[bong-water-water-bong]

qodo-ai/pr-agent is now The-PR-Agent/pr-agent (same repository, transferred; PyPI pr-agent points there; latest release v0.49.0, 2026-10-25). This moves our review job to it.

Changes

• .github/workflows/pr-agent.yml: installs pr-agent[github]==0.49.0 into a per-version venv (~/pr-agent-venv-0.49.0) on the runner. The old command -v pr-agent check would have kept 0.43.0 forever after a bump.
• The log check also fails on Failed to process the command and Failed to get git provider (PR-Agent exits 0 on both).
• .pr_agent.toml: drops [openai] api_base/key. 0.49 ignores them in repo settings as host-only and logs a warning; the endpoint already comes from the workflow env (OPENAI__API_BASE, OPENAI__KEY), as the usage guide describes.

Measured on the strixhalo runner box (workflow env, local coder-llm on :8090, CONFIG__PUBLISH_OUTPUT=false, target PR #1224):

install	result
pr-agent==0.49.0	Git provider 'github' could not be loaded because module 'github' is not installed → Failed to process the command, exit 0 (the old check passed it)
pr-agent[github]==0.49.0	review ran (6053 tokens, full diff), Review output is not published, no errors

This PR's own pr_agent run is the first live run of the new version.

🤖 Generated with Claude Code

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 11657ad)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 2 🔵🔵⚪⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Versioned Venv Usage The workflow now uses a versioned virtual environment (~/pr-agent-venv-0.49.0) for PR-Agent installation, which is good for avoiding stale dependencies. However, the version is hardcoded in the workflow file, making it prone to drift if not updated alongside the PR-Agent version bump. This could lead to inconsistencies if the version in the workflow is not kept in sync with the actual version being used. PRA_VERSION=0.49.0 venv="$HOME/pr-agent-venv-$PRA_VERSION" export PATH="$venv/bin:$PATH" if ! "$venv/bin/pr-agent" --help > /dev/null 2>&1; then python3 -m venv "$venv" "$venv/bin/pip" install -q "pr-agent[github]==$PRA_VERSION" fi Log Check Coverage The updated log check now includes more error patterns, such as "Failed to get git provider", which is important for catching failures specific to the new GitHub provider requirement. However, the regex pattern might be too broad and could potentially mask other legitimate errors or warnings that are not related to the core functionality failure. # PR-Agent exits 0 even when every model call failed or the command never ran, so check its log run() { pr-agent --pr_url "$pr" "$@" 2>&1 | tee "$RUNNER_TEMP/pr-agent.log" if grep -qE "Failed to generate prediction|Error during LLM inference|Failed to (review|describe|generate|process the command)|Failed to get git provider" "$RUNNER_TEMP/pr-agent.log"; then echo "::error::PR-Agent failed (no model answer, or the command never ran): see the log above"; exit 1 fi Configuration Clarity The comment explaining that [openai] api_base/key are ignored since PR-Agent 0.49 is helpful, but the configuration file itself no longer contains these settings. This change is correct, but it's important to ensure that all users or systems relying on this configuration file are aware of this change and that the new configuration is properly documented. # own OpenAI-compatible server, no cloud API. The endpoint (OPENAI__API_BASE, OPENAI__KEY) is set in the # workflow's env: since PR-Agent 0.49 an [openai] api_base/key in this file is ignored as host-only.

[github-actions]

Persistent review updated to latest commit b2b5364

[github-actions]

Persistent review updated to latest commit 11657ad