📝 docs(spec): accept S-0096 and S-0097, with five amendments

[Misery7100]

This accepts both 0.2 design documents with the leans the owner agreed to ("Agreed with your leans").

Amendments first (commit 1):

Amendment	Row	Change
S-0092/A-1	D-3 (ASSUMED)	A halt is locked_conflict only when it cites a LOCKED row; every other halt escalates underspecified (S-0096/D-1)
S-0093/A-1	D-3, D-4 (ASSUMED)	The battery's round exists only where a review leg runs; elsewhere a red battery escalates at once (S-0096/D-2)
S-0090/A-1	D-2 (ASSUMED)	An underspecified halt continues from its checkpoint like a locked one (see below)
S-0084/A-3	D-9 (LOCKED, the owner's call)	HTML comments are set aside with <details> (S-0097 Q-1); a thread on a landing record or AGENTS.md projection gets one fixed reply and is resolved when a configured bot opened it (Q-2)
S-0083/A-2	D-7 (ASSUMED)	While a review wave's rounds are outstanding, a landing publishes nothing; the last one publishes (S-0097/D-6)

One addition to S-0096 that the draft missed. runner._continuable resumes only blocker_finding and locked_conflict escalations from their checkpoint (S-0090/D-2). D-1 reclassifies most halts as underspecified, so without a change they would quietly restart from base. S-0096/D-1 now carries the S-0090/D-2 amendment, and phase 1 takes runner.py and tests/test_run_loop.py. This also fits S-0094's amend-then-requeue loop: the retry resumes where it stopped.

Acceptance (commit 3):

• S-0096: Q-1 is settled by D-4, so a missing prefix-named test refuses the contract.
• S-0097: Q-1 is settled by D-1 and Q-2 by D-2.

Commit 2 corrects #32's source to its squash sha, cecce075.

spec check reports 0 problems, and dry-run plans mint both documents. Once this merges, I'll serve S-0096 first; S-0097 follows after it merges, because the two share lane.py and cli/manager.py.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EabbhASMvWNhC7x4K9NJMF