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
Enforce mode no longer hard-blocks on eval(/credential text in non-code
files. The archetype-independent eval-call and secret-detected-in-content
rules ran on raw content, so a literal eval( or a credential-shaped token in
markdown / plain-text / config PROSE (e.g. documentation that explains the rules,
or a CHANGELOG entry) was treated as a runnable sink. Under the new enforce
default this turn-trapped a session with no escape — a non-code file cannot carry
an inline // chameleon-ignore directive. eval-call is now gated to recognized
code languages, and a new block_eligible_on_file gate drops the
archetype-independent rules from the BLOCK set on any detect_language()-None
file (they remain advisory). Applied at every block/arming site, including the
with-archetype paths reachable by a legacy extension-blind paths_pattern.
Enforcement on real code (.ts/.js/.rb/.py) is unchanged: a real eval()/exec() or a committed credential still blocks.