0.9.12: perf fold, quieter chrome, compatible hosts, delete staged runtime_contract

.github/PULL_REQUEST_TEMPLATE.md

@@ -11,6 +11,7 @@
 
 ## Checklist
 
+- [ ] This PR adds a new layer/module/abstraction — it names or deletes the layer it replaces
 - [ ] Updated docs or comments as needed
 - [ ] Added or updated tests where relevant
 - [ ] Verified TUI behavior manually if UI changes

.github/workflows/ci.yml

@@ -131,7 +131,7 @@ jobs:
                 ;;
             esac
             case "${path}" in
-              crates/workflow/*|workflows/rlm_cache_change.star|.github/workflows/ci.yml)
+              crates/workflow/*|.github/workflows/ci.yml)
                 workflow=true
                 ;;
             esac
@@ -484,8 +484,8 @@ jobs:
         with:
           cache-bin: false
           save-if: ${{ github.ref == 'refs/heads/main' }}
-      - name: Run RLM cache workflow mock/replay tests
-        run: cargo test -p codewhale-workflow --locked rlm_cache_change
+      - name: Run workflow crate tests
+        run: cargo test -p codewhale-workflow --locked
 
   test:
     name: Test

.gitignore

@@ -1,5 +1,12 @@
 # Build artifacts
 /target
+/target-*
+.entire/
+.cursor/
+.plans/
+.codewhale-worktrees/
+docs/superpowers/
+codewhale-inference/
 /extensions/vscode/out/
 *.pdb
 *.exe

AGENTS.md

@@ -8,6 +8,10 @@ instructions or memory. The nearest scoped `AGENTS.md` adds path-specific rules.
 
 - Inspect status and existing consumers before editing. Preserve unrelated,
   dirty, and untracked work.
+- Before adding a module named `model_*`, `*_config`, `provider_*`, or
+  anything that "bridges", "mirrors", or "stages" an existing thing, grep
+  for the existing thing and edit it. A new layer must name the predecessor
+  it replaces in the module doc; otherwise edit the original.
 - Prefer the simplest implementation that preserves observable contracts. A
   rewrite is acceptable when justified by product intent and observed behavior,
   not as a shortcut around understanding existing code.

CHANGELOG.md

@@ -9,6 +9,20 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- Z.ai `GLM-5.3-Flash` and OpenRouter `z-ai/glm-5.3-flash` are first-class
+  picker rows (`/model GLM-5.3-Flash`). Flash is the faster/explore sibling
+  of `GLM-5.3`; the Z.ai default stays `GLM-5.3`. List price is $0.15/$0.50
+  per 1M (durable; the 50% promo through 2026-09-09 is not the catalog row).
+- Baseten, Groq, and Cerebras are bundled OpenAI-compatible setup templates
+  (`[providers.<id>] kind = "openai-compatible"`), not new `ProviderKind`
+  variants. `/provider` fills URL, model, and env from one catalog row.
+- MCP manager copy now names the server, the failure, and one recovery
+  command (`The X MCP server requires OAuth reauthentication. Run /mcp login X`).
+- Settings Advanced exposes clickable MCP Connect, Reconnect, and Diagnose
+  actions, while Extensions and Problems route recovery through the existing
+  `/mcp login`, `/mcp reload`, `/mcp validate`, and `/plugin validate` commands
+  (#5643, #5655).
+
 - Added `/import-claude` (#5557): reads `~/.claude.json` and
   `~/.claude/settings.json` read-only and renders an explicit, reviewable
   migration plan plus a written report. MCP servers route through the
@@ -72,6 +86,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- Idle session metrics omit zero facts (`0 turns`, `LLM 0s`) until the
+  runtime has evidence. Working chrome says `in the current` instead of a
+  generic `working`.
+- Deleted nine uncompiled `runtime_contract/` staging files. Live contracts
+  remain `model.rs` and `termination.rs`.
+
+- The first #5587 dead-code sweep converts audited test-only helpers to
+  `#[cfg(test)]`, keeping production builds free of test-only APIs without
+  changing runtime behavior.
 - `/plugin reload` is now discoverable when on-disk plugin bundles change: the
   next send and `/plugin list` nudge once with `Run /plugin reload to apply`
   instead of silently keeping the stale catalog (#5579). Trust is unchanged;

config.example.toml

@@ -46,12 +46,14 @@ base_url = "https://api.deepseek.com/beta"
 #   z-ai/glm-5.2                  — OpenRouter Z.AI GLM 5.2
 #   z-ai/glm-5.3                  — OpenRouter Z.AI GLM 5.3 (live on Z.ai since 2026-08-13;
 #                                   metadata inherited from 5.2, unpriced)
-#   z-ai/glm-5-turbo              — OpenRouter Z.AI GLM 5 Turbo (scout fast sibling)
+#   z-ai/glm-5.3-flash            — OpenRouter Z.AI GLM 5.3 Flash (1M multimodal; $0.15/$0.50 list)
+#   z-ai/glm-5-turbo              — OpenRouter Z.AI GLM 5 Turbo (scout fast sibling of 5.2)
 #   GLM-5.3                       — default direct Z.AI Coding Plan model (live since 2026-08-13;
 #                                   metadata inherited from 5.2, unpriced)
+#   GLM-5.3-Flash                 — direct Z.AI GLM 5.3 Flash (faster/explore sibling of 5.3)
 #   GLM-5.2                       — direct Z.AI GLM 5.2 (previous default; explicit selections keep it)
 #   GLM-5.1                       — direct Z.AI smaller model
-#   GLM-5-Turbo                   — direct Z.AI fast model (scout fast sibling)
+#   GLM-5-Turbo                   — direct Z.AI fast model (scout fast sibling of 5.2)
 #   step-3.7-flash                — default direct StepFun / StepFlash model ID
 #   kimi-k3                         — direct Moonshot K3 model ID (1M context)
 #   kimi-k2.7-code                  — default direct Moonshot/Kimi K2.7 model ID
@@ -674,12 +676,13 @@ max_subagents = 10 # optional (default 64, clamped to 1-128)
 # base_url = "https://api.z.ai/api/coding/paas/v4"
 # # General API endpoint, if you are not using the Coding Plan:
 # # base_url = "https://api.z.ai/api/paas/v4"
-# model = "GLM-5.3"                             # default; GLM-5.2 is the previous default, GLM-5.1 the smaller model, GLM-5-Turbo the fast sub-agent sibling
+# model = "GLM-5.3"                             # default; GLM-5.3-Flash is the fast sibling, GLM-5.2 the previous default, GLM-5.1 the smaller model, GLM-5-Turbo the 5.2 fast sibling
 # # GLM-5.3 is live on the Z.ai Coding Plan (2026-08-13). Its catalog metadata
 # # (limits, reasoning options) is inherited from GLM-5.2 until Z.ai publishes
-# # distinct 5.3 numbers, and it carries no price. An explicit model = "GLM-5.2"
-# # keeps sending GLM-5.2; only the default moved. Accounts not provisioned for
-# # 5.3 can still see a 429 with entitlement code 1311.
+# # distinct 5.3 numbers, and it carries no price. GLM-5.3-Flash (2026-08-26)
+# # is the 1M multimodal picker row (`model = "GLM-5.3-Flash"`). An explicit
+# # model = "GLM-5.2" keeps sending GLM-5.2; only the default moved. Accounts
+# # not provisioned for 5.3 can still see a 429 with entitlement code 1311.
 
 # StepFun / StepFlash direct OpenAI-compatible endpoint (https://platform.stepfun.ai)
 [providers.stepfun]

crates/agent/src/lib.rs

@@ -516,6 +516,13 @@ impl Default for ModelRegistry {
                 supports_tools: true,
                 supports_reasoning: true,
             },
+            ModelInfo {
+                id: "z-ai/glm-5.3-flash".to_string(),
+                provider: ProviderKind::Openrouter,
+                aliases: vec!["glm-5.3-flash".to_string(), "zai-glm-5.3-flash".to_string()],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
             ModelInfo {
                 id: "z-ai/glm-5-turbo".to_string(),
                 provider: ProviderKind::Openrouter,
@@ -535,6 +542,18 @@ impl Default for ModelRegistry {
                 supports_tools: true,
                 supports_reasoning: true,
             },
+            ModelInfo {
+                id: "GLM-5.3-Flash".to_string(),
+                provider: ProviderKind::Zai,
+                aliases: vec![
+                    "glm-5.3-flash".to_string(),
+                    "glm-5-3-flash".to_string(),
+                    "zai-glm-5.3-flash".to_string(),
+                    "zai-glm-5-3-flash".to_string(),
+                ],
+                supports_tools: true,
+                supports_reasoning: true,
+            },
             // The first Z.ai row is the provider default. Keep this ordering
             // aligned with `DEFAULT_ZAI_MODEL` in codewhale-config.
             ModelInfo {
@@ -2321,6 +2340,10 @@ mod tests {
             ("glm-5.3", "GLM-5.3"),
             ("glm-5-3", "GLM-5.3"),
             ("zai-glm-5-3", "GLM-5.3"),
+            ("GLM-5.3-Flash", "GLM-5.3-Flash"),
+            ("glm-5.3-flash", "GLM-5.3-Flash"),
+            ("glm-5-3-flash", "GLM-5.3-Flash"),
+            ("zai-glm-5.3-flash", "GLM-5.3-Flash"),
             ("GLM-5-Turbo", "GLM-5-Turbo"),
             ("glm-5-turbo", "GLM-5-Turbo"),
             ("zai-glm-5-turbo", "GLM-5-Turbo"),
@@ -2555,6 +2578,7 @@ mod tests {
             ("glm-5.1", "z-ai/glm-5.1"),
             ("glm-5.2", "z-ai/glm-5.2"),
             ("glm-5.3", "z-ai/glm-5.3"),
+            ("glm-5.3-flash", "z-ai/glm-5.3-flash"),
             ("minimax-m3", "minimax/minimax-m3"),
             ("minimax-2.7", "minimax/minimax-m2.7"),
             ("openrouter-mimo-v2.5-pro", "xiaomi/mimo-v2.5-pro"),

crates/config/assets/models_dev.bundled.json

@@ -9,8 +9,9 @@
     "curated": "qwen3.8-max (GA) is curated ahead of upstream Models.dev, which as of 2026-08-03 lists only qwen3.8-max-preview; facts verified against the owner's Token Plan console (2026-08-03): ~1M context, 128K output, image understanding, always-on reasoning. deepseek-v4-flash-0731 keeps the console/in-repo wire id for the row upstream serves as deepseek-v4-flash. Coding Plan rows for qwen3.8-max-preview, deepseek-v4-pro, deepseek-v4-flash-0731, and glm-5.2 are curated from the Token Plan upstream entries (upstream alibaba-coding-plan does not list them yet); the in-repo route layer already offers the same model set on both plans. Upstream provider ids alibaba-token-plan(-cn) / alibaba-coding-plan(-cn) were merged onto the CodeWhale provider ids (live refresh normalizes them via ProviderKind aliases; the -cn regional variants stay upstream-id browse rows until Codewhale ships China endpoints).",
     "qwen_3_8_flash_2026_08_26": "OpenRouter qwen/qwen3.8-flash recorded 2026-08-26 against https://models.dev/api.json: release_date 2026-08-26, 1,000,000 context / 131,072 output, text+image+video input / text output, reasoning true, tool_call true, family qwen. Durable list prices (no promo annotation): input 0.16, output 0.47, cache_read 0.016, cache_write 0.20 per 1M. Alibaba first-party lists only qwen3.8-max (already curated, unpriced on Token Plan); this seed curates the OpenRouter row only and does not invent a first-party flash id. Live refresh supersedes on (provider, wire_model_id) identity. The flash suffix is not a GA family default, so DEFAULT_MODELSTUDIO_TOKEN_PLAN_MODEL stays qwen3.8-max and this row is not default: true.",
     "pending_release_metadata": "GLM-5.3 is live on the Z.ai Coding Plan (docs.z.ai/devpack/overview and docs.z.ai/devpack/latest-model, recorded 2026-08-13) and is the default direct Z.ai model (DEFAULT_ZAI_MODEL); explicit GLM-5.2 selections keep their own id. First-party wire id is GLM-5.3; OpenRouter mirror is z-ai/glm-5.3. Capability/limit/dialect values still inherit from GLM-5.2 until Z.ai publishes distinct 5.3 numbers. Pricing stays absent: Coding Plan publishes credit multipliers, not a USD PAYG row we can stand behind. Z.ai may auto-route GLM-5.2/GLM-5.1 requests to GLM-5.3 on their side; Codewhale still sends the selected picker id. Do not send a [1m] suffix. Scope stays first-party Z.ai plus the OpenRouter mirror; add third-party gateway rows only against that gateway's own published roster.",
+    "glm_5_3_flash_2026_08_26": "GLM-5.3-Flash recorded 2026-08-26 against https://docs.z.ai/guides/overview/pricing: natively multimodal (text/image/video), 1,000,000 context / 131,072 output, reasoning + tools. First-party wire id GLM-5.3-Flash; OpenRouter mirror z-ai/glm-5.3-flash. Durable list prices (not the 50% promo ending 2026-09-09 UTC+8): input 0.15, output 0.50, cache_read 0.03 per 1M. Flash is a picker row and the faster/explore sibling of GLM-5.3; DEFAULT_ZAI_MODEL stays GLM-5.3.",
     "currency_sweep_2026_08_17": "Rows re-verified against official pages on 2026-08-17 (#5470 follow-up): gpt-5.6-terra 2.00/12.00 (cache read 0.20, write 2.50) and gpt-5.6-luna 0.20/1.20 (0.02/0.25) per developers.openai.com model pages; claude-sonnet-5 2.00/10.00 (0.20/2.50) is now Anthropic's standard price (the 2026-09-01 increase was cancelled) and claude-opus-5 5.00/25.00 (0.50/6.25) was added; kimi-k3 3.00/15.00 (0.30) and kimi-k2.7-code-highspeed 1.90/8.00 (0.38) per platform.kimi.ai; MiniMax-M2.7-highspeed 0.60/2.40 (0.06/0.375) per platform.minimax.io; grok-4.5 (500K) and grok-4.3 (1M) carry limits only because xAI doubles their rates past 200K (same rule as grok-4.6); OpenRouter dots-studio/dots-3-note-preview:free carries limits only (its single free endpoint publishes $0, which this seed does not restate as a price).",
-    "coverage": "20 providers, 90 model rows (offline seed only)."
+    "coverage": "20 providers, 92 model rows (offline seed only)."
   },
   "models": {
     "deepseek-v4-pro": {
@@ -104,6 +105,18 @@
           "modalities": { "input": ["text"], "output": ["text"] },
           "limit": { "context": 1000000, "output": 131072 }
         },
+        "GLM-5.3-Flash": {
+          "id": "GLM-5.3-Flash",
+          "name": "GLM-5.3-Flash",
+          "family": "glm",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [{ "type": "effort", "values": ["high", "max"] }],
+          "tool_call": true,
+          "modalities": { "input": ["text", "image", "video"], "output": ["text"] },
+          "limit": { "context": 1000000, "output": 131072 },
+          "cost": { "input": 0.15, "output": 0.50, "cache_read": 0.03 }
+        },
         "glm-5.1": {
           "id": "glm-5.1",
           "name": "GLM-5.1",
@@ -1099,6 +1112,17 @@
           "modalities": { "input": ["text"], "output": ["text"] },
           "limit": { "context": 1000000, "output": 131072 }
         },
+        "z-ai/glm-5.3-flash": {
+          "id": "z-ai/glm-5.3-flash",
+          "name": "GLM-5.3-Flash (OpenRouter)",
+          "family": "glm",
+          "attachment": true,
+          "reasoning": true,
+          "tool_call": true,
+          "modalities": { "input": ["text", "image", "video"], "output": ["text"] },
+          "limit": { "context": 1000000, "output": 131072 },
+          "cost": { "input": 0.15, "output": 0.50, "cache_read": 0.03 }
+        },
         "dots-studio/dots-3-note-preview:free": {
           "id": "dots-studio/dots-3-note-preview:free",
           "name": "Dots Studio Dots3-Note Preview (OpenRouter, free endpoint)",

crates/config/src/catalog.rs

@@ -32,6 +32,7 @@
 //! [`ProviderCatalogCache`] tests).
 
 use std::collections::BTreeMap;
+use std::sync::OnceLock;
 use std::time::{SystemTime, UNIX_EPOCH};
 
 use serde::{Deserialize, Serialize};
@@ -186,16 +187,30 @@ impl CatalogOffering {
 /// honesty rule on omitted pricing (`UnknownOrStale`, never a fabricated zero).
 pub const BUNDLED_MODELS_DEV_JSON: &str = include_str!("../assets/models_dev.bundled.json");
 
+/// Parse-once cache for the committed bundled Models.dev snapshot.
+///
+/// The bundled asset is compile-time constant (`include_str!`), so its parsed
+/// form is immutable and safe to share process-wide. Before this cache, every
+/// call site parsed the full snapshot independently — the client route path,
+/// pickers, provider lake, and fleet identity each paid a full serde parse of
+/// ~50KB on their own first use (perf-attributed during the 0.9.x perf
+/// gauntlet: `ModelsDevCost` serde frames in startup profiles).
+static BUNDLED_MODELS_DEV_CATALOG: OnceLock<ModelsDevCatalog> = OnceLock::new();
+
 /// Parse the committed bundled Models.dev snapshot.
 ///
+/// The first call parses; later calls return the shared parsed catalog.
+///
 /// # Panics
 /// Panics only if the committed asset is not valid Models.dev JSON. The
 /// `tests::bundled_asset_parses` guard makes that a build-time failure, so this
 /// never panics in shipped builds.
 #[must_use]
-pub fn bundled_models_dev_catalog() -> ModelsDevCatalog {
-    ModelsDevCatalog::parse_json(BUNDLED_MODELS_DEV_JSON)
-        .expect("committed bundled Models.dev asset must be valid JSON")
+pub fn bundled_models_dev_catalog() -> &'static ModelsDevCatalog {
+    BUNDLED_MODELS_DEV_CATALOG.get_or_init(|| {
+        ModelsDevCatalog::parse_json(BUNDLED_MODELS_DEV_JSON)
+            .expect("committed bundled Models.dev asset must be valid JSON")
+    })
 }
 
 /// Bundled-layer [`CatalogOffering`] rows from the offline snapshot (#4188).
@@ -205,7 +220,7 @@ pub fn bundled_models_dev_catalog() -> ModelsDevCatalog {
 /// rows override these on `(provider, wire_model_id)` when available.
 #[must_use]
 pub fn bundled_catalog_offerings() -> Vec<CatalogOffering> {
-    bundled_offerings_from_models_dev(&bundled_models_dev_catalog())
+    bundled_offerings_from_models_dev(bundled_models_dev_catalog())
 }
 
 /// Hydrate bundled [`CatalogOffering`] rows from a parsed Models.dev catalog.

crates/config/src/catalog/tests.rs

@@ -648,7 +648,7 @@ fn bundled_asset_parses() {
         "bundled asset must carry provider rows"
     );
     // The helper returns the same parsed catalog.
-    assert_eq!(bundled_models_dev_catalog(), catalog);
+    assert_eq!(*bundled_models_dev_catalog(), catalog);
 }
 
 #[test]
@@ -851,9 +851,11 @@ fn bundled_asset_pricing_is_honest() {
 
     // GLM-5.3 is live on the Coding Plan, but Z.ai has published no USD PAYG
     // rate for it. Coding Plan credit multipliers are not USD, so every
-    // glm-5.3 row stays unpriced rather than inheriting glm-5.2's rates.
+    // glm-5.3 row *except Flash* stays unpriced rather than inheriting
+    // glm-5.2's rates. GLM-5.3-Flash has a published list (2026-08-26).
     for row in &rows {
-        if row.wire_model_id.to_ascii_lowercase().contains("glm-5.3") {
+        let wire = row.wire_model_id.to_ascii_lowercase();
+        if wire.contains("glm-5.3") && !wire.contains("flash") {
             assert!(
                 row.cost.is_none(),
                 "{}/{}: glm-5.3 must stay unpriced until Z.ai publishes rates",
@@ -863,6 +865,23 @@ fn bundled_asset_pricing_is_honest() {
         }
     }
 
+    let glm53_flash = find(&rows, "zai", "GLM-5.3-Flash");
+    let cost = glm53_flash
+        .cost
+        .as_ref()
+        .expect("GLM-5.3-Flash must ship priced at durable list rates");
+    assert_eq!(cost.input, Some(0.15));
+    assert_eq!(cost.output, Some(0.50));
+    assert_eq!(cost.cache_read, Some(0.03));
+    assert_eq!(
+        glm53_flash.limit.as_ref().and_then(|l| l.context),
+        Some(1_000_000)
+    );
+    assert!(
+        !glm53_flash.default_for_provider,
+        "GLM-5.3-Flash is a picker row, not the Z.ai default"
+    );
+
     // OpenRouter qwen3.8-flash lists durable (non-promo) rates on models.dev
     // as of 2026-08-26. Unlike GLM-5.3-Flash's explicit 50% promo, this row
     // must ship priced. It is not a family default.