Filter low-severity findings from review runs in this repository

[ralphbean]

Low-severity findings in our own review runs have mostly been noise rather than useful feedback.

Could we configure REVIEW_FINDING_SEVERITY_THRESHOLD: "medium" in .fullsend/harness/review.yaml for this repository?

This is only a repo-local configuration change for our use as Fullsend maintainers. It should not change the default for other users.

[fullsend-ai-triage]

🤖 Finished Triage · ✅ Success · Started 8:18 PM UTC · Completed 8:24 PM UTC

Commit: 4851188 · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $1.01

[fullsend-ai-triage]

Recommended implementation

Create .fullsend/harness/review.yaml using base: composition:

# .fullsend/harness/review.yaml
base: <raw-github-content-url>/fullsend-ai/agents/<SHA>/harness/review.yaml#sha256=<sha256sum>
env:
  runner:
    REVIEW_FINDING_SEVERITY_THRESHOLD: "medium"
  sandbox:
    REVIEW_FINDING_SEVERITY_THRESHOLD: "medium"

Then reference it in .fullsend/config.yaml under agents::

agents:
  - source: harness/review.yaml

Per AGENTS.md section 8 and docs/review.md, base: composition with per-key merge in env.runner/env.sandbox is the correct override path — not modifying the upstream harness/review.yaml default.

Both env.runner and env.sandbox must be overridden to keep the threshold in sync (as noted in the upstream harness comments).

Proposed test

# Verify the override harness is valid YAML and the base reference resolves
yq eval ".base" .fullsend/harness/review.yaml
yq eval ".env.runner.REVIEW_FINDING_SEVERITY_THRESHOLD" .fullsend/harness/review.yaml
yq eval ".env.sandbox.REVIEW_FINDING_SEVERITY_THRESHOLD" .fullsend/harness/review.yaml
# Both should output: medium

---

Labels: Repo-local review agent configuration change - enhancement scoped to the review-agent area.

---

Next steps:

• /fs-code — agent creates a PR to implement this issue
• /fs-code <your instruction> — agent implements with your specific guidance