[aw] Prompt Optimization reported incomplete result

未关闭
  • 哪里不对？
#306 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
4/5
预计耗时
3-5 天
新手友好度
35/100
Issue 类型
缺陷
描述清晰度
需要澄清
活跃度
活跃
技术栈
github-actions, javascript, ollama
领域
ai, ci-cd, devops

调研方向

Start with .github/workflows/prompt-optimization.md and the failed run details, then read .github/skills/agentic-workflows/SKILL.md or the linked debugging guide. Inspect how the workflow provides Ollama, its model cache, GGUF proxies, and prompt-optimizer.mjs; done means the evaluation and scoring steps can run successfully in the workflow.

由索引模型根据 Issue 内容生成。

描述

agentic-workflows
Workflow Failure

Workflow: Prompt Optimization
Branch: main
Run: https://github.com/githubnext/gh-aw-wizard/actions/runs/36779212881

[!WARNING]
Task Could Not Be Completed: The agent reported that the task could not be performed due to an infrastructure or tool failure.

Reasons:

  • The claimed prerequisite infra (restored Ollama cache, downloaded GGUF proxies, running Ollama server at 127.0.0.1:11434) did not actually exist; I could not run the hill-climbing optimization loop.

    No open PR titled "Prompt optimization:" existed, so the guard step was clear to proceed.

Attempting the described setup: ollama was not on PATH (only present at /opt/hostedtoolcache/ollama/0.33.2/x64/ollama, not installed to PATH). No ~/.ollama cache existed (no models, no blobs). No .gguf files were present anywhere on disk. Port 11434 had nothing listening (connection refused).

I started ollama serve manually to probe further:

  • It reported "total blobs: 0" — the model cache was empty, not restored.
  • GPU/inference discovery failed: "llama-server binary not found" (checked all standard paths) — the ollama installation is missing its inference backend (llama-server), so even with models present, standard chat completions via /v1/... would not work.
  • Attempting ollama pull hf.co/unsloth/SmolLM2-360M-Instruct-GGUF:Q4_K_M failed: "Get (hf.co/redacted) Forbidden" — outbound access to Hugging Face is blocked/forbidden in this sandbox, so I could not download the GGUF proxies myself as a workaround.

Given eval/scoring requires --eval-url (127.0.0.1/redacted) with both the Qwen2.5-1.5B and SmolLM2-360M GGUF proxies actually loadable and inferable, none of the prompt-optimizer.mjs --evaluate/--scoresteps could run. I made no repository content changes (only harness-owned metadata reformatting appeared in.github/skills/optimize-scenario-prompt/SKILL.md`, which I did not author and left untouched), and I did not create a pull request or review comment.

This is a structured incompletion signal (report_incomplete), not a real task outcome. Any other safe outputs emitted alongside this signal (e.g., comments) describe the failure state, not a completed review or action.

Action Required

Assign this issue to an agent to debug and fix the issue.

Debug with any coding agent

Use this prompt with any coding agent (GitHub Copilot, Claude, Gemini, etc.):

Debug the agentic workflow failure using https://raw.githubusercontent.com/github/gh-aw/main/debug.md

The failed workflow run is at https://github.com/githubnext/gh-aw-wizard/actions/runs/36779212881
Manually invoke the agent

Debug this workflow failure using your favorite Agent CLI and the agentic-workflows prompt.

[!TIP]

Stop reporting this workflow as a failure

To stop a workflow from creating failure issues, set report-failure-as-issue: false in its frontmatter:

safe-outputs:
  report-failure-as-issue: false

Generated from Prompt Optimization · copilot · 32.9 AIC · ◷

主要语言
JavaScript
星标
6
派生
1
平均合并
22 小时 10 分钟
30 天内合并 PR
10

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南，环境需要你自己搭建：先看它的 README，通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

githubnext/gh-aw-wizard 的其他 Issue

查看 githubnext/gh-aw-wizard 的全部 Issue

相似的 Issue

更多 JavaScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。