Skip to content

v0.5.0

Latest

Choose a tag to compare

@thomas-endres-tng thomas-endres-tng released this 10 Nov 13:18
· 0 commits to main since this release

Please CLI v0.5.0

Release notes for v0.5.0 — changes since v0.4.4.

✨ New features

  • Clipboard without xclip: please now falls back to xsel when xclip is missing on X11, so clipboard support no longer requires a specific tool. On Wayland, nothing changes (wl-copy).
  • Configurable completion budget: The switch of the default model to a reasoning model led to an increased token-usage for explanation requests, which can lead to empty responses. The API budget for explanations and Q&A answers can now be adjusted via the new environment variable PLEASE_MAX_COMPLETION_TOKENS (default: 1000).
  • Truncation warning: when a model response is cut off because the completion budget is exhausted (reasoning models can spend it all before producing output), please now explains the cause and points at the fix instead of silently returning an empty answer.
  • Improved key storage: please -a now verifies that the key actually round-trips through the system keychain and reports failures loudly (non-zero exit, stderr message with actionable guidance) instead of claiming "API key stored successfully" when secret-tool is missing or the write failed.

📦 Packaging

  • Drop hard dependency on xclip (Debian): xclip moved from Depends to Suggests (Suggests: xclip, xsel, wl-clipboard), matching the Arch optdepends. Machines that cannot or should not have xclip installed can now install please at all — previously this was impossible (#67).
  • The Arch PKGBUILD now annotates xclip as preferred and xsel as fallback for X11 clipboard support.

🛠 Fixes & behavior changes

  • please -a no longer reports success when the key could not be stored: missing secret-tool, a failed write, or a failed round-trip lookup now exit non-zero with a diagnostic on stderr. Upgrade note: scripts/automation that relied on the previous (silent) success behavior on machines without libsecret-tools should now expect a failure.
  • The API request budget no longer uses the deprecated max_tokens parameter inconsistently; the budget for explanations and questions is unified (previously an unmotivated 500/1000 asymmetry). Command generation is left unrestricted to avoid new truncating commands.
  • Error messages from the store flow go to stderr, keeping stdout clean for piping.

📝 Documentation

  • README overhaul: fixed the secret-tool example (it previously stored the key under attributes please could never find back), migrated API-key instructions from the long-dead beta.openai.com to platform.openai.com/api-keys, fixed a broken code fence that rendered the Parameters section as code, copyright year bump, typo fixes.

🔧 Internal

  • New bats test suites for the completion budget, the validation, and the secret-tool store path — including platform-aware tests (Linux-only tests run in CI, skipped on macOS) and stub-based testing of the keychain interaction.
  • ShellCheck compliance across the script; DCO-signed commits throughout.

🙏 Credits

Thanks to @tom-mi for #68 (drop the hard xclip dependency) and for thorough review feedback on #74/#75.


Full changelog: v0.4.4...v0.5.0