Dispatch deferred code analysis

Dispatch deferred-analysis functions to codex GPT-5.5 via driver pools

Sweeping a finite backlog of deferred decompilation targets with fresh tooling and codex reasoning to escape prior plateaus.

• · deferred-analysis/*.md files (with drift notes)
• · optional: candidates list, parallel count (K), maxTargets, codex model/effort config

• · per-target: commited matches or deferred rationale
• · scoreboard + worktree cleanup
• · git log with new commit SHAs

• · git worktree
• · make build system
• · codex plugin runtime (GPT-5.5-high)
• · bash shell

• · git repo initialized at /Users/jackrickey/Dev/frog-adv-decomp
• · gitignored deps symlinked (tools/agbcc, corpus-mirrors, vendor submodules)
• · make check passes on pristine tree
• · deferred-analysis markdown files exist with prior drift notes

• · Bootstrap symlinks missing → permuter/corpus/compiler tooling unavailable
• · Parallel agents edit same file → cherry-pick conflict during integrate phase
• · make check fails mid-integration → worktree reverts, others may continue
• · codex dispatch timeout or max-sub bound hit → target deferred for next sweep

• · Codex model explicitly specified (GPT-5.5-high) — not fallback
• · Prior drift notes loaded before each codex dispatch
• · Cherry-pick + make check gates every integration
• · Worktree mode isolates parallel work, rolling pool size (K) configurable