[FEATURE] Unify injection heuristic precedence across skill families

[hello-args]

Summary

Skills in the same functional family (e.g. RFP classify vs synthesize) receive inconsistent severities depending on which analyzer fires first: rfp-classify gets prompt_injection HIGH; rfp-synthesize gets sigma_metadata HIGH only (MCTS-T-1008 tool shadowing — B7). Triage becomes unpredictable; scores vary for equivalent instruction patterns.

---

Problem

Multiple analyzers scan the same instruction surfaces:

Analyzer	Triggers on	Example pattern
prompt_injection	Bearer, extract, override language	"extract all fields"
sigma_metadata	"call", tool shadowing	"Called once per request" (B7)
skill_md	W008 credentials, fetch links	"password resets" (B8)

No precedence rules — all analyzers run independently; all findings score (D3).

API service repositories RFP skill family:

.cursor/skills/rfp-classify/SKILL.md → prompt_injection HIGH
.cursor/skills/rfp-synthesize/SKILL.md → sigma_metadata HIGH (different rule)

Same author, same domain, different analyzer outcomes.

---

Proposed solution

Document rule precedence; unify skill family scoring.

Acceptance Criteria

• rfp-classify and rfp-synthesize get consistent severity for equivalent patterns
• Precedence documented in analyzer catalog
• Suppressed findings logged in report metadata (debug mode)
• B7 sigma fix coordinated — "call" no longer false HIGH
• Unit test: overlapping triggers → single finding

---