trellis-ai

Cut and publish a trellis-ai release to PyPI via the tag-triggered GitHub workflow. Use when asked to "cut a release", "publish to PyPI", "tag a version", or "ship vX.Y.Z". Do not use for refreshing the skynet deployment — that is the skynet-hub stack rebuild (see step 6), which this skill only reminds about.

When you discover a durable fact (an API contract, a config value, a service ownership, a non-obvious gotcha), store it in Trellis as a memory and connect it to the relevant entity in the knowledge graph. This makes the fact retrievable from any future task that touches the same area.

After completing meaningful work, write a structured trace to Trellis with steps + outcome, then record success/failure feedback. This is what makes future runs of retrieve-before-task useful — without it, the institutional memory is empty.

Pull a token-budgeted context pack from Trellis (prior traces, precedents, evidence, knowledge graph) before starting non-trivial work. Use whenever the task is more than a one-line tweak — investigations, refactors, new features, debugging, design decisions.

Structured institutional memory for AI agents — traces, precedents, knowledge graph, temporal versioning, and governed mutations. Complements built-in memory with structured knowledge.