CLI Blind Proof — 2026-07-23

ACC CLI Blind Proof (public)

Overall: PASS

Strict-blind Codex CLI pack (plain English only, no coached $skill names).

Model	gpt-5.4-mini
Reasoning	low
Plugin	2.0.0-beta.5
Product story	Local habit tracker

Proven

• Core path (app built under blind prompts)
• SessionStart / Stop hooks
• Memory write + later recall
• Safety hard-block of curl | bash
• Doctor (0 FAIL)

Honest gaps

• Many skills remain NOT PROVEN under pure blind (expected)
• Compact / subagent hooks not exercised

Privacy

Redacted public report only — no home paths, no session dumps, no secrets.

Links

• PROOF.public.md
• Re-run pack
• Website section: Verified on CLI (after site deploy)

v2.0.0-beta.5 — Desktop + CLI same page

Anyone Can Code v2.0.0-beta.5

One plugin. One version. Desktop + CLI.

What’s new

• CLI auto memory when ACC hooks are trusted (same path as Desktop)
• Docs honesty: no more “CLI still 1.1.0 / no auto memory”
• Ships with hooks optimize from #37 (lazy load, tier inject, timeouts)
• Host-honest: Scheduled / Sites stay Desktop-first

Upgrade (required)

1. Marketplace upgrade Anyone Can Code
2. Re-trust every ACC hook (/hooks on CLI, Hooks UI on Desktop)
3. Restart Codex
4. Open a project → $setup if new

Hook hashes change on updates — untrusted hooks = no memory.

Install

Marketplace: https://github.com/mitunmanav/anyone-can-code

CLI:

codex plugin marketplace add mitunmanav/anyone-can-code --ref main
codex plugin add anyone-can-code@anyone-can-code-marketplace

Then /hooks → trust all ACC hooks.

Pin

• Manifest: 2.0.0-beta.5
• Merge: PR #38

Anyone Can Code v1.1.0-beta.4

Open beta. Free Codex Desktop + CLI reliability workflow plugin. Plan, build, resume, recover, verify from plain English.