8000 refactor(minimal): run session activation as named, spanned steps · gominimal/minimal@c9a38dd · GitHub
[go: up one dir, main page]

Skip to content

refactor(minimal): run session activation as named, spanned steps

Minimal Arch / arch-review succeeded Oct 20, 2026

PASS (the design is silent)

Governs: nothing. This is a PASS by absence (rubric rule 1), the same as #2452 (T34) on the same file area. The checkout at 109541a4 has no arch-map.md, so I worked from the architecture.md table of contents. I searched architecture.md and specs/ for activation, spans, tracing, RUST_LOG and client logging. The min CLI (line 321 and after) governs the command grammar, Exit codes (line 353) and Output conventions (line 375). The [otel] table (line 1100) configures OTEL for the box env, not the client log. No line governs how activate_session is structured, or the spans that the client writes to its own log.

What I checked by structure. pr.diff equals git diff 28ab14cb c9a38dd1. The whitespace-insensitive diff is +367/-18. Most of it is the new ActivationStep enum, ActivationStep::run, and the test module. The step bodies stay inline. Each early return in the wrapped bodies of the base activate_session is a return Err(...). Inside an async block it now ends the step with that Err, and .await? propagates 28E2 it unchanged. So the error values, the main.rs downcasts and the exit codes do not change. The early return cmd_attach(...) stays outside any step. ActivationStep::run logs the failure at info with the step name and the {error:#} text. That text is the same text that the CLI already prints on stderr. It is a log record, not output, and no new message prints before the shell. The only change in order is let id = created.id;, which moves up to just after the CreateSession step. That is a pure binding.

Not checked. findings.json was not staged, so I had no minimal-review findings to cross-reference. I did not build. The UploadFiles wrapper (create.rs lines 1133 to 1242) is not in the shape that rustfmt gives the other steps. The PR body says rustfmt re-indented the blocks, so just fmt --check can flag this block. That is a code-quality matter for minimal-review and CI, not a design question.

Ruled at c9a38dd against gominimal/arch@109541a48c31 by claude-opus-5-5; arch_rulings 113, job e18ab759-afca-4420-90e3-b324c643cb5b.

0