Change an assumption, reprice, and compare — without re-running Discover. This is a sidebar: off-backbone, entered only by its _trigger, and it returns control rather than advancing. It never appears as current_phase.

_gates: generate means Generate must not start while this sidebar is unresolved. Resolution includes a decline — a declined sidebar is still "completed", because _gates holds ordering, never participation. Whether the user actually engaged is a separate signal, carried by whether scenarios/index.json exists.

The cross-skill invariants for the workshop live in references/vendored/workshop/workshop-invariants.md and win on any disagreement with the unit files here.

Knobs on the sheet: region, HA, compute target, cost optimization, CPU architecture. The architecture default is x86_64 here, matching the skill default rather than the repo-wide Graviton default — see SKILL.md § Philosophy.

Status — build step 6

Implemented. The knob set (region, HA posture, compute target, cost optimization, CPU architecture), the Apply & reprice mechanics, the comparison table, and the sidebar-resolve handoff back to the decision gate are ported from gcp-to-aws‘s equivalent, adapted to this skill’s knobs and its Premium/Balanced/Optimized scenario schema — see each fragment’s own Status note for what changed versus GCP’s version.

Step: Run the sidebar

Present the sheet (workshop-sheet.md).
On Apply & reprice, run workshop-refresh.md: patch preferences → re-run Design → re-run Estimate → snapshot the result as a scenario.
On Compare, or after a successful refresh, run workshop-compare.md.
Run workshop-assemble.md to resolve the sidebar and return control.

references/phases/workshop/workshop.md

Raw file

aws/agent-toolkit-for-aws