Launch parallel agents for conjecture research and build step implementation. Use after bootstrap is verified and build steps are ready.
Before launching agents, check what work already exists:
# Check for existing research docs
for i in $(seq -w 1 8); do
test -f "docs/conjectures/c-0${i}-research.md" && echo "EXISTS: C-00${i}" || echo "MISSING: C-00${i}"
done
# Check for existing design docs
ls docs/design/*.md 2>/dev/null && echo "DESIGN_DOCS: found" || echo "DESIGN_DOCS: none"
# Check build step completion
test -f src/emcp-stdio.el && echo "emcp-stdio.el: exists" || echo "emcp-stdio.el: missing"
test -f src/introspect.el && echo "introspect.el: exists" || echo "introspect.el: missing"
test -f tests/test-emcp-stdio.el && echo "test-emcp-stdio.el: exists" || echo "test-emcp-stdio.el: missing"
docs/conjectures/c-NNN-research.md.docs/design/<component>.mdRead CLAUDE.md and spec.org to identify parallelizable work.
For each conjecture (C-001 through C-008):
docs/conjectures/c-NNN-research.mdFor the current unblocked build step:
For steps that are blocked but need design work:
docs/design/<component>.mdAs agents complete:
Per MAST taxonomy (arXiv 2503.13657): value is in isolation and determinism, not raw throughput. Each agent gets its own worktree or output directory. No shared mutable state between agents.