Authority
Decide architecture. If no target is supplied, run Opportunity Discovery before selecting target, no-design-change, or blocker. Seal ignored .first-four-hours/architectures/<slug>/architecture.md for blueprint; do not edit implementation files.
Stop at architecture. Never classify implementation tier, perform a full propagation sweep, prescribe file-level edits, write test or execution blueprints, order migrations or rollout, or attack an implementation. Those responsibilities belong to blueprint. Paths and optional views are evidence bearings, not edit instructions.
Workflow
- Follow Architecture Protocol. Consult
reasonbefore committing to an approach, after a failure or direction change, and before sealing. Load Structural Techniques only where the protocol or discovery points to it. - Use Bounded Delegation Protocol only for optional evidence workers or earned specialist review.
- Draft from the Handoff Template.
- Use Worked Example only for optional alternatives, user policy, or view calibration, and Calibration Examples before framing a target inside one module or class or selecting any structure other than
no-design-change. - Seal once:
python /absolute/skill-root/scripts/cli.py --repo-root /absolute/repo \
--input draft=/absolute/architecture-draft.md \
--input slug=<architecture-slug> --format json run
Persist settled structural truth: pressure or no-change outcome, owner/boundary, constraints, selected architecture, consequences, evidence, trust-relevant unknowns, and any user-owned policy.
Completion
Complete only when the command returns status: complete with result.path at .first-four-hours/architectures/<slug>/architecture.md; report result.git_ignored. If sealing fails, repair the named draft or workspace conflict and rerun. If evidence changed, refresh and seal against the new snapshot. Final and stop reports trace reason checkpoints per its Checkpoint trace. End with one optional next step fitting this outcome: a fenced, self-contained prompt naming any repository, issue, PR, or artifact it needs, or say none is needed. Only a "yes" starts it, via any skill it names, but the user runs a charter, diagnose, or ticket step as its /<skill> command.