Files
msd-core/tests
Tom Boucher e0e7531530 fix(#4469): anchor beginPhaseCore's focusPattern regex to line start (#4577)
* fix(#4469): anchor beginPhaseCore's focusPattern regex to line start

focusPattern (src/state-transition.cts, beginPhaseCore's first-time-
execution branch) rewrites the **Current focus:** body line via a regex
with no line anchor and \s* (crosses newlines) instead of a same-line
whitespace class -- the same defect class already fixed for a different
function, stateReplaceField, in #4243/PR #4453. A bold **Current focus:**
quoted mid-sentence anywhere in the body (e.g. an Accumulated Context
bullet documenting the field) matched first and had the rest of its line
silently overwritten with the new focus label -- the #4010 data-loss
class, applied to a different field.

Anchored using the exact same idiom PR #4453 established:
^([ \t]*\*\*Current focus:\*\*[ \t]*)(.*)$ with the /m flag. [ \t]* (not
\s*) avoids consuming the newlines before the label into the match; $
documents the match ends at end-of-line.

Added a regression test to tests/state-transition.test.cjs: a body with
NO real **Current focus:** field yet present, but a mid-sentence prose
mention of the same bold text in an Accumulated Context bullet -- proving
the unanchored pattern would corrupt the prose (empirically verified via
direct node -e execution before wiring the test) while the anchored
pattern's .test() correctly returns false and leaves it untouched.

stateReplaceProgressPercent's bold branch (~line 51-71) is explicitly NOT
touched -- the issue itself flags it as entangled with the recorded #2177
form-priority decision, needing a maintainer call.

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

* docs(#4469): backfill changeset PR number

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

---------

Co-authored-by: sim <sim@local>
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-09-09 14:02:59 -04:00
..