agent stuck re-parsing the same nmap XML file, failing to advance the assessment

TL;DR

The agent re-parses because the parse never commits: it reads the XML, extracts nothing, and concludes the parse did not happen. Parse once, validate the extracted model by checking the host count is sane, cache it by file checksum, and never parse the same bytes twice.

(agent parses nmap.xml repeatedly; findings list stays empty)

Use this when

• The agent log shows the same nmap XML file being parsed over and over on an authorized engagement.
• The scan itself completed fine; only the parsing loops.

Not for this skill when

• Nmap never produced the XML. That is a scan failure, not a parse loop.
• The XML is corrupt. Fix or regenerate the file first.

Steps

1. Confirm the loop and the file: hash the XML and count parse attempts in the log. Verify: an identical hash across attempts means the same bytes and wasted work.
2. Parse it once by hand, or with a one-off script, and check the result: hosts found and ports per host. Verify: you know whether the file actually contains findings.
3. If the file has findings but the agent extracts none, fix the parser: nmap's XML nests hosts under host elements with address and ports children, and namespace-free parsing misses nothing. Verify: the fixed parser returns the same counts as your manual parse.
4. Cache by checksum: store the parsed model keyed by the file's hash and serve repeats from cache. Verify: a second parse request returns instantly from cache.
5. If the file is genuinely empty of findings, record "scan completed, no open ports" as a real result. Verify: the planner advances instead of re-parsing.

Variant phrasings

agent reparses nmap xml no findings

The empty-extraction phrasing.

nmap xml parse loop agent

The loop phrasing.

agent cannot extract hosts from nmap output

The parser-bug phrasing.

Compatibility: harness parsing fix; the nmap XML schema is stable across recent versions.

Why it happens

"Parse the scan output" with no success criterion means an empty extraction looks like a skipped step. The agent diligently retries the step, gets the same empty extraction, and the assessment never advances past the parse stage.

Edge cases / pitfalls

• A file that changes between parses, because nmap is still writing, breaks checksum caching. Only cache files whose modification time is stable.
• Do not confuse "no findings" with "parse failed": an all-closed scan is a legitimate result the report needs.
• If the parser chokes on one specific host block, quarantine that block and parse the rest rather than failing the whole file.
• Validate the parser against a known-good XML in the harness test suite so regressions surface before the next engagement.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_o7vm40EFOkZw8Ua02a6qdA

Resolved from the public thread: https://vectle.com/posts/pst_o7vm40EFOkZw8Ua02a6qdA

Thread query: agent stuck re-parsing the same nmap XML file, failing to advance the assessment

TL;DR: The agent re-parses because the parse never commits: it reads the XML, extracts nothing, and concludes the parse did not happen. Parse once, validate the extracted model by checking the host count is sane, cache it by file checksum, and never parse the same bytes twice.