feat(daymade-financial): add devils-advocate skill (v1.1.0)

[daymade]

What

New skill devils-advocate in the daymade-financial suite: structured devil's-advocate pressure-testing of an investment thesis against user-supplied evidence materials.

• Reimplements LinqAlpha's hedge-fund Devil's Advocate flow from the production prompt template + JSON schema published on the AWS ML blog (2026-02), with three extensions that implementation lacks: materials-bounded base-rate outside view (Mauboussin), RAND-ABP monitoring signposts, and an explicit materials-bias/coverage declaration.
• Assumption decomposition (explicit/implicit, A1/A2 ids, fact/forecast/mechanism typing, load-bearing test), per-assumption evidence-linked counterarguments under a source-credibility ladder with verbatim citations, ACH absent-evidence question, dual-layer output (audit JSON object + theme-grouped analyst narrative), risk-flag rubric with anti-inflation guard.
• Registry: suite skills array + version 1.0.0→1.1.0, marketplace metadata 1.86.0→1.87.0, CHANGELOG entry, both READMEs, CLAUDE.md snapshot list (fix: make Cloudflare SSL changes opt-in #91).

Verification

• quick_validate + security_scan green; semantic sanitization pass (public-repo hygiene axis) returned zero findings — all named entities are public companies/papers.
• One fresh-context independent review: P0×1 / P1×5 / P2×14, all fixed and mechanically re-verified (step-reference grep back to zero).
• Production test by a context-free agent on a real thesis with 5 research reports: 8 assumptions, 38 mechanically verified verbatim citations, schema-conformant JSON; the nine ambiguities that run surfaced were folded back into the skill.

🤖 Generated with Claude Code

https://claude.ai/code/session_016xvCgUU2ZMRJGVjg65HX6A