memory

Memory & domains

Stella's memory is a local self-improvement loop, and most of it is user-inspectable plain text under .stella/. Reference: stella memory.

The loop

1. After turns, Stella reflects and records lessons — appended to .stella/reflections.jsonl (one JSON object per line: lesson, domains, occurred_at) and indexed in the local context DB.
2. Lessons that keep recurring are auto-promoted to skills (.stella/skills/<slug>/SKILL.md) — see ../skills/.
3. You can promote a lesson to a hard rule yourself: stella memory promote writes .stella/rules/<slug>.md.
4. Hand-written notes live in .stella/memories/*.md; the agent cites them with cite_memory when they steer a decision.

stella memory list        # what Stella has learned here
stella memory promote     # lesson → enforced rule
stella memory validate    # prune stale/contradicted lessons

Domains — .stella/domains.toml names the areas of your codebase so lessons and telemetry tag correctly. Stella infers one on stella init (stamped inferred_by = "heuristic"); domains.toml here shows a hand-tuned taxonomy for a typical web app. Everything stays on your machine — memory, reflections, and telemetry never leave .stella/.