What a stranger gets: docker run --rm -it ghcr.io/fmind-ai/a4h chat --demo answers in under 60 seconds — a real A2A round trip, no account, no credential, no config file.
Acceptance
- The repository builds: Go module and pinned toolchain live;
mise run format, check, test, build all green; hooks installed; CI runs the same tasks.
a4h chat --demo starts a terminal REPL against the bundled sample agent (an in-process A2A server): deterministic scripted answers, no model, no key, works air-gapped.a4h chat <url> runs the same REPL against any A2A endpoint (auth header via env).- Cold
docker run to first reply measured under 60 seconds.
Floor
AGENTS.md §4 applies to every slice. The bundled agent never calls a model — permanent invariant.
Non-goals
Streaming polish (S5) · platform adapters (S3, S7) · serve beyond what chat needs.
What a stranger gets:
docker run --rm -it ghcr.io/fmind-ai/a4h chat --demoanswers in under 60 seconds — a real A2A round trip, no account, no credential, no config file.Acceptance
mise run format,check,test,buildall green; hooks installed; CI runs the same tasks.a4h chat --demostarts a terminal REPL against the bundled sample agent (an in-process A2A server): deterministic scripted answers, no model, no key, works air-gapped.a4h chat <url>runs the same REPL against any A2A endpoint (auth header via env).docker runto first reply measured under 60 seconds.Floor
AGENTS.md§4 applies to every slice. The bundled agent never calls a model — permanent invariant.Non-goals
Streaming polish (S5) · platform adapters (S3, S7) ·
servebeyond whatchatneeds.