PyPI - roadmodel - Versions diffs - 0.2.43__tar.gz → 0.2.45__tar.gz - Mend

roadmodel 0.2.43tar.gz → 0.2.45tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (49) hide show

{roadmodel-0.2.43 → roadmodel-0.2.45}/PKG-INFO RENAMED Viewed

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: roadmodel
-Version: 0.2.43
+Version: 0.2.45
 Summary: BYO-key CLI that recommends the right AI model, platform, and settings for a prompt.
 Project-URL: Homepage, https://roadmodel.ai
 Project-URL: Repository, https://github.com/nathanramoscfa/roadmodel
@@ -323,7 +323,7 @@ What the loop guarantees, independent of the project:
   and a step that touches a deployed surface is done at post-deploy
   verification, not at merge.
-Setup is once per machine (`/roadmodel-update` keeps every project's
+Setup is once per machine (`/roadmodel-upgrade` keeps every project's
 copy current after that); the commands then work in any project.
 Step-by-step: [docs/planning-workflow.md](docs/planning-workflow.md).
@@ -360,15 +360,15 @@ pip install -U roadmodel
   **then** `roadmodel export-kit . --force` to refresh it. The `/roadmap-project`
   and `/roadmap-phase` commands do both as their Step 0.
-### Every project on a machine — `/roadmodel-update`
+### Every project on a machine — `/roadmodel-upgrade`
 Projects that each carry roadmodel in their **own** conda env or venv would
 otherwise be upgraded one at a time. Instead, from any Claude Code chat:
 ```
-/roadmodel-update                       # every registered project
-/roadmodel-update ~/code/app-one …      # register these dirs, then run
-/roadmodel-update --install-schedule    # once: also run daily, unattended
+/roadmodel-upgrade                      # every registered project
+/roadmodel-upgrade ~/code/app-one …     # register these dirs, then run
+/roadmodel-upgrade --install-schedule   # once: also run daily, unattended
 ```
 The command fetches [`scripts/update_projects.py`](scripts/update_projects.py)
@@ -446,14 +446,14 @@ hand; the `export-kit` command above avoids that entirely.
 `docs/claude-commands/` — `/roadmap-project`, `/roadmap-phase N`,
 `/roadmap-step P M` (executes a step), `/roadmap-refresh` (bookkeeping
 only: marks shipped steps and re-selects upcoming Settings, runs
-nothing), `/roadmodel-update` — see
+nothing), `/roadmodel-upgrade` — see
 [The planning workflow](#the-planning-workflow) above for the loop and
 [docs/planning-workflow.md](docs/planning-workflow.md) for setup and
 per-command detail. The same five are generated for **Gemini CLI**
 (`~/.gemini/commands/*.toml`), **Codex** and **Cursor** (one
 `~/.agents/skills/*/SKILL.md` both read; `$roadmap-step` in Codex,
 `/roadmap-step` in Cursor), and **OpenCode**
-(`~/.config/opencode/commands/*.md`), installed by `/roadmodel-update`
+(`~/.config/opencode/commands/*.md`), installed by `/roadmodel-upgrade`
 wherever those tools are present.
 Either way the in-editor AI runs the algorithm itself rather than calling
@@ -582,7 +582,7 @@ provider goes down, or a task suits a different model?
 (instructions, memory, skills, slash commands, MCP servers, model, effort,
 permissions) across Claude, Codex, Gemini and open-source clients in any
 direction, and carries one parameterised prompt that makes the target agent
-sync itself. `roadmodel-update` keeps the roadmap commands, the roadmodel MCP
+sync itself. `roadmodel-upgrade` keeps the roadmap commands, the roadmodel MCP
 server and the reasoning-effort default in parity automatically.
 ## Subcommands

{roadmodel-0.2.43 → roadmodel-0.2.45}/README.md RENAMED Viewed

@@ -73,7 +73,7 @@ What the loop guarantees, independent of the project:
   and a step that touches a deployed surface is done at post-deploy
   verification, not at merge.
-Setup is once per machine (`/roadmodel-update` keeps every project's
+Setup is once per machine (`/roadmodel-upgrade` keeps every project's
 copy current after that); the commands then work in any project.
 Step-by-step: [docs/planning-workflow.md](docs/planning-workflow.md).
@@ -110,15 +110,15 @@ pip install -U roadmodel
   **then** `roadmodel export-kit . --force` to refresh it. The `/roadmap-project`
   and `/roadmap-phase` commands do both as their Step 0.
-### Every project on a machine — `/roadmodel-update`
+### Every project on a machine — `/roadmodel-upgrade`
 Projects that each carry roadmodel in their **own** conda env or venv would
 otherwise be upgraded one at a time. Instead, from any Claude Code chat:
 ```
-/roadmodel-update                       # every registered project
-/roadmodel-update ~/code/app-one …      # register these dirs, then run
-/roadmodel-update --install-schedule    # once: also run daily, unattended
+/roadmodel-upgrade                      # every registered project
+/roadmodel-upgrade ~/code/app-one …     # register these dirs, then run
+/roadmodel-upgrade --install-schedule   # once: also run daily, unattended
 ```
 The command fetches [`scripts/update_projects.py`](scripts/update_projects.py)
@@ -196,14 +196,14 @@ hand; the `export-kit` command above avoids that entirely.
 `docs/claude-commands/` — `/roadmap-project`, `/roadmap-phase N`,
 `/roadmap-step P M` (executes a step), `/roadmap-refresh` (bookkeeping
 only: marks shipped steps and re-selects upcoming Settings, runs
-nothing), `/roadmodel-update` — see
+nothing), `/roadmodel-upgrade` — see
 [The planning workflow](#the-planning-workflow) above for the loop and
 [docs/planning-workflow.md](docs/planning-workflow.md) for setup and
 per-command detail. The same five are generated for **Gemini CLI**
 (`~/.gemini/commands/*.toml`), **Codex** and **Cursor** (one
 `~/.agents/skills/*/SKILL.md` both read; `$roadmap-step` in Codex,
 `/roadmap-step` in Cursor), and **OpenCode**
-(`~/.config/opencode/commands/*.md`), installed by `/roadmodel-update`
+(`~/.config/opencode/commands/*.md`), installed by `/roadmodel-upgrade`
 wherever those tools are present.
 Either way the in-editor AI runs the algorithm itself rather than calling
@@ -332,7 +332,7 @@ provider goes down, or a task suits a different model?
 (instructions, memory, skills, slash commands, MCP servers, model, effort,
 permissions) across Claude, Codex, Gemini and open-source clients in any
 direction, and carries one parameterised prompt that makes the target agent
-sync itself. `roadmodel-update` keeps the roadmap commands, the roadmodel MCP
+sync itself. `roadmodel-upgrade` keeps the roadmap commands, the roadmodel MCP
 server and the reasoning-effort default in parity automatically.
 ## Subcommands

{roadmodel-0.2.43 → roadmodel-0.2.45}/docs/benchmarks.json RENAMED Viewed

@@ -6,7 +6,7 @@
     "endpoint": "https://artificialanalysis.ai/api/v2/data/llms/models",
     "attribution": "Benchmark data © Artificial Analysis, used under its API terms with attribution."
   },
-  "generated_at_utc": "2026-09-25T16:50:05Z",
+  "generated_at_utc": "2026-09-25T22:10:15Z",
   "model_count": 50,
   "unmapped": [],
   "missing_slugs": [],
@@ -36,8 +36,8 @@
         "tau2": 0.546783625730994,
         "tau_banking": 0.0927835051546392
       },
-      "median_output_tokens_per_second": 114.956,
-      "median_time_to_first_token_seconds": 15.297
+      "median_output_tokens_per_second": 112.331,
+      "median_time_to_first_token_seconds": 14.822
     },
     "claude-fable-5": {
       "aa_id": "cd55210d-358e-4df1-ba9c-9acb5f186cc9",
@@ -92,8 +92,8 @@
         "tau2": null,
         "tau_banking": 0.472164948453608
       },
-      "median_output_tokens_per_second": 70.547,
-      "median_time_to_first_token_seconds": 137.194
+      "median_output_tokens_per_second": 71.496,
+      "median_time_to_first_token_seconds": 137.695
     },
     "claude-opus-5": {
       "aa_id": "b8fc61f7-5e9a-49e6-8547-6ac56db24627",
@@ -120,8 +120,8 @@
         "tau2": null,
         "tau_banking": 0.420618556701031
       },
-      "median_output_tokens_per_second": 51.772,
-      "median_time_to_first_token_seconds": 27.209
+      "median_output_tokens_per_second": 67.509,
+      "median_time_to_first_token_seconds": 34.244
     },
     "claude-opus-5-5": {
       "aa_id": "2f3c4dc9-a450-4303-8697-0237257cf08f",
@@ -176,7 +176,7 @@
         "tau2": null,
         "tau_banking": 0.37319587628866
       },
-      "median_output_tokens_per_second": 82.788,
+      "median_output_tokens_per_second": 86.276,
       "median_time_to_first_token_seconds": 138.497
     },
     "deepseek-flash": {
@@ -205,7 +205,7 @@
         "tau_banking": null
       },
       "median_output_tokens_per_second": 225.054,
-      "median_time_to_first_token_seconds": 0.763
+      "median_time_to_first_token_seconds": 0.77
     },
     "deepseek-v4-pro": {
       "aa_id": "83173329-c09a-41f1-a028-a282a5f908d7",
@@ -232,8 +232,8 @@
         "tau2": null,
         "tau_banking": 0.395876288659794
       },
-      "median_output_tokens_per_second": 77.754,
-      "median_time_to_first_token_seconds": 1.125
+      "median_output_tokens_per_second": 79.847,
+      "median_time_to_first_token_seconds": 1.103
     },
     "gemini-2.5-flash": {
       "aa_id": "219ed587-60c5-4a48-9517-8480e08d0ca1",
@@ -345,7 +345,7 @@
         "tau_banking": 0.214432989690722
       },
       "median_output_tokens_per_second": 130.627,
-      "median_time_to_first_token_seconds": 18.827
+      "median_time_to_first_token_seconds": 18.476
     },
     "gemini-3.5-flash": {
       "aa_id": "0097ebf5-124f-42f6-9463-33b00e711f03",
@@ -457,7 +457,7 @@
         "tau_banking": 0.449484536082474
       },
       "median_output_tokens_per_second": 277.514,
-      "median_time_to_first_token_seconds": 9.823
+      "median_time_to_first_token_seconds": 8.773
     },
     "glm-4.5-air": {
       "aa_id": "5d303dc9-c027-401f-9803-4e9aa3331007",
@@ -736,8 +736,8 @@
         "tau2": 0.859649122807018,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 120.423,
-      "median_time_to_first_token_seconds": 40.827
+      "median_output_tokens_per_second": 117.009,
+      "median_time_to_first_token_seconds": 39.444
     },
     "gpt-5.4": {
       "aa_id": "a89c4b28-2d8c-456e-88ea-255fb51fd2b6",
@@ -876,8 +876,8 @@
         "tau2": null,
         "tau_banking": 0.311340206185567
       },
-      "median_output_tokens_per_second": 104.659,
-      "median_time_to_first_token_seconds": 95.596
+      "median_output_tokens_per_second": 102.358,
+      "median_time_to_first_token_seconds": 22.543
     },
     "gpt-5.6-sol": {
       "aa_id": "d93edfe8-bf35-49ad-b56e-b18116142a1c",
@@ -904,7 +904,7 @@
         "tau2": 0.850877192982456,
         "tau_banking": 0.443298969072165
       },
-      "median_output_tokens_per_second": 62.063,
+      "median_output_tokens_per_second": 86.318,
       "median_time_to_first_token_seconds": 45.71
     },
     "gpt-5.6-terra": {
@@ -932,8 +932,8 @@
         "tau2": 0.862573099415205,
         "tau_banking": 0.402061855670103
       },
-      "median_output_tokens_per_second": 86.203,
-      "median_time_to_first_token_seconds": 176.983
+      "median_output_tokens_per_second": 86.687,
+      "median_time_to_first_token_seconds": 143.358
     },
     "gpt-6-astra": {
       "aa_id": "2f339a97-9a0d-499a-9cb5-e0db665bfa25",
@@ -960,8 +960,8 @@
         "tau2": null,
         "tau_banking": 0.414432989690722
       },
-      "median_output_tokens_per_second": 54.552,
-      "median_time_to_first_token_seconds": 206.901
+      "median_output_tokens_per_second": 54.12,
+      "median_time_to_first_token_seconds": 214.045
     },
     "gpt-6-luna": {
       "aa_id": "6fb13851-40ce-4bda-ad3f-6b38e0c5daa7",
@@ -988,8 +988,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 129.252,
-      "median_time_to_first_token_seconds": 126.505
+      "median_output_tokens_per_second": 125.845,
+      "median_time_to_first_token_seconds": 127.063
     },
     "gpt-6-sol": {
       "aa_id": "c50ea08c-88c0-4eb7-85b8-27f2bbf6d527",
@@ -1016,8 +1016,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 98.199,
-      "median_time_to_first_token_seconds": 98.756
+      "median_output_tokens_per_second": 85.316,
+      "median_time_to_first_token_seconds": 98.805
     },
     "gpt-oss-120b": {
       "aa_id": "f0083258-8646-45b8-8082-7aaf6c2ea82a",
@@ -1044,8 +1044,8 @@
         "tau2": 0.657894736842105,
         "tau_banking": 0.127835051546392
       },
-      "median_output_tokens_per_second": 172.661,
-      "median_time_to_first_token_seconds": 0.515
+      "median_output_tokens_per_second": 174.466,
+      "median_time_to_first_token_seconds": 0.512
     },
     "gpt-oss-20b": {
       "aa_id": "36f73aaf-d38a-4b56-a2b3-d04d17186910",
@@ -1072,8 +1072,8 @@
         "tau2": 0.60233918128655,
         "tau_banking": 0.0701030927835052
       },
-      "median_output_tokens_per_second": 191.591,
-      "median_time_to_first_token_seconds": 0.434
+      "median_output_tokens_per_second": 194.912,
+      "median_time_to_first_token_seconds": 0.439
     },
     "grok-4.3": {
       "aa_id": "948892b5-db03-4118-a4a8-ccd51ed871ea",
@@ -1129,7 +1129,7 @@
         "tau_banking": 0.507216494845361
       },
       "median_output_tokens_per_second": 68.306,
-      "median_time_to_first_token_seconds": 35.691
+      "median_time_to_first_token_seconds": 33.772
     },
     "grok-4.7": {
       "aa_id": "59627802-1546-422c-9a65-65b61c022d36",
@@ -1156,8 +1156,8 @@
         "tau2": null,
         "tau_banking": null
       },
-      "median_output_tokens_per_second": 58.165,
-      "median_time_to_first_token_seconds": 0.802
+      "median_output_tokens_per_second": 58.859,
+      "median_time_to_first_token_seconds": 0.806
     },
     "kimi-k2.7-code": {
       "aa_id": "8d0cb231-7303-452c-9923-a9620b948475",
@@ -1184,8 +1184,8 @@
         "tau2": 0.900584795321637,
         "tau_banking": 0.202061855670103
       },
-      "median_output_tokens_per_second": 68.728,
-      "median_time_to_first_token_seconds": 1.062
+      "median_output_tokens_per_second": 79.482,
+      "median_time_to_first_token_seconds": 1.058
     },
     "kimi-k3": {
       "aa_id": "f7d2fc3e-1f7b-405f-818c-07952a4af78f",
@@ -1212,8 +1212,8 @@
         "tau2": null,
         "tau_banking": 0.45979381443299
       },
-      "median_output_tokens_per_second": 37.356,
-      "median_time_to_first_token_seconds": 3.866
+      "median_output_tokens_per_second": 36.835,
+      "median_time_to_first_token_seconds": 3.827
     },
     "mistral-large-3": {
       "aa_id": "4928e950-7f37-4475-b0dc-c5bad781a321",
@@ -1240,8 +1240,8 @@
         "tau2": 0.245614035087719,
         "tau_banking": 0.0577319587628866
       },
-      "median_output_tokens_per_second": 72.929,
-      "median_time_to_first_token_seconds": 0.706
+      "median_output_tokens_per_second": 73.103,
+      "median_time_to_first_token_seconds": 0.698
     },
     "mistral-medium-3.5": {
       "aa_id": "dd059b25-d82a-4ead-82a4-4adceaaec48b",
@@ -1268,7 +1268,7 @@
         "tau2": 0.941520467836257,
         "tau_banking": 0.150515463917526
       },
-      "median_output_tokens_per_second": 153.138,
+      "median_output_tokens_per_second": 151.825,
       "median_time_to_first_token_seconds": 0.635
     },
     "mistral-small-4": {
@@ -1296,8 +1296,8 @@
         "tau2": 0.412280701754386,
         "tau_banking": 0.0494845360824742
       },
-      "median_output_tokens_per_second": 160.617,
-      "median_time_to_first_token_seconds": 0.567
+      "median_output_tokens_per_second": 159.678,
+      "median_time_to_first_token_seconds": 0.561
     },
     "muse-spark-1.3": {
       "aa_id": "9999672c-b687-4026-9c15-5cef02ff53bb",

{roadmodel-0.2.43 → roadmodel-0.2.45}/docs/catalog.json RENAMED Viewed

@@ -1,9 +1,9 @@
 {
   "schema_version": "3",
   "output_contract_version": 2,
-  "generated_at_utc": "2026-09-25T16:50:05Z",
+  "generated_at_utc": "2026-09-25T22:10:07Z",
   "source_doc_sha256": {
-    "model-selector.txt": "f9a14aef7559487e2f5fb5bf182f7aa8df73539b036775c2d58c5d6af08bbd1d",
+    "model-selector.txt": "dd4715585f2f251b6635b65ad3a4cc82eb9ea22b8ed14c494313416a15407c58",
     "model-tier-cost-scale.md": "9b9f57f5367b553ab11c1116ff69b89eb4d3a9251a7f1caebd813dbdc0c4716a"
   },
   "models": [
@@ -1348,7 +1348,7 @@
       "exposes_max_mode": "no",
       "exposes_thinking": "yes",
       "exposes_orchestration": "yes",
-      "best_for": "Default for Claude coding or terminal tasks when a claude.ai Max subscription is active \u2014 $0 marginal cost until the Max budget is exhausted, full tool-use surface, runs as a CLI, IDE extension inside Cursor, native VS Code extension, and as of 2.1.170+ also ships Claude Fable 5 (Mythos-class) plus a first-class Claude in Chrome browser-control integration (GA in 2.1.198), with Claude Sonnet 5 now the default model (2.1.197+) at native 1M-token context; as of 2.1.219 Claude Opus 5 ships as the new default Opus (1M context, fast mode at $10/$50 per Mtok) and Opus 4.7 is removed from fast mode so `/fast` now applies to Opus 5 and Opus 4.8; 2.1.280 ships Claude Opus 5.5 as the new default Opus (1M context, $4/$20 per Mtok with $0.20/Mtok cache reads) and changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise); 2.1.220\u20132.1.280 are bug-fix / reliability / self-hosted-runner / UI-polish / plugin-safety / VS Code / Claude apps / AGENTS.md / auto-mode-server-classifier releases with no effort/thinking surface changes beyond the 2.1.267 `maxEffortLevel` cap and the 2.1.269 `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1\u2013256) knob for the Workflow tool's per-run concurrent-agent limit noted below (2.1.224 adds `claude self-hosted-runner` for turning your own machines into places where Claude Code web/mobile/desktop sessions can run, cross-session `SendMessage`, and archive-plugin sources; 2.1.225 adds gateway spend-limit reporting to the usage warning; 2.1.227 polishes the slash-command menu and fixes feature-flag evaluation for expired login tokens; 2.1.257 ships Claude Fable 5.1 as the new default Fable at 1M context and $10/$50 per Mtok; 2.1.259 adds `managedMcpServers` policy and `--permission-prompts none` for unattended headless hosts; 2.1.260 adds an inline `/diff` panel and a headless text form of `/advisor`; 2.1.261 adds `/skill-doctor` and `bashOutputMaxChars` / `taskOutputMaxChars` settings; 2.1.263 is a reliability release; 2.1.265 adds a `--plugin-dir` folder-of-plugins mode and tightens plugin path containment; 2.1.266 restores gateway/proxy behavior when `CLAUDE_CODE_USE_GATEWAY` is set alongside an API key or `apiKeyHelper`; 2.1.267 adds a `maxEffortLevel` setting \u2014 top-level or per model under `modelSettings` \u2014 that CAPS the effort level on every provider including Bedrock, Vertex, and Foundry while still letting users pick any lower level, plus `--system-prompt-snapshot off` for iterating on prompt text and a fix for `effort:` frontmatter on custom commands, skills, and subagents being ignored on models whose default effort is still pinned; 2.1.269 adds `claude plugin eval` for scored plugin eval suites, `/output-style [name]` to list and switch output styles (including over Remote Control and in headless sessions), a Bash-tool file-edit diff (setting `bashEditDiffEnabled`), `OTEL_METRICS_INCLUDE_REPOSITORY` for tagging OpenTelemetry metrics and events with `vcs.*` attributes, `CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS` to extend the LLM gateway `/v1/models` discovery timeout, and `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1\u2013256) to raise the Workflow tool's per-run concurrent-agent limit for inference-bound fan-outs \u2014 none of which change the `/effort` vocabulary; 2.1.270 fixes read-only git commands in Bash unexpectedly asking for permission after a session had been running for a while, a 2.1.269 regression, with no surface changes; 2.1.271 adds fast mode in Claude Code Remote (cloud and self-hosted runner) sessions where organization policy allows it, mouse support in the `/config` panel, per-command Bash / PowerShell / Monitor `allowed_domains` in auto mode with sandboxing, agent-frontmatter and `--agents` JSON `omitClaudeMd` to let custom / plugin subagents run without CLAUDE.md files (managed policy still loads), and `--accept-command <sha256>` for `claude plugin install` / `update`; 2.1.272 is a bug-fix / reliability release; 2.1.273 adds `x-claude-code-*` LLM-gateway request-hint headers behind `CLAUDE_CODE_GATEWAY_HINT_HEADERS=1`, forking a `--remote-control` / `/remote-control` session from the Claude app as a local background session, and MCP-server-disconnected notifications, plus fixes for permission handling, credential/auth error messages, prompt-cache retention across `/login` / `/upgrade` / `/extra-usage`, and gateway 401 messaging \u2014 no effort/thinking surface changes; 2.1.274 adds a critical-memory-usage warning with recovery steps, `CLAUDE_CODE_MCP_STARTUP_WAIT_MS` to bound how long the first non-interactive turn waits for connecting MCP servers, an `effort` attribute on the `claude_code.llm_request` OpenTelemetry trace span, plus assorted VS Code, Claude Code on the web, Claude Tag, and Code Review fixes and improvements \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.275 adds signed-in account confirmation for Claude apps gateway sign-in, a send-now key (ctrl+enter, or ctrl+x ctrl+s) that flushes queued messages mid-turn, syncing of claude.ai-account skills and plugins into terminal sessions (opt out with `syncClaudeAiSkills: false` / `syncClaudeAiPlugins: false`), `/plugin install <plugin> --marketplace <source>`, and a startup warning when `otelHeadersHelper` fails, plus a broad set of reliability and rendering fixes \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.276 is a hotfix for a 2.1.275 regression where every request behind a proxy or gateway failed with `400 \u2026 Input tag 'advisor_20260301'`, with no surface changes; 2.1.277 adds AGENTS.md fallback support (in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead \u2014 configurable under \"Project instructions\" in `/config`; not yet on Bedrock, Vertex or Foundry), `CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1` for Claude apps gateways whose only egress is a forward proxy, an optional `headers:` map on Claude apps gateway upstreams, plus assorted reliability, plugin, background-task, and VS Code / Claude Code on the web / Claude Tag fixes \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.278 changes auto mode for Claude API and Enterprise users, and on Bedrock, Vertex, Foundry and gateways, to default to the server-side classifier that does not charge for classifier overhead (opt out on Bedrock, Vertex, Foundry and gateways with `CLAUDE_CODE_AUTO_MODE_SERVER=0`; warns on billed fallback) and adds an `Auto mode server` row to `/status` showing whether this session's auto mode classifier runs on the server \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.280 ships Claude Opus 5.5 as the new default Opus model at 1M context and $4/$20 per Mtok, changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise), and stops Opus 4.7, Opus 4.8 and Fable 5 from holding their launch-default effort over an explicit `/effort`, `effortLevel`, project or managed setting \u2014 the `/effort` vocabulary itself is unchanged). Exposes the full `/effort` dial (low/medium/high/xhigh/max \u2014 Opus 4.6 and Sonnet 4.6 top out at max with no xhigh step; Opus 5, Opus 5.5, Opus 4.7, Opus 4.8, Sonnet 5, Fable 5, and Fable 5.1 expose the full range; effort levels a model does not support fall back to the highest supported level at or below the requested one; deployments may set `maxEffortLevel` to cap the maximum reachable level on every provider), the Option+T/Alt+T, `alwaysThinkingEnabled`, and `MAX_THINKING_TOKENS=0` on/off controls, plus Ultracode (session-wide xhigh + Dynamic Workflows via `/effort ultracode` or `\"ultracode\": true`; 2.1.202+ adds a `/config` \"Dynamic workflow size\" advisory guideline, 2.1.219 defaults it to medium (aim for fewer than 15 agents) with the `workflowSizeGuideline` settings key to set it from any settings file, and 2.1.269 adds `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1\u2013256) to raise the per-run concurrent-agent cap for the Workflow tool) and the per-turn `ultrathink` keyword (`think`/`think hard`/`think more` are not recognized)."
+      "best_for": "Default for Claude coding or terminal tasks when a claude.ai Max subscription is active \u2014 $0 marginal cost until the Max budget is exhausted, full tool-use surface, runs as a CLI, IDE extension inside Cursor, native VS Code extension, and as of 2.1.170+ also ships Claude Fable 5 (Mythos-class) plus a first-class Claude in Chrome browser-control integration (GA in 2.1.198), with Claude Sonnet 5 now the default model (2.1.197+) at native 1M-token context; as of 2.1.219 Claude Opus 5 ships as the new default Opus (1M context, fast mode at $10/$50 per Mtok) and Opus 4.7 is removed from fast mode so `/fast` now applies to Opus 5 and Opus 4.8; 2.1.280 ships Claude Opus 5.5 as the new default Opus (1M context, $4/$20 per Mtok with $0.20/Mtok cache reads) and changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise); 2.1.220\u20132.1.280 are bug-fix / reliability / self-hosted-runner / UI-polish / plugin-safety / VS Code / Claude apps / AGENTS.md / auto-mode-server-classifier releases with no effort/thinking surface changes beyond the 2.1.267 `maxEffortLevel` cap and the 2.1.269 `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1\u2013256) knob for the Workflow tool's per-run concurrent-agent limit noted below (2.1.224 adds `claude self-hosted-runner` for turning your own machines into places where Claude Code web/mobile/desktop sessions can run, cross-session `SendMessage`, and archive-plugin sources; 2.1.225 adds gateway spend-limit reporting to the usage warning; 2.1.227 polishes the slash-command menu and fixes feature-flag evaluation for expired login tokens; 2.1.257 ships Claude Fable 5.1 as the new default Fable at 1M context and $10/$50 per Mtok; 2.1.259 adds `managedMcpServers` policy and `--permission-prompts none` for unattended headless hosts; 2.1.260 adds an inline `/diff` panel and a headless text form of `/advisor`; 2.1.261 adds `/skill-doctor` and `bashOutputMaxChars` / `taskOutputMaxChars` settings; 2.1.263 is a reliability release; 2.1.265 adds a `--plugin-dir` folder-of-plugins mode and tightens plugin path containment; 2.1.266 restores gateway/proxy behavior when `CLAUDE_CODE_USE_GATEWAY` is set alongside an API key or `apiKeyHelper`; 2.1.267 adds a `maxEffortLevel` setting \u2014 top-level or per model under `modelSettings` \u2014 that CAPS the effort level on every provider including Bedrock, Vertex, and Foundry while still letting users pick any lower level, plus `--system-prompt-snapshot off` for iterating on prompt text and a fix for `effort:` frontmatter on custom commands, skills, and subagents being ignored on models whose default effort is still pinned; 2.1.269 adds `claude plugin eval` for scored plugin eval suites, `/output-style [name]` to list and switch output styles (including over Remote Control and in headless sessions), a Bash-tool file-edit diff (setting `bashEditDiffEnabled`), `OTEL_METRICS_INCLUDE_REPOSITORY` for tagging OpenTelemetry metrics and events with `vcs.*` attributes, `CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS` to extend the LLM gateway `/v1/models` discovery timeout, and `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1\u2013256) to raise the Workflow tool's per-run concurrent-agent limit for inference-bound fan-outs \u2014 none of which change the `/effort` vocabulary; 2.1.270 fixes read-only git commands in Bash unexpectedly asking for permission after a session had been running for a while, a 2.1.269 regression, with no surface changes; 2.1.271 adds fast mode in Claude Code Remote (cloud and self-hosted runner) sessions where organization policy allows it, mouse support in the `/config` panel, per-command Bash / PowerShell / Monitor `allowed_domains` in auto mode with sandboxing, agent-frontmatter and `--agents` JSON `omitClaudeMd` to let custom / plugin subagents run without CLAUDE.md files (managed policy still loads), and `--accept-command <sha256>` for `claude plugin install` / `update`; 2.1.272 is a bug-fix / reliability release; 2.1.273 adds `x-claude-code-*` LLM-gateway request-hint headers behind `CLAUDE_CODE_GATEWAY_HINT_HEADERS=1`, forking a `--remote-control` / `/remote-control` session from the Claude app as a local background session, and MCP-server-disconnected notifications, plus fixes for permission handling, credential/auth error messages, prompt-cache retention across `/login` / `/upgrade` / `/extra-usage`, and gateway 401 messaging \u2014 no effort/thinking surface changes; 2.1.274 adds a critical-memory-usage warning with recovery steps, `CLAUDE_CODE_MCP_STARTUP_WAIT_MS` to bound how long the first non-interactive turn waits for connecting MCP servers, an `effort` attribute on the `claude_code.llm_request` OpenTelemetry trace span, plus assorted VS Code, Claude Code on the web, Claude Tag, and Code Review fixes and improvements \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.275 adds signed-in account confirmation for Claude apps gateway sign-in, a send-now key (ctrl+enter, or ctrl+x ctrl+s) that flushes queued messages mid-turn, syncing of claude.ai-account skills and plugins into terminal sessions (opt out with `syncClaudeAiSkills: false` / `syncClaudeAiPlugins: false`), `/plugin install <plugin> --marketplace <source>`, and a startup warning when `otelHeadersHelper` fails, plus a broad set of reliability and rendering fixes \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.276 is a hotfix for a 2.1.275 regression where every request behind a proxy or gateway failed with `400 \u2026 Input tag 'advisor_20260301'`, with no surface changes; 2.1.277 adds AGENTS.md fallback support (in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead \u2014 configurable under \"Project instructions\" in `/config`; extended by 2.1.281 to Bedrock, Vertex, Foundry, LLM gateways and telemetry-disabled sessions), `CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1` for Claude apps gateways whose only egress is a forward proxy, an optional `headers:` map on Claude apps gateway upstreams, plus assorted reliability, plugin, background-task, and VS Code / Claude Code on the web / Claude Tag fixes \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.278 changes auto mode for Claude API and Enterprise users, and on Bedrock, Vertex, Foundry and gateways, to default to the server-side classifier that does not charge for classifier overhead (opt out on Bedrock, Vertex, Foundry and gateways with `CLAUDE_CODE_AUTO_MODE_SERVER=0`; warns on billed fallback) and adds an `Auto mode server` row to `/status` showing whether this session's auto mode classifier runs on the server \u2014 no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.280 ships Claude Opus 5.5 as the new default Opus model at 1M context and $4/$20 per Mtok, changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise), and stops Opus 4.7, Opus 4.8 and Fable 5 from holding their launch-default effort over an explicit `/effort`, `effortLevel`, project or managed setting \u2014 the `/effort` vocabulary itself is unchanged). Exposes the full `/effort` dial (low/medium/high/xhigh/max \u2014 Opus 4.6 and Sonnet 4.6 top out at max with no xhigh step; Opus 5, Opus 5.5, Opus 4.7, Opus 4.8, Sonnet 5, Fable 5, and Fable 5.1 expose the full range; effort levels a model does not support fall back to the highest supported level at or below the requested one; deployments may set `maxEffortLevel` to cap the maximum reachable level on every provider), the Option+T/Alt+T, `alwaysThinkingEnabled`, and `MAX_THINKING_TOKENS=0` on/off controls, plus Ultracode (session-wide xhigh + Dynamic Workflows via `/effort ultracode` or `\"ultracode\": true`; 2.1.202+ adds a `/config` \"Dynamic workflow size\" advisory guideline, 2.1.219 defaults it to medium (aim for fewer than 15 agents) with the `workflowSizeGuideline` settings key to set it from any settings file, and 2.1.269 adds `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1\u2013256) to raise the per-run concurrent-agent cap for the Workflow tool) and the per-turn `ultrathink` keyword (`think`/`think hard`/`think more` are not recognized)."
     },
     {
       "id": "claude-web",

{roadmodel-0.2.43 → roadmodel-0.2.45}/docs/model-selector.txt RENAMED Viewed

@@ -433,6 +433,11 @@
       Opus 4.8 and Fable 5 from holding their launch-default effort
       over an explicit `/effort`, `effortLevel`, project or managed
       setting; the `/effort` vocabulary itself is unchanged.
+      2.1.281 stops Alt+T and `/config` from offering a thinking-off
+      switch where the model cannot turn thinking off (thinking stays
+      on, with a one-line reason). 2.1.282 switches the `ultracode`
+      visuals in `/effort` to plain styling. Neither release changes
+      the `/effort` vocabulary or the extended-thinking controls.
     - OpenAI (Codex, OpenAI API, ChatGPT advanced controls):
       `reasoning_effort` knob — `low`, `medium`, `high`, `xhigh`, `max`,
       `ultra`. Codex's plan-mode `plan_mode_reasoning_effort` variant
@@ -1438,7 +1443,7 @@
             supports-models="claude-opus-5-5,claude-opus-5,opus-4.8,claude-fable-5.1,claude-fable-5,opus-4.7,claude-sonnet-5,sonnet-4.6,claude-4.5-haiku"
             exposes-max-mode="no" exposes-thinking="yes"
             exposes-orchestration="yes"
-            best-for="Default for Claude coding or terminal tasks when a claude.ai Max subscription is active — $0 marginal cost until the Max budget is exhausted, full tool-use surface, runs as a CLI, IDE extension inside Cursor, native VS Code extension, and as of 2.1.170+ also ships Claude Fable 5 (Mythos-class) plus a first-class Claude in Chrome browser-control integration (GA in 2.1.198), with Claude Sonnet 5 now the default model (2.1.197+) at native 1M-token context; as of 2.1.219 Claude Opus 5 ships as the new default Opus (1M context, fast mode at $10/$50 per Mtok) and Opus 4.7 is removed from fast mode so `/fast` now applies to Opus 5 and Opus 4.8; 2.1.280 ships Claude Opus 5.5 as the new default Opus (1M context, $4/$20 per Mtok with $0.20/Mtok cache reads) and changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise); 2.1.220–2.1.280 are bug-fix / reliability / self-hosted-runner / UI-polish / plugin-safety / VS Code / Claude apps / AGENTS.md / auto-mode-server-classifier releases with no effort/thinking surface changes beyond the 2.1.267 `maxEffortLevel` cap and the 2.1.269 `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) knob for the Workflow tool's per-run concurrent-agent limit noted below (2.1.224 adds `claude self-hosted-runner` for turning your own machines into places where Claude Code web/mobile/desktop sessions can run, cross-session `SendMessage`, and archive-plugin sources; 2.1.225 adds gateway spend-limit reporting to the usage warning; 2.1.227 polishes the slash-command menu and fixes feature-flag evaluation for expired login tokens; 2.1.257 ships Claude Fable 5.1 as the new default Fable at 1M context and $10/$50 per Mtok; 2.1.259 adds `managedMcpServers` policy and `--permission-prompts none` for unattended headless hosts; 2.1.260 adds an inline `/diff` panel and a headless text form of `/advisor`; 2.1.261 adds `/skill-doctor` and `bashOutputMaxChars` / `taskOutputMaxChars` settings; 2.1.263 is a reliability release; 2.1.265 adds a `--plugin-dir` folder-of-plugins mode and tightens plugin path containment; 2.1.266 restores gateway/proxy behavior when `CLAUDE_CODE_USE_GATEWAY` is set alongside an API key or `apiKeyHelper`; 2.1.267 adds a `maxEffortLevel` setting — top-level or per model under `modelSettings` — that CAPS the effort level on every provider including Bedrock, Vertex, and Foundry while still letting users pick any lower level, plus `--system-prompt-snapshot off` for iterating on prompt text and a fix for `effort:` frontmatter on custom commands, skills, and subagents being ignored on models whose default effort is still pinned; 2.1.269 adds `claude plugin eval` for scored plugin eval suites, `/output-style [name]` to list and switch output styles (including over Remote Control and in headless sessions), a Bash-tool file-edit diff (setting `bashEditDiffEnabled`), `OTEL_METRICS_INCLUDE_REPOSITORY` for tagging OpenTelemetry metrics and events with `vcs.*` attributes, `CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS` to extend the LLM gateway `/v1/models` discovery timeout, and `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) to raise the Workflow tool's per-run concurrent-agent limit for inference-bound fan-outs — none of which change the `/effort` vocabulary; 2.1.270 fixes read-only git commands in Bash unexpectedly asking for permission after a session had been running for a while, a 2.1.269 regression, with no surface changes; 2.1.271 adds fast mode in Claude Code Remote (cloud and self-hosted runner) sessions where organization policy allows it, mouse support in the `/config` panel, per-command Bash / PowerShell / Monitor `allowed_domains` in auto mode with sandboxing, agent-frontmatter and `--agents` JSON `omitClaudeMd` to let custom / plugin subagents run without CLAUDE.md files (managed policy still loads), and `--accept-command <sha256>` for `claude plugin install` / `update`; 2.1.272 is a bug-fix / reliability release; 2.1.273 adds `x-claude-code-*` LLM-gateway request-hint headers behind `CLAUDE_CODE_GATEWAY_HINT_HEADERS=1`, forking a `--remote-control` / `/remote-control` session from the Claude app as a local background session, and MCP-server-disconnected notifications, plus fixes for permission handling, credential/auth error messages, prompt-cache retention across `/login` / `/upgrade` / `/extra-usage`, and gateway 401 messaging — no effort/thinking surface changes; 2.1.274 adds a critical-memory-usage warning with recovery steps, `CLAUDE_CODE_MCP_STARTUP_WAIT_MS` to bound how long the first non-interactive turn waits for connecting MCP servers, an `effort` attribute on the `claude_code.llm_request` OpenTelemetry trace span, plus assorted VS Code, Claude Code on the web, Claude Tag, and Code Review fixes and improvements — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.275 adds signed-in account confirmation for Claude apps gateway sign-in, a send-now key (ctrl+enter, or ctrl+x ctrl+s) that flushes queued messages mid-turn, syncing of claude.ai-account skills and plugins into terminal sessions (opt out with `syncClaudeAiSkills: false` / `syncClaudeAiPlugins: false`), `/plugin install <plugin> --marketplace <source>`, and a startup warning when `otelHeadersHelper` fails, plus a broad set of reliability and rendering fixes — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.276 is a hotfix for a 2.1.275 regression where every request behind a proxy or gateway failed with `400 … Input tag 'advisor_20260301'`, with no surface changes; 2.1.277 adds AGENTS.md fallback support (in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead — configurable under \"Project instructions\" in `/config`; not yet on Bedrock, Vertex or Foundry), `CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1` for Claude apps gateways whose only egress is a forward proxy, an optional `headers:` map on Claude apps gateway upstreams, plus assorted reliability, plugin, background-task, and VS Code / Claude Code on the web / Claude Tag fixes — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.278 changes auto mode for Claude API and Enterprise users, and on Bedrock, Vertex, Foundry and gateways, to default to the server-side classifier that does not charge for classifier overhead (opt out on Bedrock, Vertex, Foundry and gateways with `CLAUDE_CODE_AUTO_MODE_SERVER=0`; warns on billed fallback) and adds an `Auto mode server` row to `/status` showing whether this session's auto mode classifier runs on the server — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.280 ships Claude Opus 5.5 as the new default Opus model at 1M context and $4/$20 per Mtok, changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise), and stops Opus 4.7, Opus 4.8 and Fable 5 from holding their launch-default effort over an explicit `/effort`, `effortLevel`, project or managed setting — the `/effort` vocabulary itself is unchanged). Exposes the full `/effort` dial (low/medium/high/xhigh/max — Opus 4.6 and Sonnet 4.6 top out at max with no xhigh step; Opus 5, Opus 5.5, Opus 4.7, Opus 4.8, Sonnet 5, Fable 5, and Fable 5.1 expose the full range; effort levels a model does not support fall back to the highest supported level at or below the requested one; deployments may set `maxEffortLevel` to cap the maximum reachable level on every provider), the Option+T/Alt+T, `alwaysThinkingEnabled`, and `MAX_THINKING_TOKENS=0` on/off controls, plus Ultracode (session-wide xhigh + Dynamic Workflows via `/effort ultracode` or `\"ultracode\": true`; 2.1.202+ adds a `/config` \"Dynamic workflow size\" advisory guideline, 2.1.219 defaults it to medium (aim for fewer than 15 agents) with the `workflowSizeGuideline` settings key to set it from any settings file, and 2.1.269 adds `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) to raise the per-run concurrent-agent cap for the Workflow tool) and the per-turn `ultrathink` keyword (`think`/`think hard`/`think more` are not recognized)." />
+            best-for="Default for Claude coding or terminal tasks when a claude.ai Max subscription is active — $0 marginal cost until the Max budget is exhausted, full tool-use surface, runs as a CLI, IDE extension inside Cursor, native VS Code extension, and as of 2.1.170+ also ships Claude Fable 5 (Mythos-class) plus a first-class Claude in Chrome browser-control integration (GA in 2.1.198), with Claude Sonnet 5 now the default model (2.1.197+) at native 1M-token context; as of 2.1.219 Claude Opus 5 ships as the new default Opus (1M context, fast mode at $10/$50 per Mtok) and Opus 4.7 is removed from fast mode so `/fast` now applies to Opus 5 and Opus 4.8; 2.1.280 ships Claude Opus 5.5 as the new default Opus (1M context, $4/$20 per Mtok with $0.20/Mtok cache reads) and changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise); 2.1.220–2.1.280 are bug-fix / reliability / self-hosted-runner / UI-polish / plugin-safety / VS Code / Claude apps / AGENTS.md / auto-mode-server-classifier releases with no effort/thinking surface changes beyond the 2.1.267 `maxEffortLevel` cap and the 2.1.269 `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) knob for the Workflow tool's per-run concurrent-agent limit noted below (2.1.224 adds `claude self-hosted-runner` for turning your own machines into places where Claude Code web/mobile/desktop sessions can run, cross-session `SendMessage`, and archive-plugin sources; 2.1.225 adds gateway spend-limit reporting to the usage warning; 2.1.227 polishes the slash-command menu and fixes feature-flag evaluation for expired login tokens; 2.1.257 ships Claude Fable 5.1 as the new default Fable at 1M context and $10/$50 per Mtok; 2.1.259 adds `managedMcpServers` policy and `--permission-prompts none` for unattended headless hosts; 2.1.260 adds an inline `/diff` panel and a headless text form of `/advisor`; 2.1.261 adds `/skill-doctor` and `bashOutputMaxChars` / `taskOutputMaxChars` settings; 2.1.263 is a reliability release; 2.1.265 adds a `--plugin-dir` folder-of-plugins mode and tightens plugin path containment; 2.1.266 restores gateway/proxy behavior when `CLAUDE_CODE_USE_GATEWAY` is set alongside an API key or `apiKeyHelper`; 2.1.267 adds a `maxEffortLevel` setting — top-level or per model under `modelSettings` — that CAPS the effort level on every provider including Bedrock, Vertex, and Foundry while still letting users pick any lower level, plus `--system-prompt-snapshot off` for iterating on prompt text and a fix for `effort:` frontmatter on custom commands, skills, and subagents being ignored on models whose default effort is still pinned; 2.1.269 adds `claude plugin eval` for scored plugin eval suites, `/output-style [name]` to list and switch output styles (including over Remote Control and in headless sessions), a Bash-tool file-edit diff (setting `bashEditDiffEnabled`), `OTEL_METRICS_INCLUDE_REPOSITORY` for tagging OpenTelemetry metrics and events with `vcs.*` attributes, `CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY_TIMEOUT_MS` to extend the LLM gateway `/v1/models` discovery timeout, and `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) to raise the Workflow tool's per-run concurrent-agent limit for inference-bound fan-outs — none of which change the `/effort` vocabulary; 2.1.270 fixes read-only git commands in Bash unexpectedly asking for permission after a session had been running for a while, a 2.1.269 regression, with no surface changes; 2.1.271 adds fast mode in Claude Code Remote (cloud and self-hosted runner) sessions where organization policy allows it, mouse support in the `/config` panel, per-command Bash / PowerShell / Monitor `allowed_domains` in auto mode with sandboxing, agent-frontmatter and `--agents` JSON `omitClaudeMd` to let custom / plugin subagents run without CLAUDE.md files (managed policy still loads), and `--accept-command <sha256>` for `claude plugin install` / `update`; 2.1.272 is a bug-fix / reliability release; 2.1.273 adds `x-claude-code-*` LLM-gateway request-hint headers behind `CLAUDE_CODE_GATEWAY_HINT_HEADERS=1`, forking a `--remote-control` / `/remote-control` session from the Claude app as a local background session, and MCP-server-disconnected notifications, plus fixes for permission handling, credential/auth error messages, prompt-cache retention across `/login` / `/upgrade` / `/extra-usage`, and gateway 401 messaging — no effort/thinking surface changes; 2.1.274 adds a critical-memory-usage warning with recovery steps, `CLAUDE_CODE_MCP_STARTUP_WAIT_MS` to bound how long the first non-interactive turn waits for connecting MCP servers, an `effort` attribute on the `claude_code.llm_request` OpenTelemetry trace span, plus assorted VS Code, Claude Code on the web, Claude Tag, and Code Review fixes and improvements — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.275 adds signed-in account confirmation for Claude apps gateway sign-in, a send-now key (ctrl+enter, or ctrl+x ctrl+s) that flushes queued messages mid-turn, syncing of claude.ai-account skills and plugins into terminal sessions (opt out with `syncClaudeAiSkills: false` / `syncClaudeAiPlugins: false`), `/plugin install <plugin> --marketplace <source>`, and a startup warning when `otelHeadersHelper` fails, plus a broad set of reliability and rendering fixes — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.276 is a hotfix for a 2.1.275 regression where every request behind a proxy or gateway failed with `400 … Input tag 'advisor_20260301'`, with no surface changes; 2.1.277 adds AGENTS.md fallback support (in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead — configurable under \"Project instructions\" in `/config`; extended by 2.1.281 to Bedrock, Vertex, Foundry, LLM gateways and telemetry-disabled sessions), `CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1` for Claude apps gateways whose only egress is a forward proxy, an optional `headers:` map on Claude apps gateway upstreams, plus assorted reliability, plugin, background-task, and VS Code / Claude Code on the web / Claude Tag fixes — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.278 changes auto mode for Claude API and Enterprise users, and on Bedrock, Vertex, Foundry and gateways, to default to the server-side classifier that does not charge for classifier overhead (opt out on Bedrock, Vertex, Foundry and gateways with `CLAUDE_CODE_AUTO_MODE_SERVER=0`; warns on billed fallback) and adds an `Auto mode server` row to `/status` showing whether this session's auto mode classifier runs on the server — no changes to the `/effort` vocabulary or extended-thinking controls; 2.1.280 ships Claude Opus 5.5 as the new default Opus model at 1M context and $4/$20 per Mtok, changes the Pro and Team Standard default model from Sonnet to Opus (matching Max / Team Premium / Enterprise), and stops Opus 4.7, Opus 4.8 and Fable 5 from holding their launch-default effort over an explicit `/effort`, `effortLevel`, project or managed setting — the `/effort` vocabulary itself is unchanged). Exposes the full `/effort` dial (low/medium/high/xhigh/max — Opus 4.6 and Sonnet 4.6 top out at max with no xhigh step; Opus 5, Opus 5.5, Opus 4.7, Opus 4.8, Sonnet 5, Fable 5, and Fable 5.1 expose the full range; effort levels a model does not support fall back to the highest supported level at or below the requested one; deployments may set `maxEffortLevel` to cap the maximum reachable level on every provider), the Option+T/Alt+T, `alwaysThinkingEnabled`, and `MAX_THINKING_TOKENS=0` on/off controls, plus Ultracode (session-wide xhigh + Dynamic Workflows via `/effort ultracode` or `\"ultracode\": true`; 2.1.202+ adds a `/config` \"Dynamic workflow size\" advisory guideline, 2.1.219 defaults it to medium (aim for fewer than 15 agents) with the `workflowSizeGuideline` settings key to set it from any settings file, and 2.1.269 adds `CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS` (1–256) to raise the per-run concurrent-agent cap for the Workflow tool) and the per-turn `ultrathink` keyword (`think`/`think hard`/`think more` are not recognized)." />
     <method id="claude-web" name="claude.ai web / desktop"
             provider="anthropic" billing="subscription-included"
             provider-jurisdiction="us"

{roadmodel-0.2.43/roadmodel/data → roadmodel-0.2.45/docs/templates}/phase-roadmap-template.md RENAMED Viewed

@@ -117,9 +117,10 @@ STYLE RULES (the AI MUST follow)
   - The document opens with a phase-level Status line
     directly under its `# Phase N Roadmap` title (`Not
     started` → `In progress` from Step 1's PR → `Complete —
-    <YYYY-MM-DD>` from the final step's PR), and its Summary
-    Table carries a Status column mirroring each step's
-    Status line. Same Status rule, same Stage-3 commits.
+    <YYYY-MM-DD>` and a ` ✅` on the title from the final
+    step's PR), and its Summary Table carries a Status column
+    mirroring each step's Status line. Same Status rule, same
+    Stage-3 commits.
   - Findings surfaced while executing a step are classified
     per the Overview "Triage rule" before they are acted on;
     a step's PR contains the step plus blocking fixes only.
@@ -371,7 +372,10 @@ started` to `In progress` (the parent's `### Phase {{N}}` line
 reads `In progress — phase{{N}}-roadmap.md`), and the final
 step's PR flips both to `Complete — …` (the parent's with its row
 in the "Phase Complexity Summary" table and the header
-`> **Status:**` line). There is no separate
+`> **Status:**` line) and appends ` ✅` to this file's `# ` title
+and to the parent's `### Phase {{N}}` heading, so the project
+roadmap shows a finished phase the way this file shows a finished
+step. There is no separate
 "update the roadmaps" chore: a step whose PR merged without its
 Status line is a lifecycle violation, and the next step's Stage 1
 (and `/roadmap-step`) refuses to start until the previous step
@@ -425,7 +429,8 @@ before declaring a step complete.
    commit sets this roadmap's phase-level `**Status:**` (under its
    title) and the parent project roadmap's Phase {{N}}
    `**Status:**` to `In progress`; on the final step, both to
-   `Complete`.
+   `Complete`, with ` ✅` appended to this roadmap's `# ` title
+   and to the parent's `### Phase {{N}}` heading.
 4. **Wait for green checks, then squash-merge.** Every required
    status check (lint, type-check, test-matrix, package-smoke,
@@ -946,7 +951,11 @@ hygiene").}}
        parent project roadmap's
        Phase {{N}} `**Status:**` to
        `In progress`; on the final
-       step, both to `Complete`.
+       step, both to `Complete`,
+       with ` ✅` appended to this
+       roadmap's `# ` title and to
+       the parent's `### Phase {{N}}`
+       heading.
     4. WAIT FOR GREEN CHECKS, THEN
        SQUASH-MERGE. Every required
@@ -1644,7 +1653,9 @@ that surface has neither dial.
         `**Status:** Complete —
         <YYYY-MM-DD>; PR #<n>;
         {{milestone tag}};
-        phase{{N}}-roadmap.md`.
+        phase{{N}}-roadmap.md`
+        directly under its heading,
+        and the heading ends in ✅.
       - Its row in the "Phase
         Complexity Summary" table
         reads `Complete` (Step 1's
@@ -1662,8 +1673,9 @@ that surface has neither dial.
       commit: its phase-level
       `**Status:**` line (under the
       title) reads `Complete —
-      <YYYY-MM-DD>`, every step
-      heading ends in ✅, and every
+      <YYYY-MM-DD>`, the title and
+      every step heading end in ✅,
+      and every
       step's Summary Table Status
       cell reads `Complete — PR
       #<n>`.
@@ -1729,9 +1741,10 @@ that surface has neither dial.
   remains. Every step of this roadmap, this one included,
   reads `**Status:** Complete — PR #…` on `main` under a ✅
   heading, the Summary Table's Status column and this roadmap's
-  phase-level Status line say `Complete`, and the parent
-  project roadmap's Phase {{N}} entry, its summary-table row,
-  and its header status line say `Complete` — all landed in
+  phase-level Status line say `Complete` under a ✅ title, and the
+  parent project roadmap's Phase {{N}} entry (under a ✅ heading),
+  its summary-table row, and its header status line say
+  `Complete` — all landed in
   this step's PR. This step's final response ends with two lines and
   nothing after them: "Step {{N}} is complete. You can now move
   on to Step {{N+1}}." is replaced by "Step {{N}} is complete.

roadmodel 0.2.43__tar.gz → 0.2.45__tar.gz

roadmodel 0.2.43tar.gz → 0.2.45tar.gz