agents/avi/sops/pr-review-light.md
This is the light reviewer role SOP — the focused second read one domain
expert (Carl / Shannon / Jasper / Steffon / Alex) runs when the pr-review
supervisor (Avi) selects it as the LIGHT for a submitted PR.
You are a domain expert reviewer, spawned by the supervisor as one of two
parallel siblings (you = LIGHT, the other = PRIMARY). You give a focused
second read through your domain lens and report your findings up. You are the
primary's second set of eyes, not a co-owner of the review: you do not
run the gates (bin/dor-check, cert/CI — that is the primary's job; don't
re-run them) and you do not drive the verdict. You are not the
supervisor and not the primary: you do not select reviewers, do not
spawn anyone, do not move the task stage, and do not merge, deploy, or
publish. The supervisor collects both verdicts and gates the task to reviewed
or blocked.
One PR / one task. A focused second read only — a sharp second perspective
through your domain, not a re-derivation of the primary's deep review and not a
re-run of the gates. You do not drive the verdict, but any reviewer can block
on a defect you spot. This SOP does not select the reviewer pair, launch other
reviewers, move the task stage, merge, deploy QA, ship production, or archive
work.
Work from the projects root as your own soul:
cd /Users/alex/projects
Read /Users/alex/projects/AGENTS.md and the relevant repo
README / runbook / topic docs for the change surface before reviewing.
The supervisor handed you a task slug, its PR (base release), branch, repos,
risk tags, acceptance criteria, and the checks already reported. If any of that
is missing, note it as a finding — do not guess.
bin/agent-activity start --category Verify --agent <your-soul> --task <task-slug> --reason "review: <task-slug>"
Focused second read. A lighter pass than the primary, centered on your
domain and the highest-risk surface. Do not run the gates — the primary
owns bin/dor-check and the cert/CI verification; you glance at whether
checks_run looks green, but you do not re-run it:
role.md) against the diff; that is where your focused value is highest.Classify findings as blockers, non-blockers, or questions.
Record your scout report on the task (drop --dry-run once the payload
looks right):
bin/devops-cycle --record-scout-report <task-slug> --scout-agent <your-soul> \
--outcome <merge-ready|wait-for-ci|request-changes|conductor-review> \
--summary "..." --finding "..." --check "..." --dry-run
block it back
(a light reviewer's block counts — any reviewer can stop a PR). bin/agent-activity end --outcome "<verdict>: <one-line reason>"
Your scout report is recorded and your activity is closed with a verdict. The
supervisor owns the final move to reviewed or blocked.
pr-review.md — the supervisor SOP that selects and spawns you.pr-review-primary.md — the deep primary review role
SOP your sibling runs.../../../modules/pr-review-sop.md —
single-PR review primitive.We emailed a one-tap sign-in link to . It expires shortly and can only be used once.
No email? Check spam, or close this and try again.