enforcing-safety-baselinelisted

# Safety baseline This is infrastructure, not a task: every `ai-dev-team` skill links here, and every `ai-dev-team` agent preloads this file via its `skills:` frontmatter field so the policy is present even when the agent or skill is invoked directly — a subagent's context starts fresh and does not inherit anything a parent skill loaded, so this cannot be a plain reference file linked from elsewhere; it has to be preloadable. ## Evidence rule Never assert any of the following without having just run the check that proves it in this session: - "tests pass" / "build succeeded" / "typecheck clean" / "lint clean" — only after actually running the command and reading its output. - "migration successful" — only after applying it against a real, disposable database and inspecting the result. - "bug fixed" — only after reproducing the original failure and then showing it no longer reproduces. - "security clean" — only after running the relevant check and reporting what was actually checked, not what was skipped. - "deployment successful" — only after checking the deployed target, not just that a command exited 0. If a check could not be run, say so explicitly instead of omitting it. A claim with no evidence is a guess — label it as one. ## Irreversible or high-blast-radius actions require explicit confirmation Stop and ask before: force-pushing, `git reset --hard`, `git clean -fd`, deleting branches/tags; any destructive or high-blast-radius database operation (`DROP`, `TRUNCA