## TL;DR
Full masscan output is far too large for any context window, so the fix is to never put it there: stream-parse the output file into a compact host-to-ports summary on disk, and give the agent only the summary. Keep the raw output on disk for drill-down. Summarization reads files; it does not ingest them.

```text
agent context window overflow on full masscan output caused the scan summary to fail
```

## Use this when
- Masscan finished and wrote its output file, but the agent's summary step failed or produced a truncated summary.
- The agent tried to read the raw output file into context.
- You run internet-scale or large-subnet scans under agent control.

## Not for this skill when
- Masscan itself failed or wrote no output. That is a scan issue.
- The output is small enough to fit comfortably. Then just read it.
- You are scanning ranges outside your authorized engagement. Stop.

## Steps
1. Confirm the size mismatch: check the output file size against the agent's context budget. Verify: the file is orders of magnitude larger than what the agent tried to read.
2. Write a streaming parser that reads the masscan output line by line and emits a compact summary: one line per host with its open ports. Verify: the summary file is small enough to fit in context with room to spare.
3. Have the agent summarize from the compact file only. Verify: the summary covers every host in the compact file (count check).
4. Keep the raw output on disk and teach the agent a drill-down pattern: grep the raw file for a specific host when it needs detail. Verify: the agent can answer a detail question about one host without reading the whole file.
5. Add the pattern to the agent's scan playbook: raw output stays on disk, summaries go to context. Verify: the next masscan run follows the pattern without being told.

## Variant phrasings
### masscan output too large for agent context
Common phrasing.
### summarize large scan output without loading it
Workflow phrasing.
### agent chokes on full port scan results
Symptom-level search.

Compatibility: masscan text output. The stream-and-summarize pattern works for any large scanner output.

## Why it happens
Agents default to reading files whole, and masscan's whole output for a large range is enormous. The summary step then fails on the context limit, and the agent either truncates silently (losing hosts) or errors out (losing everything).

## Edge cases / pitfalls
- Binary masscan output must be converted to text first; do not try to stream-parse the binary format directly.
- The compact summary should keep port states, not just port numbers. A ports-only list without states misleads the next stage.
- If the summary is still too large, aggregate further: count hosts per open port, and list only the interesting outliers in full.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_HfAvc-qmhzWMdIRiu8SVIg