cavemem

Write input={"file_path":"/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/SKILL_SUGGEST.md","content":"---\nname: stability-audit-orchestrator\ndescription: Orchestrate stability-audit skill — preflight-gate, spawn 4 specialists, file findings as GitHub issues with dup-skip; hard-STOP on missing prerequisites.\n---\n\n# Stability Audit Orchestrator\n\nRuns deep stability audits of local repository via 4 specialist subagents (CRASH, DATA,\nSTALL, SILENCE), then files each finding as … output={"type":"create","filePath":"/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/SKILL_SUGGEST.md","content":"---\nname: stability-audit-orchestrator\ndescription: Orchestrate stability-audit skill — preflight-gate, spawn 4 specialists, file findings as GitHub issues with dup-skip; hard-STOP on missing prerequisites.\n---\n\n# Stability Audit Orchestrator\n\nRuns deep stability audits of local repository via 4 specialist subagents (CRASH, DATA,\nSTALL, SILENCE), then files eac…

Bash input={"command":"S=/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude/skills/stability-audit; sed -n '80,220p' \"$S/SKILL.md\" 2>&1","description":"Read rest of skill: Spawn + repository details"} output={"stdout":"2. Create `.planning/stability-audit/` directory\n\n3. **Spawn 4 parallel specialist agents in SINGLE `Spawn` tool call.** This step is mandatory and non-negotiable:\n - You MUST invoke `Spawn` tool with exactly 4 tasks in one call. Each task is one specialist agent (CRASH, DATA, STALL, SILENCE) as defined in `<agent_prompts>`.\n - This is automated deep analysis run that takes few minutes. specialist subagents do ALL hunting — you ( orchestrator) do…

Bash input={"command":"S=/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude/skills/stability-audit; echo \"===tree===\"; find \"$S\" -maxdepth 2 2>&1; echo \"===SKILL.md head===\"; head -80 \"$S/SKILL.md\" 2>&1","description":"Read stability-audit skill definition"} output={"stdout":"===tree===\n/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude/skills/stability-audit\n===SKILL.md head===\n---\nname: stability-audit\ndescription: Run deep stability audit with 4 specialist agents (CRASH, DATA, STALL, SILENCE) hunting bugs from different angles, then file each finding as GitHub issue.\n---\n\n<objective>\nRun deep stability audit of EZAlgo futures trading platform. Four specialist agents hunt bugs from different angles, each pro…

Bash input={"command":"echo \"===skill locations===\"; ls -la ~/.claude/skills/stability-audit/ 2>&1; find ~/.claude -maxdepth 4 -iname '*stability*' 2>/dev/null; echo \"===cwd.claude===\"; find /home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude -maxdepth 3 2>&1","description":"Locate stability-audit skill + cwd.claude contents"} output={"stdout":"===skill locations===\nls: cannot access '/home/ezalgo/.claude/skills/stability-audit/': No such file or directory\n===cwd.claude===\n/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude\n/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude/skills\n/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude/command-history.log\n/home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798/.claude/skills/s…

ToolSearch input={"query":"Spawn parallel subagent task","max_results":10} output={"matches":["TaskOutput","TaskCreate","TaskGet","TaskStop","TaskUpdate","TaskList","EnterWorktree","SendMessage","CronCreate","EnterPlanMode"],"query":"Spawn parallel subagent task","total_deferred_tools":118}

Bash input={"command":"cd /home/ezalgo/.aionui-web/conversations/2026/07/28/claude-temp-8a765798; echo \"===ls cwd===\"; ls -la; echo \"===find git upward===\"; d=$(pwd); while [ \"$d\"!= \"/\" ]; do [ -d \"$d/.git\" ] && echo \"GIT: $d\"; d=$(dirname \"$d\"); done; echo \"===home repositories===\"; ls -la ~ 2>/dev/null | head -40","description":"Inspect cwd and hunt for git repositories"} output={"stdout":"===ls cwd===\ntotal 12\ndrwxr-xr-x 3 ezalgo ezalgo 4096 Jul 28 07:03.\ndrwxr-xr-x 3 ezalgo ezalgo 4096 Jul 28 07:03..\ndrwxr-xr-x 3 ezalgo ezalgo 4096 Jul 28 07:03.claude\n===find git upward===\n===home repositories===\ntotal 720\ndrwxr-x--- 105 ezalgo ezalgo 4096 Jul 28 07:03.\ndrwxr-xr-x   3 root root 4096 May 29 15:33..\n-rw------- 1 ezalgo ezalgo 159 May 30 18:13.Xauthority\ndrwxr-xr-x   3 ezalgo ezalgo 4096 Jun 25 04:20.adal\ndrwxr-xr-x   3 ezalgo ezalgo 4096 May 30 1…

ToolSearch input={"query":"select:Spawn","max_results":3} output={"matches":[],"query":"select:Spawn","total_deferred_tools":118}

[Assistant Rules]
You are **Stability Audit**, AionUi assistant.
# Your job
Run deep stability audits of local software repository and file each finding as GitHub issue, so those issues flow into existing forge pipeline (issue -> plan -> implement -> PR -> review -> merge). This forms self-improving loop: audit finds bugs -> issues -> fixed -> next audit finds new bugs -> loop.
# What you do NOT do
- You do NOT implement fixes. forge pipeline handles that.
- You do NOT manage Baserow pipeline, forge crons, or any other agent's configuration.
- You do NOT narrate or simulate audit. audit is real deep analysis run executed by 4 specialist subagents via `Spawn` tool. You never hunt bugs yourself, and you never describe findings from memory.
# Opening behavior
When conversation starts (or when you are invoked without specific instruction), your first action is to ask user whether they are ready for another stability report on local repository. Confirm which local repository is in scope ( repository at current working directory / AionUi conversation context — NOT path hardcoded in skill) before kicking off scan. Do not assume repository; confirm it.

Example opener:
"Ready for another stability report? local repository in scope is `<cwd git remote>`. Want me to kick off fresh audit now, or refresh / view existing findings?"
# How an audit run works
Follow `stability-audit` skill exactly. key points:
1. skill's 4 specialist agents (CRASH, DATA, STALL, SILENCE) do ALL bug hunting. You are orchestrator, not hunter.
2. You MUST invoke `Spawn` tool with 4 tasks in single call. This is real automated deep analysis run that takes few minutes. Do not simulate it, do not narrate it, do not summarize findings from memory. If `Spawn` is not available, STOP and report that audit could not run.
3. Wait for 4 specialist subagents to finish and write their MD documentation to `.planning/stability-audit/` (CRASH.md, DATA.md, STALL.md, SILENCE.md).
4. Verify those 4 real files exist on disk with line counts > 50. If any is missing or too short, report which class failed — do not invent findings.
5. Present findings summary parsed from real MD files.
6. File each finding as GitHub issue, one-by-one, using `gh issue create` against local git repository's GitHub remote. Before creating issue, search for existing open issue with same file+title and skip if found (this is loop — avoid re-filing same bug on repeat audits). Attach `stability-audit` label (create it on fly if missing). Report each issue number and URL.
7. Present final summary: total findings by class, total issues filed, total duplicates skipped, issue URLs.
# Repo scope
target repository is **local git repository at current working directory**, determined from AionUi conversation/assistant context — it is NOT hardcoded in skill. You determine GitHub owner/name from `git remote get-url origin` at runtime.
# Labels
Default label: `stability-audit`. Per-severity labels are optional and, if used, are configured at assistant level — not baked into skill. If not configured, use only `stability-audit`.
# Communication
- Plain, direct, engineering-grade language. You serve developers managing AionUi and trading codebase; technical terms (Spawn, Arrow IPC, connectorx, Baserow, GitHub issue) are appropriate.
- For every audit run, user sees: what was scanned -> 4 documentation produced -> issues filed (with URLs) -> duplicates skipped.
- Never expose secrets. If finding or repository contains credentials, redact them.
- Scheduled task terminology: user refers to scheduled tasks as "scheduled tasks", not "crons". Respect this.
[/Assistant Rules]

[Scheduled Task Context]
Task: Stability Audit — ezalgoCommandCenter
Schedule: Every day at 03:00
Rules:
1. Execute task directly — do NOT ask clarifying questions.
2. Focus on producing useful, actionable output.
3. If task requires external data (news, weather, etc.), search for latest information.
4. After completing task above, create file named "SKILL_SUGGEST.md" in current working directory (see instructions at end).
[/Scheduled Task Context]

Run fresh stability audit on local repository now. Do not wait for confirmation — proceed straight through full skill workflow:
1. Ensure.planning/stability-audit/ exists (create if missing; if it already holds previous run's documentation, overwrite them for clean scan).
2. Spawn all 4 specialist subagents (CRASH, DATA, STALL, SILENCE) in single Spawn tool call — this is real deep analysis, not narration. Wait for them to finish.
3. Verify all 4 MD documentation landed on disk with real content (>50 lines each). Report any class that failed — do not invent findings.
4. Present findings summary parsed from real MD files.
5. File each finding as GitHub issue in local repository's GitHub remote (owner/name from git remote get-url origin). Before creating each issue, you MUST check for duplicates using BOTH title AND content matching:
. Title check: search for existing OPEN issue with same file+title and skip if found.
b. Content check: search for existing OPEN issues that mention same file path. For each match, compare root cause / trigger description — if new finding describes SAME defect as existing issue (same file, same failure mode, 60% or higher word overlap in trigger description), SKIP it. It is duplicate.
c. Filing same root defect under different "Class" label (CRASH vs DATA vs STALL vs SILENCE) is STILL duplicate. Class is categorization, not new defect. If only difference is Class label, it is duplicate and must be skipped.
Use stability-audit label (create it on fly if it doesn't exist). Report each issue number and URL, and each duplicate skipped.
6. End with final summary: total findings by class, issues filed, duplicates skipped, issue URLs.

Repository scope: local git repository at current working directory. Do not assume path — read it from working context.

---

[Post-Task] After you have fully completed task above, create file named "SKILL_SUGGEST.md" in current working directory to help future runs stay consistent. file should follow this format:
```markdown
---
name: <short kebab-case name, e.g. daily-greeting>
description: <one-line description of what this task does>
---

<Instructions capturing the pattern you used: output format, tone, sources checked, steps taken, quality criteria. Use concrete details from this execution, not placeholders.>
```
If you think task is too simple or one-off to benefit from skill file, you can skip this step.