Open questions

19 of 24 entries end with something unresolved. Each is quoted as its entry left it.

• whether the derived first-sentence summary is good enough in practice, or whether entries will need summary: written by hand often enough that the tool should prompt for it.

  No. 1 — The format is the asset, so the spec is written down first2026-09-20

• whether migrate can find prose corrections like entry 50's reliably enough to suggest supersedes edges, or whether that stays a manual pass.

  No. 2 — What hellbender's own log said about the spec2026-09-20

• whether any other project's index has drifted in ways this test would not catch, since hellbender is the only real log to test against until amber is migrated.

  No. 3 — The index generates, and hellbender's own index proves it2026-09-20

• whether the marker is discoverable enough. A project that never adds it never gets the generic half updated, and nothing will remind them - check does not currently look at the skill at all.

  No. 4 — The write path, and an init that nearly ate hellbender's skill2026-09-20

• whether per-entry dates can be recovered from git history - each heading first appears in some commit, and that commit has a date. It would turn 192 identical dates into something true, and nothing else can.

  No. 7 — Migrate, and the backup that never fired2026-09-20

• whether echoing the client's protocol version is the right hedge or a way to fail confusingly later, since it has only been driven by a client this repo wrote.

  No. 8 — Distribution, and a worklog a model can read2026-09-20

• whether the serif stack resolves well away from macOS. Iowan and Charter are Apple faces; elsewhere it should fall to Palatino or Georgia, but I have only looked at it on one machine.

  No. 9 — The viewer felt cheap, and the filters had never worked2026-09-20

• whether a question ever wants partially answering. The format allows one trailer per entry and one resolves closes it entirely, so a partial answer has to be a new entry with its own open question - which may turn out to be the right shape or may turn out to be a nuisance.

  No. 10 — A question can be answered, not just a claim overturned2026-09-20

• whether check should follow cross-entry links in prose and fail on one that points at no entry. It would want parsing markdown in the core, which so far has stayed out of it.

  No. 11 — The log's own links were dead on its own site2026-09-20

• whether the light theme holds up. Headless here inherits the system's dark mode, so every screenshot in this entry is the dark one, and the light palette has still never been looked at by anything.

  No. 13 — Working blind, and the render loop I should have had first2026-09-20

• whether the same is true of covers, which hellbender fills with glob-ish strings like ART*.ACT, FOG*.MAP and which cairns currently passes through untouched. It may want to be structured, or it may be prose that is better left alone.

  No. 14 — Hellbender's own docs tree crashed the renderer2026-09-20

• whether a source install should be able to know its version at all. git describe in a build script would do it, at the cost of a build script and of behaving differently inside and outside a git checkout.

  No. 15 — Two install instructions that were not true2026-09-20

• whether a reference page's covers deserves the same. It is in extra and passes through untouched; hellbender fills it with glob-ish strings like ART*.ACT, FOG*.MAP that say what a page accounts for, which is the same kind of claim files makes and is equally invisible.

  No. 16 — A field carried all the way through and shown nowhere2026-09-20

• whether the same argument applies to files, which is optional and which [[16]] has only just made visible. An entry about code with no files is as silently incomplete as one with no open question - but unlike the trailer there is no word for "this entry is not about any file".

  No. 17 — An entry that says nothing about what it does not know2026-09-20

• whether the generated formula should be checked by CI rather than only by me running brew install once. brew audit wants a tap, and a tap wants the formula pushed, so checking it before it is published is awkward - which is exactly when a broken formula is cheapest to catch.

  No. 18 — The tap is the repo2026-09-20

• whether check should also follow the filename form of a cross-entry link, which remains unvalidated. [[11]] raised that and only half of it is answered.

  No. 19 — References, and a parser that reads them in pieces2026-09-21

• whether a project will want an icon that is not in the ten - a Mastodon mark, its own logo. A path to an SVG file in the repository would cover it, but it would also mean the renderer reading a file, which it has so far never done.

  No. 21 — Links the generated nav could never know about2026-09-21

• whether five is the right point to fold - it is a guess, and a log with many three-to-seven-file entries would say.

  No. 23 — Areas in one line each and files that fold2026-09-24

• whether a model that is told about --body stops reaching for the stub - the skill says so, but only use will show it

  No. 24 — An entry written in one command, prose on stdin2026-09-24