| name | paleo-auto |
| description | Use when you want automatic token-saving without manual skill selection. Auto-detects long sessions (>15 turns), high token usage, or bulky context — then auto-enables paleo + paleo-trim-context + paleo-converse + paleo-budget at safe defaults. User does not need to think about which skill is active. Off: "disable paleo-auto" / "manual paleo". |
| version | 2.5.0 |
| license | MIT |
| metadata | {"hermes":{"tags":["tokens","auto","compression","efficiency","automation","session"],"related_skills":["paleo","paleo-trim-context","paleo-converse","paleo-budget"]}} |
paleo-auto
Zero-touch token savings. Enable once, stay on — 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 — 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 — 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 — only lite/full/soft.
- If user disables a skill (e.g. "stop paleo"), respect it for the rest of the session.
- Report changes: "🦴 paleo-auto: enabled paleo + trim-context (session length 22 turns)" — one line, non-intrusive.
- Re-evaluate every 5 turns or on context-warning signal.
Switch
- On:
paleo-auto · auto paleo · enable auto-save
- Off:
disable paleo-auto · manual paleo · stop auto
- Status check:
paleo-auto status → list currently active skills + why
Gotchas
- paleo-auto is an ORCHESTRATOR, not a replacement. It delegates to the 6 sibling skills — 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) — 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).