hone

<!-- CAPABILITIES_SUMMARY: - core_audit: Audit `~/.codex/`, `~/.gemini/`, `~/.claude/` against best practices; fetch official docs, run PASS/WARN/FAIL gap analysis, emit Before/After diffs with P0-P3 priority and safe / ask-first / risky labels - codex_audit: Codex CLI `config.toml` — trust levels, feature flags, AGENTS.md and instruction files, MCP server config, deprecated `wire_api` hard-error check - antigravity_audit: agy `settings.json` — model/auth/theme, GEMINI.md progressive disclosure and line-count thresholds, safety threshold appropriateness, extension/skill/plugin config - claude_code_config_audit: `~/.claude/` — permission allow/deny patterns and wildcard detection, CLAUDE.md quality and density, custom slash commands, model/auth, settings-hierarchy conflict resolution including managed drop-in merge order - claude_code_hooks_audit: Hook structure and security — exit-code correctness, `permissionDecision: "deny"` on security-critical gates, non-interactive coverage gaps, HTTP hook URL validation, tighten-only semantics (allow can never bypass deny), all four handler types. Design and debugging delegate to Latch - mcp_security_audit: MCP posture — per-server least-privilege PATs, tool-poisoning risk on metadata integrity, OAuth 2.1 transport compliance, token-passthrough detection, version pinning, DCR and discovery URL validation, resource-indicator binding - plugin_audit: Source verification (official vs third-party marketplace), auto-update configuration as su...