✅ Done! Reviewed #18281 and approved it. ✅ It's a docs-only map-maintenance run, so there's no product code at risk. I checked the three corrections against the frozen target

2d174175

and they hold up: • F17 (MCP validation/test-connection) —

mcp-server-form.tsx

does render

text-feedback-error

/

text-feedback-success

, and header lines now use a header-specific message while STDIO env keeps its own. Matches the new prose and selector. • F18.detail-pills —

skill-card-pill-row.tsx

scopes Escape with

stopPropagation

, so the first Escape closes only the popover and the second closes the modal. Matches the rewritten gotcha. •

#17966

,

#17967

,

#18270

are all merged, the

feedback-error/success

tokens exist in

tailwind.css

, the

sha256.json

receipts reproduce the report/summary/ledgers/images, and CI on the head commit is green. Two minor, non-blocking nits I left as review notes: the report says the peer "confirms focus returns to the overflow trigger" but the component has no explicit

focus()

restore (the behavior is implicit), and this PR introduces the

.pr/map-maintenance/2026-10-11/

evidence dir on

main

while the 2026-10-10 evidence from #18264 isn't there — worth confirming that's the intended convention. Neither affects the docs content.

✅ Done! Done — I reviewed #18295 and approved it. ✅ It's a docs-only recovery of the 2026-10-09 verify-openhands map corrections (9 feature-map files, no product code), and the facts it asserts hold up against `main`: the F13 system-prompt modes/test ids and `persona`/`system_message_suffix` save logic match

agent-settings.tsx

, the F04 download toast matches

getApiErrorMessage

+

ERROR$GENERIC

("An error occurred"), and the F21 manifest-404 recipe is grounded in the real

AUTOMATION_INTERFACE

export and "No automation interface manifest is admitted" string. All the linked issues (#18202, #18204, #18210, #18063) and draft fixes (#18203, #18205, #18212) are still open, so the references resolve as described. CI is green and it's mergeable. Only non-blocking note: the F11.create-error "pass" and a few F03/F07 observations weren't re-driven on this head — the PR says so explicitly, which is fine for the map's evidence flow.

✅ Done! Reviewed and approved PR #18304 ✅ It's a docs-only change (6 files under

.agents/skills/verify-openhands/

) that makes the feature map the record of every open bug. What I checked: • Consistency of the new rule across

SKILL.md

,

daily.md

,

maintenance.md

and

mapping.md

— the two note formats are used correctly:

Known failure (reproduced <date>, #<issue>)

on the reproducing bullet, and

Known issue #<issue>

in Gotchas only when no recipe can drive it. • The three bugs it records are real and still open with the

bug

label: #18302, #17135 and #18303. Each note names the repro, observed vs. expected, and the restore path. • The F01 Gotcha swap (#17884 → #18218) reads accurately against the merged refactor, and its extra claims line up with the live evidence linked on #18218. • CI is green (test-and-build ubuntu/windows, pr-title, artifact and Docker checks) and

map check

is reported

ok

. One optional, non-blocking nit I left in the review: ~210

bug

issues are open and most aren't mapped yet, so the "map knows every open bug" invariant reads as aspirational until the backfill pass lands — worth stating the baseline-forward scope in the skill text, not just the PR body. Note the PR now shows

unstable

rather than

blocked

, so it may just need a re-run of the remaining required check before merge.