fix(describe): drop the blank card and keep diagrams left-to-right

[huhamhire]

Two regressions from the pr-agent 0.45.0 upgrade (#225). Both come from new upstream defaults I failed to pin, and neither is reported as an error by anything — they were found by looking at a real result.

1. Every /describe result grew a blank card

0.45.0 stamps <!-- pr-agent-generated --> at the top of its output (0.39.0 did not). That lands in the leading, title-less section, whose body therefore counts as non-empty — but an HTML comment renders to nothing, so the card came out blank.

Reproduced from the reported output: 6 findings parsed, the first being key=general, title=null, body exactly that comment.

trimNoise already discarded blank lines, horizontal rules and the internal branch-name leak; it now also discards a line that is only an HTML comment, which lets the existing "no title and no content → skip" rule cover this. 6 findings → 5, the rest unchanged. Two regression tests, including the case where the comment sits next to real prose and the section must be kept.

2. Diagrams turned top-down

0.45.0 added pr_description.pr_diagram_direction='adaptive', which draws any chain longer than pr_diagram_direction_threshold (5) nodes top-down. The reported diagram had 6.

Pinned to LR, verified against the installed runtime. Upstream tunes that heuristic for a full-width page; here the diagram is read in a narrow chat column, where a tall diagram pushes the rest of the result off screen, while a wide one scrolls sideways and can be opened in the preview view.

What I got wrong in the upgrade review

#225 pinned three new defaults, using one test: does it append to the output we parse? That test is incomplete. An upstream default can be wrong here in two distinct ways:

• it adds output — which lands as a bogus finding;
• it changes presentation for a surface upstream does not know about — pr-agent assumes a full-width page, this app renders into a narrow column.

The diagram default is the second kind, so it passed the only check I applied. 05-pragent-runtime.md now states both, so the next upgrade is screened against both.

Also worth noting for the next upgrade: neither defect surfaced in lint / typecheck / test / build, nor in the runtime probes I ran for #225 (imports, patch application, config overrides). Both needed a real /describe result to be looked at.

Verification

lint / typecheck / test / build pass; the parsing fix was verified by re-running the reported output through the parser, and the diagram pin by reading the setting back out of the embedded runtime under the override.

🤖 Generated with Claude Code