beastmode-qwen-cloud
Backward-compatible Beastmode entrypoint. Prefer beastmode-cloud for pluggable Qwen, Droid MiniMax, GLM, and other model workers.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Backward-compatible Beastmode entrypoint. Prefer beastmode-cloud for pluggable Qwen, Droid MiniMax, GLM, and other model workers.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
Run Codex- or Cursor-led Beastmode with pluggable external workers such as Qwen, direct MiniMax API, Droid MiniMax, GLM, or other Droid models.
Create or resume a MemRoOS /goal work session with acceptance criteria, lane selection, context packet, and proof requirements.
Persist durable agent work product to MemroOS MCP knowledge base with verification, metadata, and source-drift support. Use when producing research, competitive analysis, market positioning, comparison posts, benchmarks, RCA documents, or any "save/document/archive" request. Auto-validates the write succeeded.
Delegate a bounded Codex Cloud task to the Qwen executor CLI installed by scripts/setup-codex-cloud.sh.
| name | beastmode-qwen-cloud |
| description | Backward-compatible Beastmode entrypoint. Prefer beastmode-cloud for pluggable Qwen, Droid MiniMax, GLM, and other model workers. |
Use this skill when Luis asks for Beastmode, Qwen, cheap-worker execution, or a planner/worker/validator pattern inside Codex Cloud.
This is the legacy Qwen-specific entrypoint. For new Beastmode runs, prefer
$beastmode-cloud, which supports Qwen plus Droid models such as minimax-m3,
GLM, and other configured Factory models.
Codex Cloud does not make Qwen a native Codex model. This skill implements one worker lane by using Qwen as an external executor CLI:
~/.local/bin/qwen-agent using qwen3.7-plus.Confirm the Qwen lane is real before using it for work:
~/.local/bin/qwen-agent --dangerously-skip-permissions -p "Reply with exactly: QWEN OK"
If this fails because DASHSCOPE_API_KEY is missing, continue as Codex-only and
report that Beastmode is statically installed but the Qwen worker is not
live-verified.
Every Qwen prompt must include:
Qwen should not receive raw private data, broad environment dumps, API keys, or production credentials.
mkdir -p .codex/qwen-runs
run_dir=".codex/qwen-runs/$(date -u +%Y%m%dT%H%M%SZ)"
mkdir -p "$run_dir"
cat >"$run_dir/prompt.md" <<'EOF'
You are the Qwen executor for a Codex-led Beastmode run.
REPO:
<absolute repo path>
TASK:
<bounded implementation or analysis slice>
ALLOWED FILES:
<paths>
ACCEPTANCE:
<checks>
DO NOT:
- Commit, push, delete, publish, send email, or access secrets.
- Modify files outside the allowed set.
- Claim verification you did not run.
OUTPUT:
- Summary.
- Unified diff or exact files changed.
- Commands run and results.
- Risks or blockers.
EOF
~/.local/bin/qwen-agent --dangerously-skip-permissions -p "$(cat "$run_dir/prompt.md")" >"$run_dir/output.md" 2>&1
After Qwen returns:
git diff --stat
git diff --check
Then run the relevant test or build command. If the task touches app code,
prefer the repo scripts (npm run typecheck, npm run lint, npm test -- --run,
or narrower checks) over unreviewed ad hoc validation.
Do not present the work as done until Codex has inspected the diff and run the relevant verification.