Dogfood plan

f0rge manages its own repo, locally, no docker. Contract first: every agent reads f0rge --skill before acting.

Loop (every change)

1. f0rge issue create (label dogfood) for the defect/idea
2. agent fixes on a branch -> f0rge pr create --issue <id>
3. orchestrator f0rge merge -> local CI (.forge/ci/steps/)
4. green -> git commit to main; f0rge issue close conflicts stop; agent resolves (bitter lesson)

Bootstrap (#2, dog-boot)

• binary on PATH; init skeleton committed
• docs -> wiki: AGENTS OUTLINE PLAN build-log audit -> wiki/docs/
• CI local: steps 00_audit 01_test 02_budget 03_release

Domains (waves, disjoint cmd modules)

• issues/labels/milestones: real tracker for this dogfood
• prs/merge: the fix-loop is the test
• tasks: kanban for the sprint
• wiki: docs live here, index derived
• releases: tag + notes + build steps
• ci: every merge gates on it (rerun/cancel/log)
• skill/security: --skill + SEC fixtures + error discipline
• scale: 10k list, 500 wiki within budget

Exit

all domains exercised for real on this repo + its history; no derived state stored; budgets + cargo audit green.