Codex Capability Registry v0.2.0

What changed

• Narrowed automatic skill activation to explicit missing, unavailable, authentication-blocked, unhealthy, repeated-discovery, or fallback symptoms.
• Made capability lookup compact by default; use show --verbose for the complete cached entry.
• Kept schema-1 cache compatibility with v0.1.
• Added a deterministic, redacted context-overhead benchmark.

Measured fixture overhead

Artifact	Estimated tokens
Skill metadata	52
Skill body	238
Scan output	39
Compact single-capability output	38
Verbose single-capability output	107
Full disk cache	161

The full disk cache is not automatically loaded into model context. These figures use ceil(characters / 4) and measure deterministic plugin artifacts; they do not prove universal end-to-end token savings.

Verification

• 15 unit tests passed
• Codex Skill validator passed
• Codex Plugin validator passed