No Memory
Explicit opt-out of persistent memory. Installs a short rule telling the agent that no memory backend exists in this project — so it stops trying (or pretending) to use one.
What it installs
| Artifact | Path (in your project) | Purpose |
|---|---|---|
| Rule | .claude/rules/no-memory.md | One-paragraph directive: no search_memories/add_memory tools available, rely on conversation context + repo files + git history |
Why opt out explicitly
Picking "no memory" in the wizard could just leave the slot empty — but then nothing documents the choice, and a future agent (or teammate) might install a memory bundle out of habit. Installing no-memory makes the decision explicit and visible in harness.json + .claude/rules/.
It also prevents a subtle failure mode: without the rule, an agent may hallucinate memory tools or plan workflows assuming a memory MCP is available. The rule short-circuits that.
When to choose this
- Single-session / throwaway projects where memory wouldn't pay off
- Strict context hygiene workflows where every fact should come from the repo, not an opaque vector store
- Compliance / data-residency reasons that forbid third-party memory services
If you later want memory, harness-kit add mem0 (or mempalace, local-memory) — they'll replace no-memory in harness.json.
Pairs well with
docs-as-code— the natural complement: if memory isn't an option, documentation is the memory. Writing decisions intodocs/becomes the discipline.context-discipline— both push the same philosophy: keep the agent working from fresh, observable sources rather than cached state.