betamatter 0.1.0

Augmented reality for LLM agents: YAML frontmatter, content fencing, and fire-once tips for MCP tool responses.

betamatter

Structured beta for the agent driving the tools.

betamatter is the YAML-frontmatter layer for MCP tool responses whose user is the LLM. Every tool response has two channels — the body (the answer) and the frontmatter (a trusted sidechannel about the answer) — and this library owns the mechanics that keep that line honest: an entries container (FMEntries) that guards multi-contributor guidance keys against silent clobbering, the sole sanctioned --- block builder (build_frontmatter), a box-drawing datamark fence for untrusted content (fence_content), and a fire-once tip ledger (TipLedger) for once-per-process educational guidance. Extracted from parkour-mcp, whose frontmatter discipline it generalizes; consuming projects declare their own trust polarity, field vocabulary, and tip registry as deltas.

The conventions themselves — handles/shape/steering, phrasing doctrine, trust rules, tip semantics — are specified in docs/standard.md.