Tighten publish_work_checkpoint tool Description and schema field descriptions so evidence modes have clear required/forbidden fields; material_record defaults to no evidence.
Adjust the material task prompt: 原样记载 uses material_record without evidence; assumptions wording scoped to analysis.
Evaluation allows one checkpoint correction (stop only after two rejected checkpoints; reason repeated rejected work checkpoint); add label approved-o2-a01-evidence.
Why
O2 P1 proved the route contract, but A01 stopped on the first structurally invalid evidence payload. Approved design (Dev #662) is zero-paid guidance + one eval correction.
Test plan
go test ./internal/runtime/ ./internal/runmgr/ ./internal/server/ -count=1
go test -tags live -run '^TestFollowupA01KeepsItsOwnOneShotResult$' ./eval/live/ -count=1 + go vet -tags live ./eval/live/
go test ./... -count=1 + go vet ./... (host, HEAD 837480d)
## Summary
- Tighten `publish_work_checkpoint` tool Description and schema field descriptions so evidence modes have clear required/forbidden fields; material_record defaults to no evidence.
- Adjust the material task prompt: 原样记载 uses material_record without evidence; assumptions wording scoped to analysis.
- Evaluation allows one checkpoint correction (stop only after two rejected checkpoints; reason `repeated rejected work checkpoint`); add label `approved-o2-a01-evidence`.
## Why
O2 P1 proved the route contract, but A01 stopped on the first structurally invalid evidence payload. Approved design (Dev #662) is zero-paid guidance + one eval correction.
## Test plan
- [x] `go test ./internal/runtime/ ./internal/runmgr/ ./internal/server/ -count=1`
- [x] `go test -tags live -run '^TestFollowupA01KeepsItsOwnOneShotResult$' ./eval/live/ -count=1` + `go vet -tags live ./eval/live/`
- [x] `go test ./... -count=1` + `go vet ./...` (host, HEAD 837480d)
- [x] Independent code-reviewer PASS (fresh session)
## Refs
- Design: Agent-System-Dev `docs/superpowers/specs/2026-10-20-fast-slow-o2-a01-evidence-design.md` (#662)
- Plan unit: `o2a01e-i1-backend-evidence-guidance`
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Summary
publish_work_checkpointtool Description and schema field descriptions so evidence modes have clear required/forbidden fields; material_record defaults to no evidence.repeated rejected work checkpoint); add labelapproved-o2-a01-evidence.Why
O2 P1 proved the route contract, but A01 stopped on the first structurally invalid evidence payload. Approved design (Dev #662) is zero-paid guidance + one eval correction.
Test plan
go test ./internal/runtime/ ./internal/runmgr/ ./internal/server/ -count=1go test -tags live -run '^TestFollowupA01KeepsItsOwnOneShotResult$' ./eval/live/ -count=1+go vet -tags live ./eval/live/go test ./... -count=1+go vet ./...(host, HEAD837480d)Refs
docs/superpowers/specs/2026-10-20-fast-slow-o2-a01-evidence-design.md(#662)o2a01e-i1-backend-evidence-guidance