| name | spx-auto |
| description | State-aware workflow continuation for advanced coding models. Use when the user wants to resume or execute the safest next SPX step without naming the exact workflow. |
SPX Auto
Read references/project-learning.md and apply its consume-only policy.
Read references/project-cognition.md, using cognition intent ask, and
references/routing-contract.md.
Read references/ui-quality-gate.md when the request or active state is
UI-bearing.
Inspect only enough state to identify the highest trustworthy stage: current
diff, active feature/lane/quick/debug state, recoverable discussion handoff,
task completion evidence, and cognition readiness. Treat status labels and
checked boxes as claims until their required artifacts and live evidence agree.
Select exactly one next SPX workflow, read that Skill, and continue immediately
when its boundary is clear. Prefer resuming existing valid state over creating a
new lane. Do not use auto to bypass a user-owned decision, external-write
authorization, destructive action, or a blocked validation gate.
Stop after that workflow returns; do not invoke a second workflow in the same
invocation. Report any downstream route as a handoff only.
Pass auto_default_recommendation: true only to that selected workflow. It may
accept one explicit safe default that preserves current intent and scope. Never
use it for scope reduction, destructive or irreversible action, security or
credential decisions, external cost, or unresolved planning ambiguity.
Keep this hint turn-local; never persist it as the canonical next route.
Create no separate auto state, report, or orchestration layer. If evidence is
ambiguous, state the competing routes and ask only for the decision the
repository cannot resolve. Report the selected route and the concrete state
signal that justified it.
