ponytail

AI Council — Autonomous Orchestration Skills

A self-looping, never-stopping multi-agent orchestration system inspired by Karpathy's autoresearch.

The council cycles through 5 stages (Think → Plan → Create → Review → Verify) in an autonomous loop. All 14 superpower patterns are embedded inline — zero external Skill dependencies. The loop runs until the objective is fully resolved.

Architecture

┌─────────────────────────────────────────────���───────────────────┐
│                   MAIN ORCHESTRATION LOOP                        │
│                                                                  │
│  while not objective_fully_satisfied:                            │
│    ┌──────┐   ┌──────┐   ┌──────┐   ┌──────┐   ┌──────┐        │
│    │THINK │→  │PLAN  │→  │CREATE│→  │REVIEW│→  │VERIFY│        │
│    │  +   │   │  +   │   │  +   │   │  &   │   │  &   │        │
│    │CRITIC│   │CRITIC│   │CRITIC│   │ TEST │   │DELIVER│        │
│    └──┬───┘   └──┬───┘   └──┬───┘   └──┬───┘   └──┬───┘        │
│       │           │          │          │          │             │
│       ◄───────────┴──────────┴──────────┴──────────┤             │
│       │  loop back if critiqued or issues found     │             │
│       └─────────────────────────────────────────────┘             │
│                    │  if verify passes but objective unmet         │
│                    └──────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────┘

Skills

council-orchestration ⭐ Single-Model

All agents run on your current active model. No model switching needed. Works in any CLI.

Stage	Embedded Patterns Used
1 — Think	Brainstorming Pattern + Critic (embedded inline)
2 — Plan	Writing Plans Pattern (embedded inline)
3 — Create	TDD + Subagent-Driven Dev + Parallel Dispatch (all embedded inline)
4 — Review & Test	Code Review + Systematic Debugging (embedded inline)
5 — Verify & Deliver	Verification + Finishing Branch (embedded inline)

ai-council-orchestration Multi-Model

Same loop, but switches to the best model per stage (Claude Opus for thinking, Sonnet for planning, etc.).

loop 🔁 Cross-Platform Recurrent Task Runner

Run any command on a recurring interval — fixed or self-paced. Works across Claude Code, Gemini CLI, Copilot CLI, and plain shell.

Adapts to the platform: Claude Code uses ScheduleWakeup, Gemini CLI uses activate_skill chaining, Copilot CLI uses subagent dispatch, and the shell fallback uses watch/cron.

Mechanism	Claude Code	Gemini CLI	Copilot CLI	Any Shell
Fixed interval	/loop 5m /cmd	loop run "cmd" -i 30	loop run "cmd" -i 30	./loop.sh run "cmd" -i 30
Self-paced	/loop /cmd	activate_skill("loop")	subagent loop	./loop.sh interactive "cmd"
Status	loop status	loop status	loop status	./loop.sh status

ponytail 🦎 Over-Engineering Prevention (KISS/YAGNI)

Forces the simplest, shortest solution that actually works: standard library first, native platform features before dependencies, and one line before fifty. Fully embedded in the Thinking, Planning, Creating, and Reviewing stages of the AI Council.

Command	What it does
/ponytail [lite | full | ultra | off]	Set the intensity, or turn it off. No argument reports the current level.
/ponytail-review	Review the current diff for over-engineering, hands back a delete-list.
/ponytail-audit	Audit the whole repo for over-engineering, not just the diff.
/ponytail-debt	Harvest the ponytail: shortcuts you've deferred into a ledger, so "later" doesn't become "never".
/ponytail-gain	Show the measured impact scoreboard (less code, less cost, more speed) from the benchmark.
/ponytail-help	Quick reference for the commands above.

Files