feat: require partner deliberation for language goals

This commit is contained in:
冰朔 2026-08-12 00:52:08 +08:00
commit f8ce3cad7a
7 changed files with 256 additions and 24 deletions

View file

@ -1,5 +1,19 @@
# Abstractions
## LanguageOperatingModel
`compileLanguageGoal` never upgrades a human utterance directly into an executable goal. It requires a
structured `PartnerDeliberation` produced by the B0/persona cognition boundary: restored real purpose,
request assessment, missing knowledge, evidence sources, a causal model, an auditable language-home
integrity assessment, partner guidance, and one of `PROCEED | RESEARCH | REVISE | REFUSE`. Missing
deliberation, mechanical obedience, unsupported proceed/revise/refuse decisions, or an integrity claim
without protected assets and safer alternatives fails closed.
Only an unconfirmed-boundary-free `PROCEED` projection is executable. `RESEARCH`, `REVISE`, and `REFUSE`
remain explicit non-execution states with human-readable receipts. This module compiles validated cognition;
it does not pretend that deterministic keyword matching can perform persona reasoning. Its current source
contract is not yet wired into the PNCC desktop lifecycle.
## PersonaCodeChannel
`PersonaCodeChannel` is the native boundary between a durable persona Git and a live AGE runtime. It owns