How can I find an unreachable commit hash in a GIT repository by keywords?

When you delete a branch, you also delete its reflog. There's a separate reflog for HEAD that will retain a reference to commits that were on the deleted branch, but only if you've had them checked-out.

The difference between --lost-found and --unreachable is subtle:¹ see the git glossary, and/or the illustration below. In general, using --lost-found and/or --unreachable will find such commit(s) (and with --lost-found, also write IDs into the .git/lost-found/commit directory, which I think has the side effect of protecting them from garbage collection).

In this particular case, the commit you were looking for was not the tip-most commit of the deleted branch. That is, suppose before deleting feature/something we have this, with the two most recent commits made on the feature branch:

A <- B <- C   <-- master
  \
    D <- E    <-- feature/something

Now we delete feature/something, losing the IDs of commits E and D both. Both IDs will show up in the output of git fsck --unreachable, but only E's ID will show up (and be saved) by git fsck --lost-found, because commit D is "reachable" from E if/when you restore that commit.

Finding your commit

how could I have found this lost commit by keyword?

It's a bit tricky. Probably your best bet is using git show on all unreachable commits, something like:

git show $(git fsck --unreachable | git cat-file --batch-check |
    awk '/commit/ { print $3 }')

Now you can search for the keyword(s) in the log messages (or the diffs). The internal $(...) sequence is the method for extracting all the candidate IDs: we just want commits, not tags, trees, and blobs. Once you have the IDs, all regular git commands (git log -1, git show, etc) can be used with those.

---

¹In fact, I just learned it myself writing up this answer.