GanyuanRan/Aegis v2.12.1

v2.12.1: Codex skill-read evidence

on GitHub

Codex skill-read evidence

• Reject failed Codex SKILL.md reads instead of treating an attempted command
  as proof that a skill loaded.
• Require a successful command result containing matching skill name and
  description frontmatter before reporting a load event.
• Add regression coverage for failed reads, parallel completion order,
  mismatched frontmatter, and failed-then-successful retries. Historical
  successful Codex logs remain recognized.

Product relationship and distribution

• Introduce Autoloom in the English and Chinese public entrypoints and clarify
  the boundary between the independently usable Aegis Method Pack and the
  separate product carrying the runtime-core direction into execution.
• Synchronize all eight package and plugin version fields at 2.12.1. Update
  through the existing host installation route and restart or reload as that
  host requires. No manual routing-prefix re-copy is needed.
• Keep issues #67, #68, and #75 open under their documented evidence limits;
  this release does not represent those bounded follow-ups as resolved.

Validation scope

• bash tests/e2e/run-all.sh --full --host-profile fast passed in isolated
  Linux: Layer 1 43/0, governance completion contract passed, and Layer 2 and
  Layer 3 each 6/0.
• Both representative Codex smokes passed on Codex CLI 0.146.0 with successful
  skill-read command results; all 20 parser unit tests passed.
• Version drift and undeclared-version audits passed for all eight package and
  plugin fields.
• GitHub Actions CI passed for release commit 4edf34e.