Verify each finding against the current code and only fix it if needed.

Inline comments:
In @.agents/skills/nemoclaw-maintainer-day/HOTSPOTS.md:
- Around line 1-46: Do not edit HOTSPOTS.md directly; instead update the
original Hotspots source document that feeds the markdown generator, then run
the docs-to-skills.py generator to regenerate the skill artifact (ensure the
change appears in the generated HOTSPOTS.md), and commit both the source doc and
the regenerated output so CI remains consistent.

In @.agents/skills/nemoclaw-maintainer-day/MERGE-GATE.md:
- Around line 1-47: This file is autogenerated and must not be edited directly;
revert any changes to .agents/skills/nemoclaw-maintainer-day/MERGE-GATE.md and
instead make your content edits in the canonical source docs, then run the
generator script scripts/docs-to-skills.py to regenerate the skill file;
reference the autogenerated path
(.agents/skills/nemoclaw-maintainer-day/MERGE-GATE.md) and the generator
(scripts/docs-to-skills.py) when updating so future regenerations don't
overwrite manual edits.

In @.agents/skills/nemoclaw-maintainer-day/SALVAGE-PR.md:
- Around line 1-84: The SALVAGE-PR.md content was edited manually but must be
generated; revert manual edits to the "Salvage PR Workflow" markdown and
regenerate it from the canonical source using the docs-to-skills generator (run
the docs-to-skills.py tool to rebuild the skill artifact), then commit the
autogenerated output instead of hand-editing; ensure the generator is run and
its output replaces the current SALVAGE-PR.md so the file matches the canonical
source.

In @.agents/skills/nemoclaw-maintainer-day/scripts/bump-stragglers.ts:
- Around line 36-37: The current extraction of the repo value from args uses
repoIdx without ensuring an argument exists after the flag (repoIdx >= 0 ?
args[repoIdx + 1] : ...), which can read undefined or throw; update the check in
bump-stragglers.ts so you only access args[repoIdx + 1] when repoIdx >= 0 &&
repoIdx + 1 < args.length, otherwise fall back to the default "NVIDIA/NemoClaw";
reference the variables repoIdx, args and repo when making the change.

In @.agents/skills/nemoclaw-maintainer-day/scripts/hotspots.ts:
- Around line 112-116: Validate the parsed CLI args before use: check that
daysIdx >= 0 and the token at args[daysIdx + 1] exists and parseInt(args[daysIdx
+ 1], 10) yields a finite number (not NaN) before assigning to days, otherwise
throw a clear error or fallback to the default 30; similarly ensure repoIdx >= 0
and args[repoIdx + 1] is a non-empty string before assigning to repo, otherwise
throw a clear error or fallback to "NVIDIA/NemoClaw". Update the code around
daysIdx/days and repoIdx/repo and add explicit validation checks prior to
calling gitChurn() and openPrFileOverlap() so downstream git/gh invocations
never receive NaN or undefined values and instead fail fast with a helpful
message.

In @.agents/skills/nemoclaw-maintainer-day/scripts/state.ts:
- Around line 179-192: The cmdSetQueue function currently calls JSON.parse on
stdin without error handling; wrap the readFileSync + JSON.parse logic in a
try-catch inside cmdSetQueue (around the input/triageOutput parse) to catch
SyntaxError and other JSON errors, log a clear user-friendly error via
console.error including the raw input or the parse error message, and exit
non-zero (process.exit(1)) so callers see failure; reference cmdSetQueue,
readFileSync(0, "utf-8"), JSON.parse(input), loadState, saveState and ensure
normal state update only proceeds when parsing succeeds.
- Around line 194-205: The cmdSetHotspots function lacks error handling around
reading and parsing stdin; wrap the readFileSync/JSON.parse of input in a
try-catch inside cmdSetHotspots, catch and log a descriptive error via
console.error (including the caught error.message or error object) and exit with
a non-zero status (e.g., process.exit(1)) on failure; also validate that
hotspotOutput.hotspots is an array before assigning to state.hotspots.files and
log a clear error if the structure is invalid so callers get a helpful message.
- Around line 146-154: cmdExclude currently always writes exclusions to
state.excluded.prs, but the function is documented as "Add PR/issue to permanent
exclusion list" so it must distinguish PRs vs issues; update cmdExclude to
accept a type flag or detect item kind and write to either state.excluded.prs or
state.excluded.issues accordingly (e.g., add a boolean parameter like isIssue or
check a CLI --issue flag from the caller), ensure the CLI handler is updated to
pass that flag, and persist to state.excluded.issues with the same shape
(reason, excludedAt) when the item is an issue, leaving existing PR behavior
unchanged.

In @.agents/skills/nemoclaw-maintainer-day/scripts/triage.ts:
- Around line 165-167: The --approved-only filter in fetchOpenPrs is ineffective
because fetchOpenPrs filters on pr.reviewDecision === "APPROVED" before enrichPr
populates reviewDecision; move the approved-only logic out of fetchOpenPrs and
into main() after calling enrichPr (or remove the flag entirely). Specifically,
stop filtering by reviewDecision inside fetchOpenPrs, call enrichPr for each PR
(using enrichPr), then if approvedOnly is true filter the enriched PR array by
pr.reviewDecision === "APPROVED" before further processing.
- Around line 302-342: The scoreItem function has a type/logic mismatch: the
bucket variable and the function return type should include "merge-now" and
"review-ready" (not the currently declared "ready-now"), and the security-path
conditional is dead/wrong (it checks bucket === "ready-now"); update the bucket
type union in scoreItem and the function return type to "merge-now" |
"review-ready" | "salvage-now" | "blocked", and change the security-files branch
to test whether bucket is one of the merge-capable buckets (e.g., bucket ===
"merge-now" || bucket === "review-ready") so nextAction becomes "security-sweep
→ merge-gate" for those buckets and "security-sweep → salvage-pr" otherwise;
keep references to bucket, nextAction, scoreItem, and riskyFiles.

In @.agents/skills/nemoclaw-maintainer-day/scripts/version-progress.ts:
- Around line 63-65: The repo lookup currently reads args[repoIdx + 1] into repo
without validating it, so when `--repo` is passed with no value you end up
passing undefined/"undefined" to gh; change the logic around repoIdx/repo to
validate that args[repoIdx + 1] exists and is a non-flag string (e.g., not
starting with "--" and not the literal "undefined"), and if the value is
missing/invalid throw an Error or call process.exit(1) with a clear message;
update the code that defines repo (the repoIdx and repo variables in
version-progress.ts) to perform this check before using repo in the gh query.

In @.agents/skills/nemoclaw-maintainer-day/scripts/version-target.ts:
- Around line 110-112: The repo extraction logic using args, repoIdx and repo is
missing a bounds check for the value at args[repoIdx + 1]; update the code that
computes repo to verify repoIdx >= 0 and repoIdx + 1 < args.length and that
args[repoIdx + 1] is a non-empty string, otherwise fall back to the default
"NVIDIA/NemoClaw"; modify the expression that sets repo (or add a short
conditional) so passing "--repo" as the last argument won't set repo to
undefined.

In @.agents/skills/nemoclaw-maintainer-day/SKILL.md:
- Around line 1-79: The SKILL.md for the skill named "nemoclaw-maintainer-day"
is autogenerated and must not be hand-edited; revert your manual edits to
SKILL.md and instead apply your changes to the upstream documentation source
used by docs-to-skills.py, then run the docs-to-skills.py regeneration step to
produce an updated "nemoclaw-maintainer-day" SKILL.md (verify helper scripts
like version-target.ts, version-progress.ts and state.ts still work), and commit
the regenerated skill output rather than committing direct edits to the
generated SKILL.md.

In @.agents/skills/nemoclaw-maintainer-day/TEST-GAPS.md:
- Around line 1-65: The TEST-GAPS.md file is autogenerated and must not be
edited directly; update the original source documentation that feeds the
generator and then run the generator script to regenerate this skill file:
modify the upstream docs (the source that maps to TEST-GAPS content), then run
scripts/docs-to-skills.py to produce the updated
.agents/skills/nemoclaw-maintainer-day/TEST-GAPS.md, committing the source
changes and the regenerated output rather than editing TEST-GAPS.md by hand.

In @.agents/skills/nemoclaw-maintainer-evening/SKILL.md:
- Around line 22-28: bump-stragglers.ts can silently continue when run() returns
an empty string, so update the script to detect empty/invalid run() results and
fail fast: after each call to run() in bump-stragglers.ts validate the return
(and any parsed JSON) and if missing or malformed (e.g. no "bumped" array) write
a clear error to stderr and exit non-zero; also ensure any label-creation or
PR/issue-editing failures from run() are surfaced (log the error output and
include it in the process exit) and add a final check that the emitted JSON
contains the expected "bumped" items before reporting success.

In @.agents/skills/nemoclaw-maintainer-morning/SKILL.md:
- Around line 1-65: The SKILL.md file
(.agents/skills/nemoclaw-maintainer-morning/SKILL.md) is autogenerated and must
not be edited directly; revert any manual changes here and update the
authoritative source content or the generator instead (update the original doc
used by scripts/docs-to-skills.py or modify scripts/docs-to-skills.py to produce
the desired change), then re-run the docs-to-skills.py generation to regenerate
SKILL.md so the change persists.
- Around line 29-32: Update the SKILL.md paragraph to reflect actual triage
behavior: replace the claim that the script "calls `gh-pr-merge-now --json`"
with a statement that triage.ts uses the GitHub CLI REST API (`gh api
--paginate`) to list PRs and enriches top candidates with `gh pr view --json`
(review status, CI status), performs risky-file detection, applies the scoring
model from PR-REVIEW-PRIORITIES.md, and also uses find-review-pr to add PRs
labeled `security` + `priority: high` into the candidate pool.

---

Nitpick comments:
In @.agents/skills/nemoclaw-maintainer-day/scripts/bump-stragglers.ts:
- Line 62: There are silent catch blocks (the empty "catch { /* ignore */ }" in
bump-stragglers.ts and the similar one around line ~80) that swallow parse
failures; replace each with a parameterized catch (e.g., "catch (err)") and log
the error to stderr with context using console.error (include what was being
parsed or the offending value and the error.message/stack) so failures are
visible for debugging, keeping execution flow otherwise unchanged.

In @.agents/skills/nemoclaw-maintainer-day/scripts/check-gates.ts:
- Around line 219-243: The current checkRiskyCodeTested function treats any test
file in the PR as sufficient for risky-file coverage, which allows unrelated
tests to pass the gate; update checkRiskyCodeTested to require tests that are
likely to exercise the risky files by matching test filenames or paths to the
riskyFiles (e.g., test file path contains the risky file basename, same
directory, or follows a naming convention like foo.test.ts for foo.ts) or
otherwise map riskyFiles to corresponding test files using isTestFile and
filename matching; ensure you reference and use the riskyFiles array and the
existing isTestFile/isRiskyFile helpers to compute a more precise hasTests value
and adjust the returned details and hasTests boolean accordingly.

In @.agents/skills/nemoclaw-maintainer-day/scripts/handoff-summary.ts:
- Around line 30-36: AREA_LABELS in handoff-summary.ts duplicates the risky file
patterns defined as RISKY_PATTERNS in shared.ts which risks future drift;
consolidate to a single source by either exporting AREA_LABELS (or the pattern
map) from shared.ts and importing it into handoff-summary.ts, or modify
shared.ts to export a single RISKY_PATTERNS/AREA_LABELS constant and update
isRiskyFile to derive its checks from that exported map so handoff-summary.ts
reuses the shared export instead of defining its own AREA_LABELS.

In @.agents/skills/nemoclaw-maintainer-day/scripts/shared.ts:
- Around line 34-38: The TEST_PATTERNS array currently only matches
/\.test\.[jt]sx?$/, /\.spec\.[jt]sx?$/, and paths starting with /^test\//;
expand it to cover other common test conventions by adding patterns for
/^tests?\// (both test/ and tests/), /__tests__\//, filenames ending with
/_test\.[a-z]+$/ (e.g., *_test.go), and filenames starting with
/^test_.*\.[a-z]+$/ (e.g., test_*.py); update the TEST_PATTERNS RegExp[] to
include these new regexes so test file discovery handles these conventions.

In @.agents/skills/nemoclaw-maintainer-day/scripts/triage.ts:
- Around line 398-404: The enrichment only runs for enrichCount =
Math.min(candidates.length, limit * 3), leaving later non-draft PRs without
statusCheckRollup and causing classifyPr to treat them as failing; change the
logic so all non-draft PRs are enriched before classification—either set
enrichCount = candidates.length or call enrichPr for every entry in candidates
(and await/Promise.all if enrichPr is async), ensuring statusCheckRollup is
populated for use in classifyPr and referenced checks (e.g., checksGreen).

In @.agents/skills/nemoclaw-maintainer-day/scripts/version-target.ts:
- Line 76: The empty catch in
.agents/skills/nemoclaw-maintainer-day/scripts/version-target.ts silently
swallows JSON parse errors (seen at the try/catch around the JSON.parse call at
lines ~76 and ~103); change the catch blocks to capture the error (e.g., catch
(err)) and log a warning to stderr (console.error or process.stderr.write)
including the error message and the raw gh output being parsed so malformed
output is visible for debugging before continuing.
- Around line 29-47: Extract the duplicated functions getLatestTag and bumpPatch
from version-target.ts and handoff-summary.ts into shared.ts and have both files
import them; specifically, move the implementations (which use the existing
run() helper) into shared.ts as exported functions getLatestTag and bumpPatch,
remove the local definitions from version-target.ts and handoff-summary.ts, and
update their imports to import { getLatestTag, bumpPatch } from './shared' (or
the module path already used for shared utilities).

Verify each finding against the current code and only fix it if needed.

Inline comments:
In @.agents/skills/nemoclaw-maintainer-day/scripts/triage.ts:
- Around line 406-414: The code currently only enriches the first enrichCount
(limit * 3) candidates and later only scores the first limit * 2 actionable
items, which biases results by fetch order; change the flow so all non-draft
candidates (the full candidates array) are enriched before
classification/scoring and ensure file-based scoring runs over the full set of
actionable PRs (not just a sliced subset). Specifically, remove or adjust the
enrichCount slicing and any subsequent slicing of actionable items so that
enrichPr is invoked for every item in candidates, classifyPr runs on the full
prs array as before, and the file-based scoring step processes all entries
produced by classifyPr (refer to candidates, enrichPr, enrichCount, classifyPr,
classified and the actionable/file-based scoring variables around the existing
limit-based slices).
- Around line 281-287: fetchPrFiles currently fetches only the first 100 PR
files and drops the rest; update fetchPrFiles to paginate like fetchOpenPrs by
calling ghApi with the --paginate option (or the same pagination helper used by
fetchOpenPrs), accumulate filenames across all pages, keep the return type
Array<{ filename: string }> | null handling, and return the concatenated list of
filename strings (preserve the existing Array.isArray check and empty-array
fallback).

---

Duplicate comments:
In @.agents/skills/nemoclaw-maintainer-day/scripts/triage.ts:
- Around line 326-329: The current block treating riskyFiles sets nextAction to
"security-sweep → review" for any non-"blocked" bucket, which incorrectly routes
bucket === "salvage-now" into review; update the conditional in triage.ts so
that when riskyFiles.length > 0 and bucket !== "blocked" you still add
SCORE_SECURITY_ACTIONABLE to score but set nextAction based on bucket values:
preserve the existing "security-sweep → merge-gate" for bucket === "merge-now",
set "security-sweep → salvage" (or the project's salvage label) for bucket ===
"salvage-now", and otherwise set "security-sweep → review"; adjust the
nextAction assignment near the riskyFiles check (reference variables:
riskyFiles, bucket, nextAction, SCORE_SECURITY_ACTIONABLE).
- Around line 167-169: The --approved-only flag is ignored because you filter by
pr.reviewDecision before enrichment (pr.reviewDecision is only set during
enrichment) and you also hardcoded false where the parsed flag is used; move the
approved-only filtering to run after the enrichment step that populates
reviewDecision (i.e., after the enrich/enrichPullRequests/enrichPRs call that
processes non-draft PRs) and remove the hardcoded false so the code uses the
parsed approvedOnly variable from argument parsing (the flag parsed around Line
390). Ensure the post-enrichment filter uses prs.filter(pr => pr.reviewDecision
=== "APPROVED") so --approved-only works correctly.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In @.agents/skills/nemoclaw-maintainer-day/scripts/version-target.ts:
- Around line 58-76: The empty catch around JSON.parse of prOut (in the block
that populates stragglers by iterating prs and labels) swallows parse errors;
change it to surface failures by capturing the parse error and rethrowing or
returning a rejected Error so the maintainer loop fails fast (e.g., throw new
Error with context including prOut and the original error). Apply the same
change to the corresponding issueOut parsing block (the catch that currently
ignores parse errors when building issue-based stragglers) so both parse
failures no longer produce silent empty/partial straggler lists.
- Around line 65-66: The code currently treats any semver label not equal to
targetVersion as a "straggler" (using label.name !== targetVersion), which
incorrectly includes future release labels; change the logic in
version-target.ts to parse label.name and targetVersion into semver tuples
(major, minor, patch) and compare them numerically so only labels strictly older
than targetVersion are pushed into stragglers; update both occurrences (the
block that pushes into stragglers and the similar check at lines referenced
92-93) to use a helper compareSemver or inline tuple comparison rather than
string inequality, ensuring future-version labels are excluded and only
older-version labels are collected for bump-stragglers.ts to consume.
- Around line 111-116: The script reads the --repo value via parseStringArg but
still resolves tags from the local git checkout using run("git", ["fetch",
"origin", "--tags", "--prune"]) and getLatestTag(); update the logic so the repo
flag actually controls version resolution: if a non-default --repo is provided,
query that remote repository for tags/releases (e.g., use the GitHub CLI gh
release list or gh api to list tags) and derive latestTag and targetVersion
(bumpPatch) from that output; otherwise keep the existing local run("git",
["fetch", ..."]) and getLatestTag() path. Refer to parseStringArg, run,
getLatestTag, and bumpPatch to locate and branch the code, or alternatively
remove the --repo flag and its uses if you decide to restrict the script to the
local checkout.