[bong-water-water-bong]

Adds PR-Agent (0.43.0) to the engine repo.

• Model: Qwen3-Coder-30B-A3B-Instruct UD-Q5_K_XL, served by 1bit serve on Vulkan, 64K context. It runs on strixhalo as the coder-llm user service at 127.0.0.1:8090.
• Runner: self-hosted strixhalo-engine (label pr-agent). No code leaves the box, and there is no API key or Docs7 quota.
• Opened or updated PRs get a review. Owners, members and collaborators can comment /review, /describe, /improve or /ask <question>.
• .pr_agent.toml carries engine-specific review instructions: correctness on NPU/HRX/Vulkan, C++ ownership, measured claims only, the Apache notice, and docs updated in the same PR.

This PR itself is the first test run.

🤖 Generated with Claude Code

[context7]

Docs7 for 1bit-monster/engine

Result	Status	Action
Deployment	➖ Not used	—
Content review	➖ Did not run. This site has no agent runs available this month. Wait for the monthly reset or check your Docs7 plan.	—

Commit 4dca26c

[github-actions]

Preparing review...

[github-actions]

PR Reviewer Guide 🔍

(Review updated until commit 4dca26c)

Here are some key observations to aid the review process:

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review Potential Timeout Issue in Model Readiness Check The model readiness check in the workflow uses a loop with a 10-second sleep between attempts. If the model takes longer than 300 seconds (30 attempts * 10 seconds) to become ready, the workflow will fail. This could be problematic if the model takes longer to load, especially in a resource-constrained environment. Consider increasing the timeout or making it configurable. ready=0 for i in $(seq 1 30); do if curl -sf --max-time 60 http://127.0.0.1:8090/v1/chat/completions -H 'Content-Type: application/json' \ -d '{"model":"qwen3-coder","messages":[{"role":"user","content":"ok"}],"max_tokens":1}' > /dev/null; then ready=1; break fi sleep 10 done [ "$ready" = 1 ] || { echo "::error::the local model (coder-llm on :8090) does not answer"; exit 1; } Possible Command Injection Vulnerability The command parsing logic in the workflow directly uses printf '%s' "$COMMENT_BODY" without sanitizing the input. If a malicious user crafts a comment with a specially crafted payload, it could potentially lead to command injection. While the workflow uses sed and case to parse commands, it's still risky to trust user input directly. Consider adding input validation or sanitization. cmd=$(printf '%s' "$COMMENT_BODY" | sed -E 's/^[[:space:]]*\/?//; s/[[:space:]].*//' | tr '[:upper:]' '[:lower:]') case "$cmd" in review) run review ;; describe) run describe ;; improve) run improve ;; ask) run ask "$(printf '%s' "$COMMENT_BODY" | sed -E 's/^[[:space:]]*\/?ask[[:space:]]*//')" ;; *) echo "not a PR-Agent command: $cmd" ;; esac

[github-actions]

Persistent review updated to latest commit 4dca26c