Canonical skills
sdlc-definition, sdlc-migration, and sdlc-audit remain harness-neutral source.
AgentFlow 1.0 · Work-surface integration
One SDLC model, multiple agent harnesses.
AgentFlow 1.0 contains canonical skills, native adapter and plugin manifests, routing contracts, and structural settings merge for Claude Code, Codex, Agy, and Pi.
Native adapters and routing
Claude Code, Codex, Agy, and Pi can resolve supported execution targets without making generated harness folders canonical product source.
Safe settings merge
Structural merge preserves project-owned keys, backs up changed files, and blocks unsafe roots.
Identity is not the same as execution.
AgentFlow records the platform that produced evidence separately from execution target, transport, delegation boundary, context boundary, and model.
ChatGPT, Cowork, Antigravity, human work, and future registered platforms can provide truthful evidence identity without implying that a native adapter or routable executor exists.
Validation commands
node bin/cli.mjs plugins validate --harness all --json
node bin/cli.mjs settings validate --harness all --json
node bin/cli.mjs settings merge --harness all --dry-run