Warning: This build has drifted.
The coverage report for this pull request build may be inaccurate because its base commit is no longer the HEAD of its target branch.
This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

    • Learn more: For more information on this, see Tracking coverage changes for pull request builds.
    • Fix now: For a quick fix, rebase this PR at GitHub. Your next report should be accurate.
    • Prevent going forward: To avoid this issue with future PRs, see these Recommended CI Configurations.

fix(transcript): match transcript by prompt prefix for large pasted prompts

Catalog.Select identifies the turn's transcript by requiring the file to
contain the entire injected prompt verbatim. Large prompts are pasted into
Claude (max-wpm-size threshold), and the terminal's paste handling can subtly
transform the pasted body (control/whitespace bytes in fetched source), so the
stored transcript copy is not byte-identical to the injected prompt. The
whole-prompt match then fails, Select never finds the transcript, and the turn
hangs until turn-timeout with no output — even though Claude completed the turn
and wrote a terminal result event.

Observed with a ~45KB headless prompt: the stored user message differed from
the injected prompt by ~144 bytes deep in the source body, while the prompt
header and tail stayed intact. Small (typed) prompts match exactly and were
unaffected.

Match on a bounded prompt prefix (256 runes) in addition to the full prompt.
The prefix is injected before any paste drift accumulates and stays a stable,
distinctive anchor (a typical headless prompt's task header and source URL fall
within it), so Select locates the transcript reliably. Short prompts keep their
existing exact-match behavior.