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。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/thewoolleyman/cxdb-graph-ui --skill cxdbstatus命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
基于 SOC 职业分类
| name | cxdb:status |
| description | Query CXDB to show pipeline context status, detect stuck agents, error loops, and suggest fixes. |
| user-invocable | true |
Show a real-time status summary of all CXDB contexts for active Kilroy pipeline runs. Detect problems (error loops, stale agents, failed branches) and suggest fixes.
The CXDB API base URL is http://$KILROY_CXDB_HOST:9120 (nginx frontend).
curl -sf http://$KILROY_CXDB_HOST:9120/v1/contexts
If the request fails (connection refused, timeout, non-200), report:
CXDB is not reachable at
http://$KILROY_CXDB_HOST:9120. Start it with:script/start-cxdb.sh
If the response is an empty list, report:
No active contexts in CXDB. Nothing to report.
Then stop.
For each context returned, fetch the 10 most recent turns:
curl -sf http://$KILROY_CXDB_HOST:9120/v1/contexts/{CONTEXT_ID}/turns?limit=10&order=desc
Look at the client_tags on each context to find the Kilroy run ID (typically formatted as kilroy/<ULID>). Then locate the pipeline DOT file at the repo root:
pipeline.dotRead the DOT file to get the list of node names so you can map node_id values from turns to human-readable pipeline stages.
For each context, extract from the most recent turns:
data.node_id — current pipeline stagedata.branch_key — parallel branch identifierdeclared_type.type_id — turn type (e.g., ToolCall, ToolResult, GitCheckpoint, Text)data.is_error — whether the turn is an errordata.status — status field (e.g., pass, fail)data.tool_name — which tool was calleddata.output — output text (scan for validation failures)Present a table with one row per context:
| Context | Node | Branch | Status | Last Activity | Depth/Head |
|---|
Where:
node_id mapped to the DOT file node namebranch_key value (or — if not on a branch)running, error, complete, stale, disconnectedScan all contexts for these issues:
Consecutive is_error: true turns on the same node_id. This often indicates a tool schema mismatch or validation failure (e.g., repeated write_file validation errors).
Contexts with status: live but no turn activity in the last 5 minutes.
GitCheckpoint turns with status: fail — indicates a branch did not pass its gate check.
Contexts that appear disconnected (no longer streaming) but were expected to still be active based on pipeline progress.
For each detected problem, print a warning block with a suggested fix:
Error loop on node X:
Agent is stuck in an error loop on node
X. The tool{tool_name}is repeatedly failing. Fix: Stop the run with/kilroy:status→ Stop, check the tool schema or input validation, then re-run.
Failed branch B at checkpoint:
Branch
Bfailed its git checkpoint. The consolidation step will proceed without this branch's changes. Fix: Review the branch output for errors. You may need to manually apply the intended changes.
Stale context (no activity for >5 min):
Context
{id}appears stale — no turns in the last 5 minutes. Fix: Check if the agent process is still running. If not, resume with/kilroy:status→ Resume.
All branches complete but consolidator not started:
All parallel branches have completed but no consolidation context is active. Fix: The pipeline may need a manual restart. Use
/kilroy:status→ Resume.
After the report, suggest relevant actions:
/kilroy:status to stop the run, fix the issue, and re-run."http://$KILROY_CXDB_HOST:9120 for the full CXDB console."/kilroy:land to merge and push."