Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/collect-gates.sh:
- Around line 38-39: The jq expressions computing ci_failure_count and
ci_pending_count fail when .statusCheckRollup is null; update both queries to
null-coalesce .statusCheckRollup to an empty array (e.g. use .statusCheckRollup
// []) so the select/iteration always runs on an array; modify the expressions
that set ci_failure_count and ci_pending_count to use (.statusCheckRollup // [])
in place of .statusCheckRollup.

In @.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/find-candidates.sh:
- Around line 36-38: The gh CLI search invocation assigned to the candidates
variable uses an invalid flag (--in:body); move the body qualifier into the
query string so gh search prs receives the full query (e.g., include "in:body
#${issue_number}" with repo_args), retry the search without suppressing stderr
while preserving piping to sort/head, and ensure the command still outputs JSON
numbers for the jq filter; update the command that builds candidates (using gh
search prs, repo_args, issue_number, MAX_CANDIDATES) accordingly so it actually
finds PRs containing the issue number in the body.

In @.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/render-verdict.py:
- Around line 113-118: tier_2_keys currently contains "workaround_vs_root_cause"
but there is no corresponding check in checks/tier-2-quality.md; fix by either
removing "workaround_vs_root_cause" from the tier_2_keys list in
render-verdict.py (so the code and docs align) or add a formal definition and
scoring for the workaround_vs_root_cause check in checks/tier-2-quality.md (and
update any max-score calculations/templates accordingly) — locate the symbol
tier_2_keys in
.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/render-verdict.py and
make the change consistently (removal from the list OR a documented check entry
and score in tier-2-quality.md).
- Around line 205-209: The current prints use the variable winner directly and
produce confusing output when winner is None; update the code that handles mode
and winner (variables mode and winner) to validate winner first and substitute a
clear fallback (e.g., "no PR selected" or "no winner") before formatting
messages, so in the "happy" branch you print "VERDICT: MERGE PR #<id>" only when
winner is not None and otherwise print "VERDICT: no PR selected", and in the
degraded branch print "Neither mergeable yet" and "No PR is closer" (or similar)
when winner is None; ensure all f-strings using winner are guarded or replaced
with the fallback string.

In @.agents/skills/nemoclaw-maintainer-pr-comparator/templates/verdict.md:
- Line 34: The example in the verdict template shows an incorrect max score
(21.0); update the "Weighted score" row in
.agents/skills/nemoclaw-maintainer-pr-comparator/templates/verdict.md to reflect
the true maximum based on actual check counts and weights (Tier 1 = 2.0× per
check, Tier 2 = 1.0× per check). Reconcile counts with render-verdict.py (which
currently includes the Tier‑2 check workaround_vs_root_cause) and compute the
correct max (e.g., 6 Tier‑1 checks ×2 + 4 Tier‑2 checks ×1 = 16.0, or adjust if
checks differ), then replace both the numerator and denominator values so the
example matches the code.

In @.agents/skills/nemoclaw-maintainer-pr-comparator/validation/backtest.md:
- Around line 22-24: The gh search prs command uses search qualifiers inside the
query string, not as flags; replace the incorrect snippet gh search prs --repo
OWNER/REPO --merged --in:body "supersedes" --limit 30 by moving the qualifier
into the quoted query, e.g. gh search prs --repo OWNER/REPO --merged "in:body
supersedes" --limit 30 (or include repo/is:merged qualifiers inside the query as
needed) so the in:body qualifier is part of the query string.

---

Nitpick comments:
In
@.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/check-coderabbit-threads.sh:
- Around line 98-105: The heredoc that emits JSON uses the unescaped variable
$bot_login which can break JSON if it contains quotes/backslashes; update the
script (check-coderabbit-threads.sh) to build the JSON using a safe tool like jq
or to explicitly JSON-escape $bot_login before emitting (e.g., construct the
object with jq --arg pr "$pr" --arg bot_login "$bot_login" --argjson
gate_coderabbit_threads_resolved "$gate_pass" --argjson details "$counts" '.pr =
($pr|tonumber?) | .bot_login = $bot_login | .gate_coderabbit_threads_resolved =
$gate_coderabbit_threads_resolved | .details = $details' or equivalent) so the
output is always valid JSON and references the same variables ($pr, $bot_login,
$gate_pass, $counts).

In
@.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/parse-supersession.sh:
- Around line 24-26: The handler for the --repo option in parse-supersession.sh
uses "$2" and shift 2 without validating that a value exists; update the --repo
case to check that a non-empty argument is present (e.g., verify $# -ge 2 or
that "${2:-}" is non-empty) before assigning repo_args=(--repo "$2") and
performing shift 2, and if the check fails print a clear error/usage message and
exit with non-zero status so the script doesn't hit an unbound-variable or
broken shift.

In @.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/render-verdict.py:
- Around line 181-186: spec["issue"] can raise a KeyError if the input dict
lacks required keys; update the code that reads spec (the block assigning issue,
criteria, prs, winner, mode, tiebreaker) to validate required keys up front
(e.g., check for "issue" and any other required fields) and either use
spec.get(...) with sensible defaults for optional fields (criteria, prs, winner,
mode, tiebreaker) or raise a clear ValueError/TypeError with a descriptive
message when a required key is missing so callers get an actionable error
instead of a cryptic traceback.

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In @.agents/skills/nemoclaw-maintainer-pr-comparator/scripts/render-verdict.py:
- Around line 129-134: The current total uses all keys from
pr.get("tier_1"/"tier_2") which lets extra/hidden keys change rankings; change
the summation to iterate over the canonical key lists (tier_1_keys and
tier_2_keys) and call score_for(pr.get("tier_1", {}).get(key), TIER_1_WEIGHT)
and similarly for tier_2, so only the fixed keys affect total; keep max_total
as-is (len(tier_1_keys)*TIER_1_WEIGHT + len(tier_2_keys)*TIER_2_WEIGHT) and
ignore any extra keys in pr when computing total.