| name | ship |
| description | Goal-oriented workflow for landing a change to agentyk safely. Use when the user asks to ship, fix and ship, take a change through validation, or drive a PR through CI to merge. |
Ship
Goal: land the requested change with evidence, and merge only after CI is green.
Read knowledge/operations/shipping.md § Required outcomes first —
it owns the bar. This skill owns how to reach it.
"Fix and ship" means implement first, then switch into shipping mode.
Working the change
Start from the goal and the risk the diff actually changes, not from checklist
order. Review the delta (git diff origin/main...HEAD, git log origin/main..HEAD),
confirm the requested behavior is really implemented, then pick the smallest
evidence a skeptical reviewer would accept: targeted diff reading, a focused
test, then the checks in AGENTS.md for the surfaces you
touched.
Two kinds of evidence, neither substituting for the other:
- The test. It drives the changed seam's real entry point — run a turn
through the executor, fold the event stream, exercise the driver's wire
mapping — not a constructor. Cover the meaningful success, failure, boundary,
regression, and integration paths. It stays offline: the scripted
SimDriver
and the canned stdio MCP server mean a missing API key is never the reason a
behavior went untested.
- The smoke test. Run the affected flow.
cargo run -p agentyk --example hello for framework changes; the example binary for anything under
examples/; a live-provider run for the http drivers, since their wire
format is the one thing SimDriver cannot prove.
For a fix, first reproduce the failure and retain before/after evidence in the
PR. Prefer a regression test plus the artifact that makes the failure easiest
to review: output, logs, screenshots, or VHS for terminal or timing-sensitive
behavior. If the regression test is the complete reproduction, say why another
artifact adds nothing.
