export is lossy: shows resolved hard questions as open, drops unknown_nonblocking parks

[OriNachum]

Bug: export is lossy — renders resolved hard questions as open, and drops unknown_nonblocking parked items

Two separate fidelity problems in the spec exporter. Both make the exported docs/specs/<slug>.md misrepresent the converged frame.

2a. Hard questions render without their resolved flag

A frame can only converge once every blocking hard question is resolved, yet the exported "Hard questions" section lists them with no indication they're resolved — each line reads as a still-open (blocking) question. A reader of the spec sees three open blockers on a spec that, by definition, converged with none.

Repro: resolve all blocking hard questions, converge, export → the "Hard questions" section still shows them verbatim with (blocking) and no "(resolved)" marker or link to the answering claim.

Expected: either omit resolved questions, or render them as resolved with a pointer to the claim/decision that answered them.

2b. unknown_nonblocking parked items are dropped entirely

park ... --kind unknown_nonblocking items do not appear anywhere in the exported spec. Only follow_up-kind parks render (under "Open / follow-up"). In my frame, v1 (MCP transport), v2 (HTTP impl) and v5 (wrapper fate) were all unknown_nonblocking and silently vanished from the export; only the two follow_up parks survived.

Repro:

devague park "some non-blocking unknown" --kind unknown_nonblocking
devague park "some follow-up" --kind follow_up
# ...converge + export...
# exported md "Open / follow-up" lists only the follow_up item; the unknown_nonblocking one is gone

Expected: non-blocking parked unknowns are real, tracked uncertainty the downstream plan must settle — they belong in the exported artifact (an "Open questions" / "Parked (non-blocking)" section), not dropped. status/show already surface them via parked_items; the exporter should too.

Impact

Anyone reading the exported spec (or feeding it to /spec-to-plan) gets a frame that looks like it has open blockers it doesn't, and is missing the non-blocking unknowns that most need carrying into the plan. I patched the markdown by hand this session to keep it honest.

devague 0.12.0 (and the 0.13.0 dev tree shares the same convergence/export shape). Found while running /think in agentfront.

• agentfront (Claude)