildcardConversion)),
-            QRegularExpression::CaseInsensitiveOption);
         if (re.isValid() && re.match(subject).hasMatch())
             return true;
     }

sources/agents/agents.qrc

@@ -18,6 +18,15 @@
         <file>google_base_chat.toml</file>
         <file>ollama_base_chat.toml</file>
         <file>ollama_base_fim.toml</file>
+        <file>mistral_base_chat.toml</file>
+        <file>codestral_base_chat.toml</file>
+        <file>codestral_fim_base.toml</file>
+        <file>openrouter_base_chat.toml</file>
+        <file>lmstudio_base_chat.toml</file>
+        <file>lmstudio_responses_base.toml</file>
+        <file>llamacpp_base_chat.toml</file>
+        <file>ollama_openai_base_chat.toml</file>
+        <file>openai_compatible_base_chat.toml</file>
 
         <file>openai_chat.toml</file>
         <file>openai_compatible_chat.toml</file>
@@ -35,8 +44,8 @@
         <file>ollama_gemma4_e4b_chat.toml</file>
         <file>ollama_codellama_7b_code_fim.toml</file>
         <file>ollama_codellama_13b_qml_fim.toml</file>
+        <file>ollama_qwen25_coder_fim.toml</file>
 
-        <file>claude_sonnet_chat.toml</file>
         <file>claude_sonnet46_chat.toml</file>
         <file>claude_haiku45_chat.toml</file>
         <file>claude_opus_max.toml</file>

sources/agents/claude_sonnet_chat.toml

@@ -1,14 +0,0 @@
-schema_version = 1
-
-extends     = "Anthropic Base Chat"
-name        = "Claude Sonnet Chat"
-description = "Anthropic Claude Sonnet 4.6 (Messages API) — coding chat assistant with thinking."
-
-model = "claude-sonnet-4-6"
-
-enable_thinking = true
-tags = ["chat", "claude", "anthropic", "cloud"]
-
-[body]
-max_tokens = 8192
-thinking   = { type = "enabled", budget_tokens = 4096 }

sources/agents/codestral_base_chat.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "Codestral Base Chat"
+description = "Mistral Codestral Chat Completions. Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "Codestral"
+endpoint          = "/v1/chat/completions"
+
+tags = ["chat", "codestral", "mistral", "cloud"]

sources/agents/codestral_chat.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "Codestral Base Chat"
 name        = "Codestral Chat"
 description = "Mistral Codestral (Chat Completions API) — coding chat assistant."
 
-provider_instance = "Codestral"
+model = "codestral-latest"
-endpoint          = "/v1/chat/completions"
-model             = "codestral-latest"
-
-tags = ["chat", "codestral", "mistral", "cloud"]

sources/agents/codestral_fim.toml

@@ -1,19 +1,9 @@
 schema_version = 1
 
+extends     = "Codestral FIM Base"
 name        = "Codestral FIM"
 description = "Mistral Codestral fill-in-the-middle code completion (/v1/fim/completions)."
 
-provider_instance = "Mistral AI"
+model = "codestral-latest"
-endpoint          = "/v1/fim/completions"
-model             = "codestral-latest"
 
-enable_thinking = false
-enable_tools    = false
 tags = ["fim", "codestral", "mistral", "cloud", "completion"]
-
-[body]
-max_tokens  = 256
-temperature = 0.2
-stream      = true
-prompt      = """{{ tojson(ctx.prefix) }}"""
-suffix      = """{% if existsIn(ctx, "suffix") %}{{ tojson(ctx.suffix) }}{% endif %}"""

sources/agents/codestral_fim_base.toml

@@ -0,0 +1,17 @@
+schema_version = 1
+
+name        = "Codestral FIM Base"
+description = "Mistral fill-in-the-middle completion (/v1/fim/completions). Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "Mistral AI"
+endpoint          = "/v1/fim/completions"
+
+tags = ["fim", "mistral", "cloud", "completion"]
+
+[body]
+max_tokens  = 256
+temperature = 0.2
+stream      = true
+prompt      = """{{ tojson(ctx.prefix) }}"""
+suffix      = """{% if existsIn(ctx, "suffix") %}{{ tojson(ctx.suffix) }}{% endif %}"""

sources/agents/llamacpp_base_chat.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "llama.cpp Base Chat"
+description = "llama.cpp server (OpenAI-compatible Chat Completions). Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "llama.cpp"
+endpoint          = "/v1/chat/completions"
+
+tags = ["chat", "llamacpp", "local"]

sources/agents/llamacpp_chat.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "llama.cpp Base Chat"
 name        = "llama.cpp Chat"
 description = "llama.cpp server (OpenAI-compatible Chat Completions) — local coding chat assistant."
 
-provider_instance = "llama.cpp"
+model = "llama"
-endpoint          = "/v1/chat/completions"
-model             = "llama"
-
-tags = ["chat", "llamacpp", "local"]

sources/agents/lmstudio_base_chat.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "LM Studio Base Chat"
+description = "LM Studio Chat Completions. Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "LM Studio (Chat Completions)"
+endpoint          = "/v1/chat/completions"
+
+tags = ["chat", "lmstudio", "local"]

sources/agents/lmstudio_chat.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "LM Studio Base Chat"
 name        = "LM Studio Chat"
 description = "LM Studio (Chat Completions API) — local coding chat assistant."
 
-provider_instance = "LM Studio (Chat Completions)"
+model = "local-model"
-endpoint          = "/v1/chat/completions"
-model             = "local-model"
-
-tags = ["chat", "lmstudio", "local"]

sources/agents/lmstudio_responses.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Responses Base"
+extends     = "LM Studio Responses Base"
 name        = "LM Studio Responses"
 description = "LM Studio (Responses API) — local coding chat assistant."
 
-provider_instance = "LM Studio (Responses API)"
+model = "local-model"
-endpoint          = "/v1/responses"
-model             = "local-model"
-
-tags = ["chat", "lmstudio", "responses", "local"]

sources/agents/lmstudio_responses_base.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Responses Base"
+name        = "LM Studio Responses Base"
+description = "LM Studio Responses API. Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "LM Studio (Responses API)"
+endpoint          = "/v1/responses"
+
+tags = ["chat", "lmstudio", "responses", "local"]

sources/agents/mistral_base_chat.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "Mistral Base Chat"
+description = "Mistral AI Chat Completions. Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "Mistral AI"
+endpoint          = "/v1/chat/completions"
+
+tags = ["chat", "mistral", "cloud"]

sources/agents/mistral_chat.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "Mistral Base Chat"
 name        = "Mistral Chat"
 description = "Mistral Large (Chat Completions API) — coding chat assistant."
 
-provider_instance = "Mistral AI"
+model = "mistral-large-latest"
-endpoint          = "/v1/chat/completions"
-model             = "mistral-large-latest"
-
-tags = ["chat", "mistral", "cloud"]

sources/agents/mistral_medium_chat.toml

@@ -1,11 +1,9 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "Mistral Base Chat"
 name        = "Mistral Medium Chat"
 description = "Mistral Medium 3.5 (Chat Completions API) — frontier coding/agentic chat."
 
-provider_instance = "Mistral AI"
+model = "mistral-medium-latest"
-endpoint          = "/v1/chat/completions"
-model             = "mistral-medium-latest"
 
 tags = ["chat", "mistral", "medium", "cloud"]

sources/agents/mistral_reasoning_chat.toml

@@ -1,12 +1,10 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "Mistral Base Chat"
 name        = "Mistral Reasoning Chat"
 description = "Mistral Magistral Medium — native chain-of-thought reasoning model."
 
-provider_instance = "Mistral AI"
+model = "magistral-medium-latest"
-endpoint          = "/v1/chat/completions"
-model             = "magistral-medium-latest"
 
 enable_thinking = true
 tags = ["chat", "mistral", "reasoning", "cloud"]

sources/agents/ollama_openai_base_chat.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "Ollama (OpenAI-compatible) Base Chat"
+description = "Ollama via its OpenAI-compatible Chat Completions endpoint. Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "Ollama (OpenAI-compatible)"
+endpoint          = "/v1/chat/completions"
+
+tags = ["chat", "ollama", "local"]

sources/agents/ollama_openai_chat.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "Ollama (OpenAI-compatible) Base Chat"
 name        = "Ollama (OpenAI-compatible) Chat"
 description = "Ollama via its OpenAI-compatible Chat Completions endpoint — local coding chat assistant."
 
-provider_instance = "Ollama (OpenAI-compatible)"
+model = "qwen2.5-coder"
-endpoint          = "/v1/chat/completions"
-model             = "qwen2.5-coder"
-
-tags = ["chat", "ollama", "local"]

sources/agents/ollama_qwen25_coder_fim.toml

@@ -0,0 +1,15 @@
+schema_version = 1
+
+extends     = "Ollama FIM Base"
+name        = "Ollama Qwen2.5-Coder FIM"
+description = "Local Qwen2.5-Coder on Ollama, FIM completion via <|fim_*|> markers."
+
+model = "qwen2.5-coder"
+
+tags = ["fim", "ollama", "local", "qwen"]
+
+[body]
+prompt = """{% if existsIn(ctx, "suffix") and length(ctx.suffix) > 0 %}{{ tojson("<|fim_prefix|>" + ctx.prefix + "<|fim_suffix|>" + ctx.suffix + "<|fim_middle|>") }}{% else %}{{ tojson("<|fim_prefix|>" + ctx.prefix + "<|fim_middle|>") }}{% endif %}"""
+
+[body.options]
+stop = ["<|endoftext|>", "<|fim_prefix|>", "<|fim_suffix|>", "<|fim_middle|>", "<|fim_pad|>", "<|repo_name|>", "<|file_sep|>"]

sources/agents/openai_compatible_base_chat.toml

@@ -0,0 +1,10 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "OpenAI Compatible Base Chat"
+description = "Any OpenAI-compatible Chat Completions endpoint. Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "OpenAI Compatible"
+
+tags = ["chat", "openai", "compatible"]

sources/agents/openai_compatible_chat.toml

@@ -1,10 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "OpenAI Compatible Base Chat"
 name        = "OpenAI Compatible Chat"
 description = "Any OpenAI-compatible Chat Completions endpoint — set the model to match your server."
 
-provider_instance = "OpenAI Compatible"
+model = "default"
-model             = "default"
-
-tags = ["chat", "openai", "compatible"]

sources/agents/openrouter_base_chat.toml

@@ -0,0 +1,11 @@
+schema_version = 1
+
+extends     = "OpenAI Base Chat"
+name        = "OpenRouter Base Chat"
+description = "OpenRouter (OpenAI-compatible Chat Completions). Abstract — extend it and set model."
+abstract    = true
+
+provider_instance = "OpenRouter"
+endpoint          = "/chat/completions"
+
+tags = ["chat", "openrouter", "cloud"]

sources/agents/openrouter_chat.toml

@@ -1,11 +1,7 @@
 schema_version = 1
 
-extends     = "OpenAI Base Chat"
+extends     = "OpenRouter Base Chat"
 name        = "OpenRouter Chat"
 description = "OpenRouter (OpenAI-compatible Chat Completions) — coding chat assistant."
 
-provider_instance = "OpenRouter"
+model = "openai/gpt-4o"
-endpoint          = "/chat/completions"
-model             = "openai/gpt-4o"
-
-tags = ["chat", "openrouter", "cloud"]

sources/settings/AgentSelectionDialog.cpp

@@ -147,10 +147,7 @@ AgentRowCard::AgentRowCard(const AgentConfig &cfg, QWidget *parent)
 
     Pill *sourcePill = nullptr;
     if (cfg.isUserSource()) {
-        sourcePill = new Pill(
+        sourcePill = new Pill(Pill::User, Tr::tr("User"), this);
-            Pill::User,
-            cfg.overridesBundled ? Tr::tr("Override") : Tr::tr("User"),
-            this);
     }
 
     auto *description = new QLabel(this);

sources/settings/AgentSlotWidget.cpp

@@ -296,9 +296,7 @@ void AgentSlotWidget::setAgentConfig(const AgentConfig &cfg)
     m_name->setText(cfg.name);
 
     if (cfg.isUserSource()) {
-        m_sourcePill->setText(cfg.overridesBundled
+        m_sourcePill->setText(Tr::tr("User"));
-                                  ? Tr::tr("User overrides bundled")
-                                  : Tr::tr("User"));
         m_sourcePill->show();
     } else {
         m_sourcePill->hide();

sources/settings/PipelinesConfig.cpp

@@ -95,10 +95,10 @@ void fillMissingFromDefaults(PipelineRosters &r, const toml::table &section)
 PipelineRosters PipelineRosters::defaults()
 {
     PipelineRosters r;
-    r.codeCompletion = {QStringLiteral("Ollama Qwen2.5-Coder Completion")};
+    r.codeCompletion = {QStringLiteral("Ollama Qwen2.5-Coder FIM")};
-    r.chatAssistant = {QStringLiteral("Ollama Chat")};
+    r.chatAssistant = {QStringLiteral("Ollama (OpenAI-compatible) Chat")};
-    r.chatCompression = {QStringLiteral("Ollama Compression")};
+    r.chatCompression = {QStringLiteral("Ollama (OpenAI-compatible) Chat")};
-    r.quickRefactor = {QStringLiteral("Ollama Quick Refactor")};
+    r.quickRefactor = {QStringLiteral("Ollama (OpenAI-compatible) Chat")};
     return r;
 }
 

test/AgentLoaderTest.cpp

@@ -0,0 +1,184 @@
+// Copyright (C) 2024-2026 Petr Mironychev
+// SPDX-License-Identifier: GPL-3.0-or-later
+// Additional attribution terms under GPLv3 §7(b) apply — see LICENSE
+
+#include <gtest/gtest.h>
+
+#include <QDir>
+#include <QFile>
+#include <QTemporaryDir>
+
+#include <AgentConfig.hpp>
+#include <AgentLoader.hpp>
+
+using QodeAssist::AgentConfig;
+using QodeAssist::Agents::AgentLoader;
+
+namespace {
+
+void writeFile(const QString &dir, const QString &name, const QByteArray &contents)
+{
+    QFile f(dir + QLatin1Char('/') + name);
+    ASSERT_TRUE(f.open(QIODevice::WriteOnly | QIODevice::Text));
+    f.write(contents);
+}
+
+QByteArray minimalAgent(const QByteArray &name, const QByteArray &extra = {})
+{
+    return "name = \"" + name + "\"\n"
+           "provider_instance = \"P\"\n"
+           "model = \"m\"\n"
+           "endpoint = \"/e\"\n"
+           + extra +
+           "[body]\n"
+           "stream = true\n";
+}
+
+const AgentConfig *findConfig(const AgentLoader::LoadResult &result, const QString &name)
+{
+    for (const auto &cfg : result.configs) {
+        if (cfg.name == name)
+            return &cfg;
+    }
+    return nullptr;
+}
+
+bool anyContains(const QStringList &list, const QString &needle)
+{
+    for (const QString &s : list) {
+        if (s.contains(needle))
+            return true;
+    }
+    return false;
+}
+
+} // namespace
+
+TEST(AgentLoaderTest, WarnsOnUnknownTopLevelAndMatchKeys)
+{
+    QTemporaryDir dir;
+    ASSERT_TRUE(dir.isValid());
+    writeFile(dir.path(), "a.toml",
+              minimalAgent("A",
+                           "enable_thinkin = true\n"
+                           "[match]\n"
+                           "file_pattern = [\"*.cpp\"]\n"));
+
+    const auto result = AgentLoader::load(QString(), dir.path());
+    EXPECT_TRUE(result.errors.isEmpty()) << result.errors.join("; ").toStdString();
+    EXPECT_TRUE(anyContains(result.warnings, QStringLiteral("enable_thinkin")));
+    EXPECT_TRUE(anyContains(result.warnings, QStringLiteral("match.file_pattern")));
+}
+
+TEST(AgentLoaderTest, WarnsOnDuplicateNameInSameLayer)
+{
+    QTemporaryDir dir;
+    ASSERT_TRUE(dir.isValid());
+    writeFile(dir.path(), "first.toml", minimalAgent("Dup"));
+    writeFile(dir.path(), "second.toml", minimalAgent("Dup"));
+
+    const auto result = AgentLoader::load(QString(), dir.path());
+    EXPECT_TRUE(anyContains(result.warnings, QStringLiteral("defined in both")));
+    const AgentConfig *cfg = findConfig(result, QStringLiteral("Dup"));
+    ASSERT_NE(cfg, nullptr);
+    EXPECT_TRUE(cfg->sourcePath.endsWith(QStringLiteral("second.toml")));
+}
+
+TEST(AgentLoaderTest, UserAgentCollidingWithBundledNameIsRejected)
+{
+    QTemporaryDir bundled;
+    QTemporaryDir user;
+    ASSERT_TRUE(bundled.isValid());
+    ASSERT_TRUE(user.isValid());
+    writeFile(bundled.path(), "a.toml", minimalAgent("A", "description = \"base\"\n"));
+    writeFile(user.path(), "a.toml", minimalAgent("A", "description = \"mine\"\n"));
+
+    const auto result = AgentLoader::load(bundled.path(), user.path());
+    EXPECT_TRUE(anyContains(result.errors, QStringLiteral("cannot be replaced")));
+    const AgentConfig *cfg = findConfig(result, QStringLiteral("A"));
+    ASSERT_NE(cfg, nullptr);
+    EXPECT_EQ(cfg->description, QStringLiteral("base"));
+    EXPECT_FALSE(cfg->isUserSource());
+}
+
+TEST(AgentLoaderTest, HiddenIsNotInherited)
+{
+    QTemporaryDir dir;
+    ASSERT_TRUE(dir.isValid());
+    writeFile(dir.path(), "parent.toml", minimalAgent("Parent", "hidden = true\n"));
+    writeFile(dir.path(), "child.toml", minimalAgent("Child", "extends = \"Parent\"\n"));
+
+    const auto result = AgentLoader::load(QString(), dir.path());
+    EXPECT_TRUE(result.errors.isEmpty()) << result.errors.join("; ").toStdString();
+    const AgentConfig *parent = findConfig(result, QStringLiteral("Parent"));
+    const AgentConfig *child = findConfig(result, QStringLiteral("Child"));
+    ASSERT_NE(parent, nullptr);
+    ASSERT_NE(child, nullptr);
+    EXPECT_TRUE(parent->hidden);
+    EXPECT_FALSE(child->hidden);
+}
+
+TEST(AgentLoaderTest, UserAgentExtendsBundledProviderBase)
+{
+    QTemporaryDir bundled;
+    QTemporaryDir user;
+    ASSERT_TRUE(bundled.isValid());
+    ASSERT_TRUE(user.isValid());
+    writeFile(bundled.path(), "base.toml",
+              "name = \"Provider Base\"\n"
+              "abstract = true\n"
+              "provider_instance = \"P\"\n"
+              "endpoint = \"/e\"\n"
+              "[body]\n"
+              "stream = true\n");
+    writeFile(bundled.path(), "a.toml",
+              "name = \"A\"\n"
+              "extends = \"Provider Base\"\n"
+              "model = \"stock-model\"\n");
+    writeFile(user.path(), "mine.toml",
+              "name = \"My A\"\n"
+              "extends = \"Provider Base\"\n"
+              "model = \"my-model\"\n"
+              "[body]\n"
+              "temperature = 0.2\n");
+
+    const auto result = AgentLoader::load(bundled.path(), user.path());
+    EXPECT_TRUE(result.errors.isEmpty()) << result.errors.join("; ").toStdString();
+    const AgentConfig *stock = findConfig(result, QStringLiteral("A"));
+    ASSERT_NE(stock, nullptr);
+    EXPECT_EQ(stock->model, QStringLiteral("stock-model"));
+    const AgentConfig *mine = findConfig(result, QStringLiteral("My A"));
+    ASSERT_NE(mine, nullptr);
+    EXPECT_EQ(mine->model, QStringLiteral("my-model"));
+    EXPECT_EQ(mine->providerInstance, QStringLiteral("P"));
+    EXPECT_TRUE(mine->body.contains(QStringLiteral("stream")));
+    EXPECT_TRUE(mine->body.contains(QStringLiteral("temperature")));
+    EXPECT_TRUE(mine->isUserSource());
+}
+
+TEST(AgentLoaderTest, ExtendsUnknownParentErrorNamesChild)
+{
+    QTemporaryDir dir;
+    ASSERT_TRUE(dir.isValid());
+    writeFile(dir.path(), "child.toml", minimalAgent("Child", "extends = \"NoSuchBase\"\n"));
+
+    const auto result = AgentLoader::load(QString(), dir.path());
+    EXPECT_TRUE(anyContains(result.errors, QStringLiteral("'Child' extends unknown agent 'NoSuchBase'")));
+    EXPECT_EQ(findConfig(result, QStringLiteral("Child")), nullptr);
+}
+
+TEST(AgentLoaderTest, ParseFileReportsWarningsForOwnFileOnly)
+{
+    QTemporaryDir dir;
+    ASSERT_TRUE(dir.isValid());
+    writeFile(dir.path(), "other.toml", minimalAgent("Other", "bogus_key = 1\n"));
+    writeFile(dir.path(), "target.toml", minimalAgent("Target", "another_bogus = 2\n"));
+
+    QString error;
+    QStringList warnings;
+    const auto cfg = AgentLoader::parseFile(
+        dir.path() + QStringLiteral("/target.toml"), QString(), &error, &warnings);
+    ASSERT_TRUE(cfg.has_value()) << error.toStdString();
+    EXPECT_TRUE(anyContains(warnings, QStringLiteral("another_bogus")));
+    EXPECT_FALSE(anyContains(warnings, QStringLiteral("bogus_key")));
+}

test/BundledAgentsTest.cpp

@@ -23,6 +23,10 @@ TEST(BundledAgentsTest, AllBundledAgentsLoadResolveAndRender)
         << "bundled agent load errors: "
         << result.errors.join(QStringLiteral("; ")).toStdString();
 
+    EXPECT_TRUE(result.warnings.isEmpty())
+        << "bundled agent load warnings: "
+        << result.warnings.join(QStringLiteral("; ")).toStdString();
+
     ASSERT_FALSE(result.configs.empty()) << "no bundled agents were loaded from :/agents";
 
     for (const auto &cfg : result.configs) {

test/CMakeLists.txt

@@ -7,6 +7,7 @@ add_executable(QodeAssistTest
     JsonPromptTemplateTest.cpp
     ResponseRouterTest.cpp
     BundledAgentsTest.cpp
+    AgentLoaderTest.cpp
     # LLMClientInterfaceTests.cpp
     unittest_main.cpp
 )