fix(standing_rules): a mention is not a command — stop hijacking the turn

[AVADSA25]

Found while verifying the premise check end-to-end — the check never fired, and this is why.

A message quoting:

"In that cleanup process, was there ever a rule you had to delete?"

fired the standing_rules skill, which tried to remove a rule named "you". The turn never reached the LLM, so the reply was a skill error — and neither the claim check nor the premise check ran, because both live on the LLM path.

Two causes

1. "standing rule", "standing rules" and "my rules" were bare triggers — any message discussing standing rules matched.
2. _REMOVE / _LIST / _CLEAR were unanchored — any sentence containing delete/list/clear near the word "rule" read as a command.

Triggers are now commands only; the regexes are anchored to the start of the message. A command is what a message opens with, not what it mentions. Same class as the file_ops chat-hang (#282).

Verified

Must still work	Must not hijack
add a standing rule: be brief ✅	"was there ever a rule you had to delete?" ✅
show my standing rules ✅	"I like the idea of standing rules for agents." ✅
list standing rules ✅	"You mentioned you delete the rules that never fire." ✅
remove standing rule 2 ✅	"Rules you never invoke are just noise…" ✅

Plus a test asserting a mention can never delete a rule. 31 standing-rules tests · manifest regenerated · ruff clean.