feat(credit): add rating-migration-watch workflow

[liyc-sys]

Summary

Adds a rating-migration-watch workflow to llmquant-credit. The existing issuer-credit-risk-review answers "how risky is this issuer today"; this workflow answers the forward-looking question "is this issuer likely to migrate across rating levels" — upgrade/downgrade candidates, outlook and watch status, market-implied rating gaps versus same-rating peer cohorts, and crossover (fallen angel / rising star) candidates. Rating migration is a distinct, repeatable fixed-income research task that none of the current credit workflows cover.

Type of change

• New category skill (skills/llmquant-<category>/)
• New workflow (workflows/<name>.md)
• Fix to an existing skill / workflow
• Script or asset change
• Docs / README only

Affected categories

llmquant-credit

Contract checklist

• Category folder is named llmquant-*.
• Root SKILL.md exists and routes to the workflow.
• Workflow file added or updated under workflows/.
• input_data_source: LLMQuant Data is present in the router frontmatter.
• Required data capabilities are described in natural language (not exact MCP tool names).
• Missing or future data capabilities are explicitly named.
• Freshness rules are explicit (dates, filing periods, observation dates, stale notices).
• Fallback behavior is explicit for missing, unsupported, or stale data.
• Guardrails prevent invented data and unsupported conclusions.
• Output format is structured (Answer, Evidence, Risks/Caveats, Data Used).
• README tables updated if a category or major workflow was added, removed, or renamed.

Evidence grounding

All external facts (rating history, outlooks, filings, fundamentals trajectory, issuer and peer-cohort spreads, CDS, sector/macro context) route through LLMQuant Data. The workflow requires reporting rating action dates, outlook dates, filing periods, and spread observation dates in the output, and falls back to fundamentals-trajectory-only analysis — with the missing rating-market inputs named — when agency outlooks or peer-cohort spreads are unavailable. Guardrails forbid presenting the view as an agency action, claiming market-implied gaps without cohort data, or extrapolating from a single reporting period.

Linked issues

None.