|
37 | 37 | `sandbox` before `main`; verify a sandbox review persists MCP findings and |
38 | 38 | a stalled stream reports a timeout rather than waiting for the job limit. |
39 | 39 |
|
40 | | -## 0.5.0 candidate |
41 | | - |
42 | | -- **Why a minor bump (0.5.0):** the release removes public surface rather than |
43 | | - adding models. Caret ranges on `^0.4.x` do not pick it up; consumers opt in. |
44 | | -- **Breaking:** |
45 | | - - The built-in provider set narrows to the 18 providers in |
46 | | - `packages/cli/src/provider/catalog.json`, shipped in the binary; models.dev |
47 | | - is no longer fetched or cached (#202). Other providers must be defined under |
48 | | - `provider` in `aictrl.json`. GitHub Copilot is removed (#193). |
49 | | - - Commands removed: `github`, `pr` (#192) and `acp` (#204) print a |
50 | | - replacement hint and exit 1; `debug lsp`, `debug snapshot` and |
51 | | - `models --refresh` are gone; `run --share` and `import <url>` fail (#195); |
52 | | - `run --agent plan` fails (#162). |
53 | | - - Subsystems and tools removed: LSP (#191), snapshots/revert/session diffs |
54 | | - (#184), session sharing (#195), the file watcher (#204), `question` and |
55 | | - plan mode (#162), `todowrite`/`todoread` (#183). |
56 | | - - Prompt changes: the `bash` description has no git procedures or rules, and |
57 | | - no default deny rules ship; callers use `permission.bash` (#161). Headless |
58 | | - runs no longer discover `~/.claude`, `~/.agents` or `~/.opencode` skills |
59 | | - (#165). |
60 | | - - Config keys `lsp`, `permission.lsp`, `snapshot`, `share`, `autoshare`, |
61 | | - `enterprise` and `watcher` still load as warned no-ops. |
62 | | -- **NDJSON v1 additions:** `session_error` reason `usage_limit`, code |
63 | | - `PROVIDER_USAGE_LIMIT` and `resetAt` (#174); `structured_output_rejected` |
64 | | - keyword `rejected` (#179). |
65 | | -- Stored sessions from 0.4.x still load and continue (covered by e2e tests in |
66 | | - #183, #184, #191, #195). |
67 | | -- Evidence (0.4.8 vs the candidate, GLM-5.3, 8 headless tasks × 3 runs, same Bun): |
68 | | - - pass rate 24/24 on both; no retries, hangs or `session_error`; |
69 | | - - uncached input tokens −25%, tool definitions −47% (36.8k → 19.4k chars), |
70 | | - no title request; output tokens median +5% per run; |
71 | | - - trivial run 558 → 389 ms, peak RSS 270 → 174 MB, binary 114 → 105 MB, |
72 | | - no network at startup. |
| 40 | +## 0.5.1 candidate |
| 41 | + |
| 42 | +- **Why a patch bump (0.5.1):** model-only support change. It restores the |
| 43 | + built-in `google` provider (Google AI Studio, `@ai-sdk/google`) that 0.5.0 |
| 44 | + left out of the static catalog, so `google/<model>` works with only |
| 45 | + `GOOGLE_GENERATIVE_AI_API_KEY`, `GEMINI_API_KEY` or `GOOGLE_API_KEY` set. |
| 46 | +- The catalog gains one provider and 22 models (18 → 19 providers, 1354 → 1376 |
| 47 | + models); every other provider's entry is byte-identical to 0.5.0. |
| 48 | +- No CLI flags, config schema or NDJSON events change. One config behaviour |
| 49 | + changes: a `provider.google` defined in `aictrl.json` on 0.5.0 now extends |
| 50 | + the built-in, so its models without `npm` use `@ai-sdk/google` again (as up |
| 51 | + to 0.4.8) and the built-in models are listed too. The CHANGELOG gives the |
| 52 | + `npm` + `whitelist` opt-out. |
73 | 53 |
|
74 | 54 | Publication and executor promotion are separate gates. Preparing this candidate |
75 | 55 | PR does not publish npm packages or deploy an executor image. |
0 commit comments