Agent Skills: Show Work

Show Work

Present investigation findings as structured markdown directly in the conversation — never as a file.

Output Format

For each piece of evidence, show the exact command run and its full unedited output using markdown code blocks. Structure the response as:

1. Title — a short heading describing what was investigated
2. Evidence sections — each section has:
   • A markdown heading describing what this check proves
   • The command run, shown as $ command inside a code block
   • The full output immediately below in the same code block
3. Conclusion — a short paragraph at the end tying the evidence together

Rules

• Output everything as markdown directly in the chat response — do NOT write files
• Show command output fully when practical — truncation is acceptable provided it does not hide information important to the investigation
• Prefix each command with $ so it reads like a terminal session
• Use fenced code blocks with appropriate language hints (yaml, json, etc.) when the output is structured
• Group related commands under descriptive headings (e.g., "### No active pods" not "### Step 3")
• If $ARGUMENTS is provided, investigate that topic. Otherwise, compile evidence from commands already run in the current conversation.
• Re-run commands if needed to get fresh output — do not rely on stale results
• Include timestamps where relevant (e.g., job start/completion times)
• If a command fails or returns empty, include that too — absence of results is evidence