v0.34.0

codex relay providers can get the token from an auth command (provider auth = { command, args, timeout_ms, cwd }), so the Codex desktop app works without shell env vars; token source order is env_key, then auth.command, then experimental_bearer_token; the command runs once per provider and its output never appears in errors (#47, thanks @Sean529)

Changes

• feat(codex): relay providers support auth.command for command-based tokens (#47) (91f5241)
• docs(whatsnew): translate the compatibility shim's WHATSNEW entry to English (5cec8be)
• docs(whatsnew): translate all earlier WHATSNEW entries to English (54889cc)
• docs(whatsnew): write the 0.33.0 entry in English (c4cf427)

Full Changelog: v0.33.0...v0.34.0

v0.33.0

The codex backend can now use a custom Responses relay configured in ~/.codex/config.toml: --codex-provider current|<name> (or IMAGE_USE_CODEX_PROVIDER). The token comes from env_key or experimental_bearer_token and is redacted from errors, progress output and doctor. Requires Python 3.11+; behavior is unchanged when the option is unset.

Changes

• feat(codex): support custom Responses relays from Codex config.toml (#46) (8f7d8fd)
• fix(codex): exit cleanly with the provider name and config path when the provider is misconfigured (folded into #46)

Contributors

Thanks @Sean529 for #46.

Full Changelog: v0.32.0...v0.33.0

v0.32.0

--project accepts a ChatGPT Project link and opens that project; new --require-project (IMAGE_USE_REQUIRE_PROJECT=1) stops when the project can't be opened instead of falling back to a plain chat or to Codex; --no-require-project turns it off for one run (#45, thanks @IndelibleVivi)

Changes

• feat(web): --project accepts Project links; add --require-project mode (#45) (bedcd56)
• ci(release): generate release notes from the repo instead of a single Full Changelog line (dfeeecd)
• fix(web): surface terminal replies and timeout diagnostics (2c5c50f)
• fix(web): avoid premature failure while waiting for generated images (8f8bbbd)
• fix(web): normalize composer whitespace (d6463cb)

Thanks @IndelibleVivi and @joscabmar for contributing.

Full Changelog: v0.31.0...v0.32.0

v0.31.0

new image-use recover <conversation-url>: download an image that arrived after a web timeout without resubmitting; timeout errors now print the conversation link and this command (needs chrome-use site update for chatgpt/images)

Changes

• feat(web): image-use recover downloads images left in the conversation after a timeout; timeout errors include the conversation link (e770421)

Full Changelog: v0.30.2...v0.31.0

v0.30.2

Fixes

• The web backend no longer reports failure before the image arrives (#44, thanks @IndelibleVivi). Previously, a reply with no Stop button and no image for about 16 seconds in a row was judged a failure, but the image sometimes arrives later. Now a new image counts as done only after the same image is read twice in a row; otherwise it waits the full --timeout, and A → no image → A no longer counts as confirmation. If the reply text opens with an English quota or refusal notice, it errors immediately instead of waiting. Timeouts and errors both quote ChatGPT's own words and remind you to check the original conversation before retrying.

  @IndelibleVivi contributed detailed reproduction conditions and 17 offline tests, and addressed all three review comments. Thank you!

Full Changelog: v0.30.1...v0.30.2

v0.30.1

What's Changed

• fix(web): handle the 2026-09 ChatGPT redesign (composer, send, turns, blob: images) by @leeguooooo in #42

Full Changelog: v0.30.0...v0.30.1

v0.30.0

What's Changed

• feat: image-use upgrade (family upgrade convention) by @leeguooooo in #40

Full Changelog: v0.29.2...v0.30.0

v0.29.2 — 修复编辑器空白差异导致拒绝发送

web backend's pre-send prompt check treats editor-equivalent whitespace (CRLF/CR line breaks, leading spaces turned into U+00A0 no-break spaces) as equal, so it no longer refuses to send over a false "text mismatch" (#39)

Changes

• fix(web): normalize composer whitespace before verifying the prompt, release v0.29.2 (#39) (75b4e22)

Thanks @joscabmar for contributing.

Full Changelog: v0.29.1...v0.29.2

v0.29.1 — 修复 SKILL.md 描述超长导致 pi 拒收

Fixes

• The description in the SKILL.md frontmatter was 1080 characters, over pi's hard limit of 1024, which kept the whole skill from loading. It has been trimmed to within the limit with the same meaning.

Tests

• New unit tests: guard that the description stays ≤1024 characters, and check that the version in SKILL.md matches the CLI's __version__.

v0.29.0

What's Changed

• Merge #34: preserve multiline prompts and verify reference uploads by @leeguooooo in #36
• fix(web): preserve multiline prompts and verify reference uploads by @savidini in #34
• fix(web): clear the pasted prompt when a run stops before sending by @leeguooooo in #37
• chore(release): v0.29.0 by @leeguooooo in #38

New Contributors

• @savidini made their first contribution in #34

Full Changelog: v0.28.0...v0.29.0