Skills

Three agent skills ship with the lexicon and load automatically in a project that uses it.

Skill	Covers
chant-fountain	The core authoring loop — declaring the three kinds, the build/apply/drift/import cycle, endpoint and auth
chant-fountain-secrets	Secrets, env_vars, and ${VAR} substitution: the order of preference, vault precedence, and what does not round-trip
chant-fountain-locked-sandboxes	The locked-down posture for untrusted or security-sensitive agents, and running conversations against them

Invoke one directly by name, or let it trigger on context (fountain, vault, sandbox, networking_type).