Bump Pi SDK dependencies (pi-ai, pi-agent-core, pi-web-ui) to 0.52.12
Session from pi-for-excel-sessions on Hugging Face by thomasmustier · MIT · condensed by Coders Talk
Click a segment to jump to that moment
19msession
0interventions
GOAL
Audit which Pi packages the Excel add-in depends on and what changed upstream since the current pinned version, then update the dependencies to the latest release.
- +0Prompt · +0Check what we import from Pi and what's changed upstream in the new version, reading pwd + ~/pi-mono
- +0mAgent did · +0mInvestigated declared vs installed Pi package versions, diffed changelogs and model registry between v0.52.5 and current pi-mono HEAD across ai/agent/web-ui packages, catalogued all imports used in src/
- +6mPrompt · +6mGo ahead and implement the update, open a PR, handle the automated Codex review, then merge when green; also look for refactor opportunities afterward
- +6mAgent did · +6mSet up an isolated git worktree tracking origin/main and ran npm install to bump pi-agent-core, pi-ai, pi-web-ui to ^0.52.12
- +6mFail · +6mnpm install didn't actually update package.json/lockfile versions; a follow-up node check to verify installed versions failed with module not found
- +7mAgent did · +7mManually edited package.json version ranges to ^0.52.12, re-ran npm install, verified lockfile diff, ran model-ordering tests, lint/typecheck/build checks, updated docs/model-updates.md verification date, then committed
- +8mFail · +8mgh pr create command failed because the multi-line body string with backticks was mis-parsed by bash, causing package names to be interpreted as commands
- +8mAgent did · +8mFound the PR had already been created (#260) with a mangled body, wrote a proper body to a temp file and used gh pr edit --body-file to fix it
- +9mAgent did · +9mPolled the PR for automated Codex review activity per the requested process: waited 240s then polled every 60s, found all CI checks green and mergeStateStatus CLEAN with no reviewer activity after ~6 minutes of polling
- +19mOutcome · +19mPR #260 opened with dependency bump to 0.52.12, all checks green (CLEAN merge state), no automated review appeared within the polling window
VERDICT
I'd start the next prompt by giving it the exact gh pr create body as a file from the start, since inline multi-line strings with backticks in bash keep breaking the command. The worktree + isolated branch approach worked well for avoiding interference with parallel work, and running the full check/build/test suite before committing caught nothing broken, which gave me confidence to let it proceed to PR and merge unattended.