Decide what a design blocking condition costs the rest of a /specs run

What surfaced

#404 decided that a design-conventions doc may declare blocking conditions and that they stop /specs. It decided nothing about what a block costs the rest of the run, and #416 had to answer that to make the prose executable.

What commands/specs.md now says, marked in the file as this command's rule rather than the doc's:

• No contract for the component that blocked. Never a partial or empty one — an empty contract reads as a component with no properties, not as one nobody could read.
• No contract for the components after it either.
• Print the Step 9 summary with the blocked: line filled and gate: not opened, then stop before the gate.
• Contracts already written this run stay on disk unstaged, and the summary lists them as written-not-staged. The PRD stays on disk, and Step 2 picks it up as a re-entry.

The decision

Fail-fast on the whole set, or write every contract that can be written and report the blocked ones at the gate? The trade is visible: fail-fast never puts a half-set in front of a human, and write-what-you-can keeps a nine-component feature moving when one component is untidy in the design file.

The current answer is fail-fast, chosen because a partial set reaching the gate is the shape a human approves without noticing. Whoever reverses it owns the summary line that makes a partial set legible.

Evidence for either side

Measured 2026-08-25 on the Consumer (#403, #405): focus is drawn nowhere on all four interactive sets and three typos stand, so the first real run blocks on the first component it reads. That makes fail-fast indistinguishable from write-what-you-can today, and the difference only appears once the design side is clean.

Start with commands/specs.md, especially the blocking-condition rule and Step 9 summary, then read issues #404 and #416 for the existing decisions. Compare fail-fast with write-what-you-can against the Consumer evidence in #403 and #405; done means the chosen policy and its summary behavior are unambiguous and recorded.