fix(security): remove inert curl-pipe deny rules from settings template

[CodeWithJuber]

What & why

A usage audit flagged the two curl-pipe deny rules in
global/settings.template.json as malformed. Verification confirmed it: Claude
Code only honors :* as a trailing wildcard at the end of a Bash(...)
specifier — any text after it turns the colon into a literal, so these rules
matched no command that ever runs. They looked like a security control and
enforced nothing, and forge init copied them verbatim into every user's
~/.claude/settings.json.

The intended protection already exists and genuinely enforces at
global/guards/protect-paths.sh — a PreToolUse Bash hook that emits a
permissionDecision:"deny", which is the mechanism the Claude Code docs
recommend for constraining piped commands.

Changes

• Remove the two inert deny rules from the settings template.
• Tighten protect-paths.sh pipe-to-shell detection: was a space-only
  substring match; now a boundary-aware regex that also catches a no-space pipe
  and a zsh target, while still allowing legit pipes such as shellcheck.
• Add test/settings_template.test.js — a regression guard: no Bash rule
  may embed a shell operator in its specifier, :* must be pattern-terminal,
  and the two removed rules must stay gone.

Note for reviewers

No enforcement is lost — the removed rules were dead; the real guard is
unchanged in behavior except for being broadened. Verified the guard blocks
| sh, no-space pipe, and zsh targets, and still allows shellcheck and
plain curl.

Checklist

• npm test passes — 663/666; the one failure (imagine.test.js worktree
  sandbox) is pre-existing and unrelated (proven by stashing this diff)
• npm run check passes for the changed files (pre-existing biome.json
  deprecation + warnings in other files are untouched)
• New test added for the template rule shape
• Conventional commit message (fix:)
• CHANGELOG.md updated under ## [Unreleased]
• No new runtime dependency
• node src/cli.js docs check passes

Risk & rollback

• Risk level: low
• Rollback plan: revert this commit — restores the two (inert) rules and the
  previous guard match; no behavioral loss either way.

Extra checks

• npm run typecheck passes
• Input validated at boundaries; errors handled
• Logs contain no secrets/PII

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.