The helper's emitted POST example still timed out against LOST112 in live
verification because the 20-second timeout budget was too short for the
observed form-submit path. Raise the generated timeout to 60 seconds,
lock that contract in the regression, and align the skill/doc wording so
users receive guidance that matches the runnable command.
Constraint: LOST112 live POST latency can exceed 20 seconds even when reachability probes succeed
Rejected: Keep 20 seconds and rely on manual retry | still ships a broken default curl_example
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: If the timeout budget changes again, rerun the exact emitted curl_example against LOST112 rather than only the lightweight reachability probe
Tested: python3 -m unittest scripts.test_subway_lost_property
Tested: python3 scripts/subway_lost_property.py --station 강남역 --item 지갑 --days 14 --verify-live
Tested: npm run ci
Not-tested: Long-term LOST112 latency drift beyond the currently observed 60-second budget
The review found that the generated subway lost-property curl example still timed out even when the helper's reachability probe passed. This change updates the helper to emit the command shape that actually succeeds against LOST112, tightens the regression test to lock that contract, and aligns the published skill/docs text with the runnable output users now receive.
Constraint: LOST112 POST flow requires a referer and does not behave reliably with the prior redirect-following command shape
Constraint: Keep the v1 scope hybrid/guidance-only without adding new dependencies or scraping guarantees
Rejected: Only add a referer header while leaving the rest of the curl shape unchanged | still reproduced timeouts with the generated command during verification
Rejected: Add a live network regression test for the emitted curl command | would make CI flaky and violate deterministic regression expectations
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: If the LOST112 request shape changes again, re-verify the emitted curl command end-to-end before updating docs or tests
Tested: python3 -m unittest scripts.test_subway_lost_property; python3 scripts/subway_lost_property.py --station 강남역 --item 지갑 --days 14 --verify-live; generated curl_example executed via shell and wrote lost112-search-result.html; npm run ci
Not-tested: Alternate LOST112 item/station combinations beyond the verified 강남역/지갑 example
Issue #98 was approved as a conservative implementation: structure the official LOST112 and 서울교통공사 workflows instead of pretending a stable public API exists. The new helper builds the documented search payload, the docs explain the v1 hybrid scope, and regression coverage locks the helper plus repo docs into the shipped flow.
Constraint: Public subway lost-property APIs remain unconfirmed, so v1 must stay guidance-first
Rejected: Full automated scraping of live result tables | unstable without confirmed API/session guarantees
Confidence: high
Scope-risk: narrow
Reversibility: clean
Directive: Keep this skill on official HTTPS entry points until a stable public API is verified end-to-end
Tested: python3 -m unittest scripts.test_subway_lost_property; python3 scripts/subway_lost_property.py --station 강남역 --item 지갑 --days 14 --verify-live; npm run ci
Not-tested: Browser-only user journey through LOST112 search submission beyond payload generation
