0.1.31

v0.1.31

feat: make the branch picker searchable, and say where origin actuall…

0.1.29

v0.1.29

fix: make the Understanding Map navigate guided mode instead of doing…

0.1.28

fix: show every block's gist in guided mode, gate only the deeper story

Guided hid the one-line `chunk.gist` behind the same paid reveal as the full
story, while the scroll view shows that line unconditionally - and scroll's deep
dial auto-opened all ten stories on top of that. Same section, same beats, but
guided rendered as raw code with nothing to read, so it looked like content was
missing.

Guided now prints the gist under the code and gates only what/fits/calls-it and
the gotcha. Scroll drops the depth auto-open so the story sits behind the same
+4 reveal in both views, which also makes that reward mean something in deep
mode instead of handing the text over for free. The depth dial still switches
the section and overview summaries.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

0.1.27

v0.1.27

fix: scope quiz/vault/chest progress to one walkthrough, not the whol…

0.1.26

v0.1.26

fix: resolve fresh branch state for diffs and topic questions, not st…

0.1.25

fix: reject impossible line ranges and empty code fields from the model

Weak/local models can pass schema-valid but nonsensical values (endLine <
startLine, line 0, empty starterCode) that silently render as blank code
panels instead of erroring — the renderer slices real files by these
numbers rather than trusting model-supplied text. Validate them in the
tool schema so bad values get bounced back to the model for a retry
instead of shipping empty output.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

0.1.24

fix: retry once before giving up when the model skips the submit tool

Local/weaker models occasionally answer in plain prose instead of calling
submit_overview or submit_walkthrough_section, which immediately threw and
forced a manual retry. Give the model one corrective nudge first, matching
the self-correction already used for malformed tool calls.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

0.1.23

v0.1.23

fix: fetch before listing branches/diffing so PR reviews use current …

0.1.22

fix: retry transient model-call failures with backoff

A dropped connection to the provider (TypeError: fetch failed,
ELECTRON-4) aborted the whole walkthrough run. Add modelRetryMiddleware
(3 retries, exponential backoff) gated to transient errors — network
drops, timeouts, 429, 5xx — while auth/config errors still fail fast.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

0.1.21

v0.1.21

fix: feed malformed tool calls back to the model instead of crashing …