Context Packs
A context pack is a compact, ranked view of project memory for the next session. Structured sections are preserved; the ranker blends seven composite signals.
Tool exports
Export a reviewed pack as the files AI coding tools already read — CLAUDE.md, AGENTS.md, a Cursor rule, or a generic Markdown handoff. Copy it or download it. Shelvia prepares the file from approved memory; it does not run the tool for you or write into your repo.
Seven composite signals
- Structured priority (kind weight)
- Source-backed bonus
- Memory health (stale + conflicted downweighted)
- Semantic similarity (against the pack purpose)
- Recency
- Purpose / target_tool match
- Kind weight (decisions > prompts > next steps, by default)
Why a pack is not a RAG dump
Context packs preserve the structured shape of project memory, decisions live in a decisions section, sources live in a sources section, prompts live in a prompts section. The receiving agent can reason about kind, not just text.
Generating a pack
await shelvia.context.generatePack(projectId, {
purpose: "Continue the v2 launch",
target_tool: "claude",
max_chars: 12000,
});For runnable code samples and the developer reference, see /developers. For the trust model in depth, see /security.