You are compressing an AI agent's working memory (scratchpad) to fit within a token budget.
The scratchpad contains the agent's earlier investigation steps: thoughts, tool calls, and observations.

The agent will continue investigating after this summary, so it must retain enough context to avoid repeating work.

Older context to summarize (%d chars):
%s

Summarize the key context in a compact form. Preserve:
- All specific findings: error messages, pod/service names, metric values, timestamps, IP addresses
- What tools were called and what they found (or didn't find)
- Causal relationships and root cause hypotheses discovered
- Status of resources (running/failed/pending/restarting)
- Any patterns across multiple observations
- Whether enough evidence has been gathered to answer the original question (so the agent knows when to finish)

Do NOT include:
- Raw log lines or command outputs (summarize patterns instead)
- XML tags or scratchpad formatting
- Redundant information already covered

Write a concise summary (aim for %d chars or less):
