Add warrant-policy-author (Security & Systems)Update README.md

[MANISH PATEL (rmanish2000-del)]

Adds warrant-policy-author, the authoring skill shipped inside the
warrant-mcp npm package
(a policy firewall for Claude Code tool calls — plain-English rules,
compiled once, enforced by deterministic code via a PreToolUse hook).

The skill teaches Claude the authoring craft: a short user interview,
sentences shaped for the closed 8-rule vocabulary, and the failure shapes
(provenance, cost/judgement, ask-me-first, cross-action state) explained
rather than avoided. Only warrant-mcp review can say whether a policy
compiles; the skill's job is to give it the best possible chance.
Evidence so far: eight policies for genuinely different project types —
three authored directly, five by sessions given nothing but the installed
skill and a vaguely-worded brief — all accepted by review on the first
attempt. A drift test in the package pins the skill's vocabulary to the
compiler's schema so the two cannot diverge silently.

• Real use case: a policy the compiler refuses is the product's main
  onboarding friction; this skill teaches the craft that avoids it and
  explains the refusals it cannot avoid.
• Not a duplicate: no existing entry targets agent-firewall policy authoring.
• License: MIT. Install: npx warrant-mcp init --skill; or
  /plugin marketplace add rmanish2000-del/warrant-mcp then
  /plugin install warrant-policy-author@warrant-mcp (no npm needed); or
  copy the folder into .claude/skills/.

Note for maintainers: please link the entry to the source folder rather than
vendoring a copy — the skill is derived from the enforcement engine's schema
and must version with it.