[Portability benchmark] Run Doubt on Claude Code

[alsoleg89]

Part of #9. This issue covers one client only: Claude Code.

Contribution

Run the frozen Agent Skills portability protocol with the unmodified Doubt v0.5.0 skill and synthetic fixture.

Client ID for the result file: claude-code

Run all three prompts in fresh sessions:

• direct.txt
• implicit.txt
• negative.txt

Deliverables

• Exact Claude Code version
• Exact model, operating system, permission mode, install path, and relevant configuration
• Sanitized raw output for all three prompts
• Generated artifact and receipt when one exists
• benchmarks/skill-portability/results/claude-code-<version>.json
• npm run benchmark:portability passes
• Skill, fixture, and prompts are unchanged

Start from result.example.json. The validator prints the canonical skill digest.

A fail or blocked run is a valid contribution when the limitation is explicit. Do not turn missing behavior into a success and do not claim cross-client equivalence.

Comment before starting so two people do not duplicate the same client/version. Do not include credentials, private repository data, personal paths, or user names in transcripts.