Skip to content

arxdsilva/loops

v0.7.0MIT

Agent-driven Go backend workflow: research → plan (+review loop) → role-based parallel implementation → QA verification loop → product proof + PR → reflection into ai-memory. Every step is traced live (status line, pane) and in an HTML report.

loops for Codex

Generated by loops port build from the canonical plugin (repository root). Do not edit; see specs/PLATFORMS.md §3.2.

Install

  1. Add the marketplace (the repository root holds .agents/plugins/marketplace.json, which points at platforms/codex):

    codex plugin marketplace add arxdsilva/ai-loops
    
  2. Install the loops plugin from the Plugins Directory (restart Codex after adding the marketplace).

  3. Install the role agents (plugins cannot bundle subagents):

    loops install --target codex            # project: .codex/agents/
    loops install --target codex --scope user   # user: ~/.codex/agents/
    
  4. Review and trust the plugin hooks with /hooks. Codex skips untrusted hooks, and re-asks when they change.

Commands are skills: $work, $init, $status, $report.

Binary

bin/loops uses loops from PATH when it reports version 0.7.0, else a cached binary, else downloads the prebuilt release binary for your OS/arch (checksum-verified, no Go needed). Only when no prebuilt binary exists does it fall back to go install github.com/arxdsilva/ai-loops/cmd/loops@v0.7.0.

Caveats

  • Hooks run only after you trust them, and plugins with hooks are not eligible for the public plugin directory.
  • No PostToolUseFailure event: failed tool calls are not traced.
  • Custom agents cannot restrict tools; read-only roles are told not to edit.
  • Subagents cannot be resumed: loop-backs spawn a new agent of the same role (PLATFORMS §4, tier B).
  • Models: roles inherit the session model unless loops port build --models <file.json> maps the tiers.