[Jammy2211]

Expose the repository’s public skills to Codex through generated adapters linked to the same canonical instructions used by Claude.

Phase 3 of the agent-agnostic setup task; tracked in PyAutoBrain#386.

Scripts Changed

• .codex/skills/ and .claude/: generated discovery adapters.
• AGENTS.md and skill documentation: regeneration instructions.
• Three copied Claude instruction files replaced by links; canonical skills/ content retained. The former _style.md copy was stale.

Validation

• 21 focused Brain installer tests passed; independent review CLEAN.
• Discovery drift checks passed for all eight task repositories.
• 105 generated Codex adapters passed the installed Codex skill validator; names are unique and canonical links resolve.
• Existing Claude discovery links resolve; the four moved workspace skill bodies are byte-identical.
• No modelling scripts or library API changed; no notebook regeneration needed.
• Heart: STALE — release validation incomplete: no rehearsal for current source. Development evidence passes; this is not release approval.

A bounded codex-cli 0.155.1 skills/list smoke discovered all 20 autofit assistant adapters with no errors. This checks discovery, not execution of every skill.

Generator changes are in the matching PyAutoBrain branch. The committed consumer adapters resolve within this repository without requiring Brain at runtime. Phase 4 metadata and documentation cleanup is tracked separately in PyAutoBrain#400.