You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Changed: removed wording in the skill that broke the skill's own rules. The
12 document rules in SKILL.md lost their bold lead-ins. Rule 11 lost its
"not X, but Y" title. The References list, the rule-catalog headings, and
four cells in word-swaps.md lost their em-dashes. The rule content did not
change.
Removed: filler and claims with no source. SKILL.md lost "a tired
mechanic", "Nothing else in this file is optional", and "Read them last,
apply them first". use-cases.md lost "a stressed reader at 2 a.m." and "STE
cuts the error rate and the cost". rule-catalog.md lost "The ones agents
break are 1.7, 1.11, and 1.13". The SKILL.md example is now labeled "AI
output" and not "real AI output", because no committed eval file holds it.
Changed: the sentence "The same rule covers a fact, not just a word" in
SKILL.md, prompts/system-prompt.md, and output-styles/simple-english.md
now starts "Also name the host". The two mirrors stay byte-identical.
No benchmark ran on this release. The published figures still describe
2.0.1.