DONNA

Universal output surface · Reports, deltas, plans, artifacts — all here
Loop engineering: 5 gaps/Evaluator cartridges: 6 gaps/Claude Code /goal native — Hermes manual only/Humanify cartridge: P1/Autonomous retry loop: P1/Composite scoring: P2/All output → delta111.pages.dev/ Loop engineering: 5 gaps/Evaluator cartridges: 6 gaps/Claude Code /goal native — Hermes manual only/Humanify cartridge: P1/Autonomous retry loop: P1/Composite scoring: P2/All output → delta111.pages.dev/
03
Entries
Published
15
Implementation
Gaps
11
Planned
Actions
Latest Entry — Jul 18, 2026

Evaluator Cartridges & Autonomous Loops

Karpathy's auto-research is now a native Claude Code feature — /goal and /loop. The real unlock: evaluator cartridges — custom scoring algorithms that grade AI output autonomously so the loop knows when to stop. No human in the loop. We have zero of this.

Open entry →
◆ Key Insight Jul 18
The human moves from checking every output to defining the scoring rubric once.

A cartridge is a self-contained scoring algorithm: text in → score out (1–10). Composite across multiple cartridges for holistic grading. The newsletter loop ran 5 attempts autonomously — 7.1 → 26.8 over 40 minutes. Zero keystrokes from the human.

Dream Labs AI · 6 implementation gaps
// Active knowledge base

All Entries

◆ Entry #003 Awaiting
Next report, delta, or artifact…

// Every Donna output — deltas, reports, plans, artifacts
// — lands here as a new entry by default.
// No exceptions unless Dwad says otherwise.

◆ Entry #002 · Delta Jul 18, 2026
Evaluator Cartridges & Autonomous Loops

Karpathy's auto-research pattern is now a native Claude Code feature/goal and /loop. The new unlock: evaluator cartridges — custom scoring algorithms that grade AI output autonomously. The combination — goal-driven agent + multi-cartridge evaluator + attempt budget — produces output that iteratively sharpens without you.

This is Entry #001's loop engineering with the inspector automated. The human graduates from checking every output to defining the scoring rubric once.

A cartridge is a self-contained scoring algorithm: output → number (1–10). Three shown:

1. Humanify Cartridge — voice, variety, predictability (AI slop detector). Only 9+ passes.
2. Hormozi AI Cartridge — scored against Alex Hormozi's public writing for marketing punch.
3. Open Rate Analytics Engine — predicted open rate from past email data + audience behavior.

Composite scoring across cartridges: newsletter required 27/30 total, no single cartridge below 8. Agent ran 5 attempts autonomously: 7.1 → 26.8 over ~40 minutes.

CapabilityHermes HasSource SaysGap
Autonomous goal-driven loops⚠️ Manual pattern/goal nativeBIG
Evaluator cartridges (scoring)❌ NoneMulti-cartridgeBIG
Condition-based wake-up⚠️ Cron (schedule)/loop (condition)Medium
Subagent swarm → single goal⚠️ Independent tasks650 experimentsMedium
Attempt tracking + scoring history⚠️ STOP AFTER onlyScoring tableMedium
Scoped file access for loops❌ Full FSScoped accessSmall
  1. P1 — Humanify cartridge. Standalone scorer script: text → 1-10 on voice, variety, predictability. Beachhead to prove the pattern.
  2. P1 — Autonomous retry in loop-engineer. Upgrade from manual inspection to subagent self-scoring against checklist, retry until pass or budget.
  3. P2 — Multi-cartridge composite. Run N cartridges, sum scores, enforce per-cartridge minimums. The "27/30 with no cartridge below 8" pattern.
  4. P2 — Scoring table in cron output. Per-attempt scoring history in deliveries so Dwad sees improvement trajectory.
  5. P3 — Condition-based cron wake-up. Extend cronjob: "run every N minutes UNTIL condition met."
  6. P4 — Swarm coordinator. N subagents optimize toward shared best-score state. The 650-experiment pattern.
◆ Entry #001 · Delta Jul 16, 2026
Loop Engineering — The End of One-Shot Prompts

Prompt engineering is dead. The new paradigm is loop engineering: AI tries → self-checks → fixes → retries → marks done. You move from being in the loop to being outside it. Four components: Goal, Checklist, Inspector, Budget.

Inner loop: Single task — try, inspect, fix, retry. AI self-corrects before delivering.
Outer loop: Recurring scheduled job running inner loops each iteration. Self-improves over time.

CapabilityHermes HasLoop Eng.Gap
Scheduled recurring tasks✅ 17 cronsOuter loopNone
Self-review within task❌ MissingInner loopBIG
Inspector/second pass❌ MissingGrader modelBIG
Hard definition-of-done⚠️ Soft goalsChecklistMedium
Skills auto-update from loops❌ MissingLoop→skillMedium
Per-task attempt budget⚠️ Repeat onlyStop after NSmall
  1. P0 — Inner loop skill. Created loop-engineer skill: GOAL + DONE WHEN + STOP AFTER → delegate → inspect → retry up to N.
  2. P0 — Inspector pass. Five-point checklist before subagent output reaches Dwad: exists, meets checklist, no hallucinations, correct format, sniff test.
  3. P1 — Upgrade crons. DONE WHEN checklists + STOP AFTER budgets on quality-critical cron jobs.
  4. P2 — Skills auto-update. LLM-driven cron prompt: "If you discover anything that improves the skill, patch it before delivering."
  5. P3 — Goal template library. Reusable DONE WHEN checklists for recurring task types.