Install in seconds
Install this skill
Copy the command and run it in your terminal. You can review the source before installing.
terminal
git clone https://github.com/mocasus/paleo

Works with Git. The repository opens in your current directory.

๐Ÿฆด
AutomationPython

Paleo Auto

by mocasus

Automatically detects long sessions, high token usage, or bulky context and enables the appropriate Paleo token-saving skills to optimize token consumption without manual intervention.

11 stars0 forksAdded 2026/07/16
agent-skillsai-agentsai-efficiencyautomationclaudeclaude-codecodexcontext-windowcost-optimizationdeveloper-toolsgeminigemini-clihermeshermes-agentllmllm-agentsprompt-engineeringskillstoken-savingtokens

Documentation

README

paleo-auto

Zero-touch token savings. Enable once, stay on โ€” paleo-auto watches your session and activates the right skills at the right time.

What it does

  • Automatic activation. No manual triggers needed. Detects session conditions and enables skills.
  • Picks the right combo. Not all skills needed all the time โ€” paleo-auto picks per-condition.
  • Safe defaults. Never applies ultra or hard-strict modes unless asked. Always starts at full or soft.

Auto-detection triggers

Condition Detection Action
Session length >15 turns Count user/assistant exchanges Enable paleo (full) + paleo-trim-context
Repeated tool output (same tool called 3+ times with similar results) Compare last 3 tool outputs Enable paleo-summary (full)
Context approaching warning (>60% capacity) Estimate context usage Enable paleo-converse (full, N=8) + paleo-trim-context
High token cost model detected (GPT-4, Claude Opus, etc.) Model name check Enable paleo-budget (soft, 3000 tokens) + paleo (full)
Bulky output returned (single tool result >2000 tokens est.) Output size check Enable paleo-summary (lite) for subsequent similar calls
User sends "long session" / "chatty" / "lagging" Keyword match Enable all 6 skills at full defaults
Session startup (fresh new session) First 2 turns Enable paleo (full) only โ€” low overhead baseline

Default combo per scenario

Scenario Active skills Why
Fresh session paleo (full) Low overhead baseline, catches wordy responses
Long session paleo (full) + paleo-trim-context + paleo-converse (N=6) Context hygiene + history compression
Expensive model paleo (full) + paleo-budget (soft, 3k) Cap costs on pricey models
Debugging/tools-heavy paleo-summary (full) + paleo (full) Shrink bulky tool output
User says "lagging" / "chatty" All 6 at full defaults Max savings, immediate

Rules

  • Never override a user's explicit skill activation. If user says paleo ultra, keep it.
  • Never apply ultra or hard/strict modes automatically โ€” only lite/full/soft.
  • If user disables a skill (e.g. "stop paleo"), respect it for the rest of the session.
  • Report changes: "๐Ÿฆด paleo-auto: enabled paleo + trim-context (session length 22 turns)" โ€” one line, non-intrusive.
  • Re-evaluate every 5 turns or on context-warning signal.

Switch

  • On: paleo-auto ยท auto paleo ยท enable auto-save
  • Off: disable paleo-auto ยท manual paleo ยท stop auto
  • Status check: paleo-auto status โ†’ list currently active skills + why

Gotchas

  • paleo-auto is an ORCHESTRATOR, not a replacement. It delegates to the 6 sibling skills โ€” don't re-implement their rules here.
  • Don't fight user overrides. Use explicitly set = respect it.
  • Detection is approximate (turn count, output size estimation) โ€” false positive is fine, false negative is tolerable. Err on the side of activating.
  • Startup cost: loading paleo-auto adds ~300 tokens of context. Only keep active if the session will benefit from it (usually >10 turns).