📖About

Author two single-file reference skills in the Sage-owned skill library so all Carol agents discover them via the shared work/.claude/skills symlink (no per-agent grant; the allow-list scans for SKILL.md). (1) agent-centric-design: the standard for agent-centric modular architecture (Design 146), the app-to-agent shim boundary (Design 173), and the shared-vs-owned test (Policy P.03.01.01.23), plus the compliance-checker command. A reference standard agents consult before designing or refactoring (Albus, Sage, Orion). (2) run-orion-bypass: the operator procedure for an end-to-end Orion bypass on top of the shared/bypass.py engine — file-planned, bypass_start with mandatory remediation linkage and the or-bx-01 droid tag, record activities, the INI-716 caller-audit waiver, bypass_end, and recover-from-blocked. Neither is a four-role build procedure; both are single SKILL.md reference skills.

⚖️Decisions

• Auto-detected remediation target INI-1002595 from title/description scan (matched CAROL-INI-0716 -> row id 1002595 (CAROL-INI-0716-00: Caller-audit gate — Elrond refuses contract-change closes wit)); override by setting remediates_initiative_id explicitly at bypass_start. (system-auto-detect)
• Caller audit waived by Orion: INI-900 only added two SKILL.md files under the Sage skill library; it did NOT modify shared/bypass.py. The INI-716 entry-point flag is the known fleet-wide uncommitted-edit false-positive cleared by the hourly Shipper auto-commit. No bypass-runtime behaviour changed. — Entry-point gate false-positive; skill-only change. (Orion)

✅Success criteria

• Both skills exist as folders with a SKILL.md and are discovered by the skill allow-list scan (must_have)
• agent-centric-design encodes the shared-vs-owned test and cites Designs 146/173 and Policy P.03.01.01.23 (must_have)
• run-orion-bypass encodes the bypass lifecycle including remediation linkage, caller-audit waiver, and blocked-recovery (must_have)