[fullsend-ai-coder]

Summary

Changes in this repository:

• behaviour/features/review/post-review-outcomes.feature: in "A rejecting review closes the pull request", the second /fs-review round now adds And the harness "review" run was skipped by its pre-script right after the workflow step.
• go.mod / go.sum: bump github.com/fullsend-ai/fullsend from 6d925afef to fullsend main 3ea9dd1b8 (v0.50.1-0.20261024212938-3ea9dd1b82e4). That version adds the step. The old pin's go.sum entries are replaced, matching previous bumps. fullsend's go.mod hash is unchanged.

Why

Before this change, the second round checked only that the review workflow succeeded and that the "Review skipped" comment was posted. Both pass even if the agent ran anyway. The new step downloads the fullsend-review artifact for that run. It requires fullsend.prescript.skipped=true on the root run span in the host telemetry, and it requires that there is no agent-result.json or behaviour-results.json. So it proves pre-review stopped the stage before the sandbox was created.

Context

• fix(#1782): skip review stage on closed PR or skipped author #1783: pre-review's closed-PR skip now stops the stage.
• test(behaviour): step to assert a stage was skipped by its pre-script fullsend#8792 / test(#8792): add step asserting a stage was skipped by its pre-script fullsend#8793 (merged as 3ea9dd1b8, currently fullsend main HEAD): add the step.

Testing

• The behaviour package (-tags behaviour) vets and compiles against the new pin, using the same -modfile + local replace setup as behaviour-tests.yml.
• The new feature line matches the registered step regex ^the harness "([^"]+)" run was skipped by its pre-script$ and captures review.
• harness/triggertest passes against the new pin. It reads the root go.mod pin's fixtures.
• make test could not run through make (the sandbox blocks /bin/sh). I ran every script-test command directly: 31 pass and the same 15 environment-only failures appear on main and on this branch.
• skillsaw --strict (the make lint command) passed.
• pre-commit could not fetch hook repos. I ran the matching pre-commit-hooks v6.0.0 checks and lint-release-cases directly and they passed. scan-secrets (gitleaks) passed.
• Not run: live behaviour scenarios. The merge criterion (Agents behaviour green with the reject scenario passing, not skipped, plus the job duration) has to be checked in CI.

🤖 Generated with Claude Code

---

Closes #1784

Post-script verification

• Branch is not main/master (agent/1784-prove-review-skip)
• Secret scan passed (gitleaks — 5543e099221f08daf2d1efdbe7cc62efa2c9252b..HEAD)
• PR body secret scan passed (gitleaks — no-git)

[waynesun09]

/fs-fix-stop

[github-actions]

Fix agent disabled for this PR. Remove the fullsend-no-fix label or use /fs-fix to re-engage.

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 10:14 PM UTC · Completed 10:21 PM UTC

Commit: 7171358 · View workflow run →

Runtime: pi · Model: sonnet → claude-sonnet-5-5 · Effort: high · Cost: $0.97

[fullsend-ai-review]

<!-- fullsend:risk-assessment -->
Risk Assessment: moderate (2/5)

Details

Small, tightly scoped 3-file PR aligning with issue requirements, with risk elevated to moderate due to dependency updates in go.mod and go.sum alongside recent churn on those manifests.

[fullsend-ai-review]

Looks good to me

[waynesun09]

Approving at 7171358: fullsend bump to include #8793; behaviour run 40137590399 passed 11/11 with the reject scenario now proving the closed-PR review skip stops the stage; bot approved, no threads.