kyleandClaude Fable 5.1
74b8af4ce0
v2 given tests compiled against a panic-only skeleton (go vet clean); no reference implementation. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
1.7 KiB
1.7 KiB
v1.1 implementation plan: review follow-ups
For the implementing model: do not work from this file. The owner gives you one task file at a time. This file is the index for the owner and the reviewer.
Goal: close findings 1 and 2 of the v1 review (docs/implementer-log.md): a request whose
client disconnects — mid-stream or while queued — must still write its accounting row (status
499), and /_crossbar/usage with no rows must answer [], not null.
How this plan was made: acceptance tests first, from the findings; no reference
implementation. Both given tests were run against master at the merge of v1: all three fail
there (the two cancel tests find no 499 row; the empty-usage test gets null).
Tasks
| # | File | Delivers | Tests that define it |
|---|---|---|---|
| 01 | 01-review-fixes.md |
499 rows on both cancel paths; [] for empty usage |
internal/proxy/cancel_test.go, internal/admin/usage_empty_test.go |
Branch v1.1. One task, one fresh OpenCode session, one commit.
For the reviewer
git log --oneline master..v1.1: one commit with the trailer.cmpboth copied tests;git diff master..v1.1 --stat -- PLAN.md AGENTS.md docs/plansempty.make gate,make smoke.- Probe: cut a stream with
curl -m 0.4 -N …against the smoke rig and confirm onestatus="499"line in/_crossbar/metrics.
Changes during the run
- 2026-09-25, task 01, first session: ended after ~8 min with no commit and no row, right after
the sandbox refused a
/tmpscratch program (the I9 pattern, third time tonight). The rule against ending a turn on a refusal lived only in v1's task 05; it is now inAGENTS.md, so every task carries it. Resumed from the working tree.