Module DD-08 — Hermes: Layered Persistent Memory

Hermes: Layered Persistent Memory

Self-evolving skills. The depth play. The memory reference. 33,000+ stars. Overtook OpenClaw on OpenRouter May 10, 2026 (224B tokens/day). On the breadth-vs-depth split, unambiguously the depth competitor.

minutes

artifacts

sub-sections

Hermes competes on memory depth: self-evolving skills that compound across sessions. The depth play's strategic thesis is that breadth is reproducible (a competitor can add channels) but a compounding memory is not — each user's accumulated skill store is a switching cost. The migrate command is the market expression: depth makes reach obsolete. But the same model-initiated writes that make skills compound make poisoning compound — the self-evolving skill store is the largest memory-poisoning surface in the roster, and the write gate is the missing defense.

Key Claims

Load-Bearing Claims

Hermes is the Module 4 (Memory) reference at 5/5 — the only harness in the roster operating at tier 4 with write-back: the episodic record is not just logged (read-only) but curated into skills and re-injected. The store grows monotonically richer with use. This is the deepest memory implementation in the roster and the load-bearing reason for reference status.

The self-evolving skill store is the feature AND the vulnerability via the same mechanism (model-initiated writes, no gate). As a feature: the agent learns from every session without throttle, skills compound, switching cost accrues. As a vulnerability: a prompt-injected model persists a poisoned skill that compounds across all future sessions. The 5/5 on Module 4 and the 2/5 on Module 6 are the same design decision read from two sides.

The compounding that makes skills valuable is the same compounding that makes poisoning dangerous. A poisoned working file is read once; a poisoned skill activates on every similar future task with no natural decay. The memory-poisoning risk scales with memory depth, and Hermes — the deepest memory — carries the largest poisoning surface in the roster.

The NemoClaw-style harness-managed write gate (model proposes, harness validates) is the single highest-value fix. It converts the compounding-poisoning surface into a compounding-capability surface with the risk closed. The depth-versus-governance contrast between Hermes (DD-08, the depth reference) and NemoClaw (DD-09, the governance reference) is load-bearing for the roster — the two define the depth-vs-governance axis every other harness sits between.

After This Module

Articulate Hermes's strategic thesis — the depth play — and explain why a compounding memory is a switching cost that breadth cannot reproduce.

Describe the self-evolving skill model: how a skill is born, retrieved, and compounds across sessions, and why it is the deepest memory implementation in the roster (Module 4 tier 4 with write-back).

Score Hermes across the 12-module rubric and defend the 36/60 — highest on Module 4 (Memory, 5/5), lowest on Modules 5 and 6 (Sandbox and Permission, 2/5 each).

Explain why the self-evolving skill store is a memory-poisoning surface, why poisoning compounds here specifically, and why model-initiated writes are both the feature and the vulnerability.

Specify the NemoClaw-style harness-managed write gate that closes the poisoning surface — the defense Hermes omits — and articulate the architect's verdict on when to build on Hermes and when not to.

Artifacts

Teaching Document

Teaching document — the depth play thesis, the self-evolving skill model (Module 4 tier 4 with write-back), the memory-poisoning surface, model-initiated writes as feature-and-vulnerability, the NemoClaw write-gate fix, the 36/60 score profile; with learning objectives, anti-patterns, key terms, references

READ

Diagrams

5 Mermaid/n8n diagrams — the self-evolving skill model (compounding flow), the poisoning surface (working file vs skill), the NemoClaw write-gate fix (vulnerable vs hardened), the 36/60 score profile (depth-specialist shape), the n8n skill-loop workflow with the model-initiated write risk callout

READ

Slide Deck

10 slides — reveal.js, dark theme, design-system teal; covers the depth thesis, the self-evolving skill model, why Hermes is the Module 4 reference, the poisoning surface, model-initiated writes as feature-and-vulnerability, the NemoClaw fix, the score profile, anti-patterns, the lab

READ

Teaching Script

Verbatim teaching transcript with [SLIDE N] cues, ~3,000 words spoken at ~140 wpm across 10 slide cues