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

Inline comments:
In `@nemoclaw/src/commands/config-show.ts`:
- Around line 32-35: The current rendering of credentialEnv in the config-show
command can echo unsafe or empty values; update the logic in the config-show
handler (where redactedCredential and lastFour are computed) to only render
`$NAME` when config.credentialEnv matches a strict env-var name regexp (e.g.
/^[A-Z0-9_]+$/); if it does not match or is empty, set redactedCredential to a
generic label like "(configured via env var)" or "(not configured)" and adjust
lastFour accordingly so raw tokens or malformed strings from loadOnboardConfig()
are never printed.

In `@nemoclaw/src/commands/shields-status.test.ts`:
- Around line 16-33: The blankState() test fixture returns an incomplete
NemoClawState missing the lastRebuildAt and lastRebuildBackupPath fields; update
blankState() to include these two properties (initialize them similarly to other
timestamp/backup fields—e.g., null) so the fixture matches the full
NemoClawState shape used in tests like slash.test.ts and avoids type/coverage
gaps when using lastRebuildAt or lastRebuildBackupPath in tests.

In `@src/lib/shields-timer.ts`:
- Around line 114-125: The catch block that calls
appendAudit("shields_up_failed", ...) currently lets execution fall into finally
which unconditionally calls process.exit(0), masking failures; modify
shields-timer.ts so the process exits with a non-zero status when an exception
occurs—e.g. set a local flag or call process.exit(1) from the catch (or set
process.exitCode = 1) and ensure the finally block uses that code or only
performs cleanup via cleanupMarker() without forcing exit(0); update the catch
handling around appendAudit(...) and the finally that calls cleanupMarker() /
process.exit(...) so errors produce a non-zero exit (use the function/variable
names appendAudit, cleanupMarker, and process.exit/process.exitCode to locate
changes).

In `@test/config-set.test.ts`:
- Around line 88-90: The test description string "accepts public http URLs" in
the it(...) block does not match the URL passed to validateUrlValue (currently
"https://example.com"); update the test so the description and input agree:
either change the description to "accepts public https URLs" or modify the URL
to "http://example.com" to match "http" — adjust the it(...) description or the
validateUrlValue argument accordingly.

---

Outside diff comments:
In `@nemoclaw-blueprint/policies/openclaw-sandbox-permissive.yaml`:
- Line 9: Update the stale inline comment describing Filesystem behavior so it
matches the current setting of include_workdir (which is false); locate the
comment that reads "Filesystem: include_workdir: true makes the sandbox home
directory writable" and change it to state that include_workdir: false makes the
sandbox home directory read-only (or otherwise describe the current behavior),
ensuring the comment and the include_workdir setting in
openclaw-sandbox-permissive.yaml are consistent.

In `@nemoclaw/src/commands/slash.test.ts`:
- Around line 64-81: The test fixture blankState() returns an incomplete
NemoClawState; add the missing fields lastRebuildAt and lastRebuildBackupPath to
the returned object so the shape matches the NemoClawState interface: include
lastRebuildAt: null (or ISO string when appropriate) and lastRebuildBackupPath:
null (or an empty string if your code expects a string) alongside the other
properties in the blankState() function to prevent type mismatches in tests.

---

Nitpick comments:
In `@agents/hermes/policy-permissive.yaml`:
- Around line 14-17: Update the comment that currently reads "Filesystem section
MUST match the default Hermes policy (policy-additions.yaml) exactly" to a
softer phrasing that conveys intent without implying absolute exactness—e.g.,
"Filesystem section should generally match the default Hermes policy
(policy-additions.yaml); avoid changing include_workdir or removing read_only
paths on live sandboxes." Ensure the new text also calls out the known exception
(/sandbox/.nemoclaw write path) so future editors aren't misled, and retain the
warning about using kubectl exec for config mutability and Landlock-related
restrictions.

In `@src/lib/sandbox-config.ts`:
- Around line 504-512: readStdin currently can hang indefinitely; update the
readStdin function to accept an optional timeoutMillis (or use a sensible
default) and implement a Promise.race between the stdin-reading promise and a
timeout promise; when the timeout fires, reject with a descriptive error, and
ensure you remove stdin listeners and clear any timers on both success and
timeout to avoid leaks (refer to the readStdin function name and its
"data"/"end"/"error" listeners when making these changes).

In `@test/e2e/test-shields-config.sh`:
- Around line 466-481: The auto-restore check using STATUS_AFTER_TIMER and the
command `nemoclaw "${SANDBOX_NAME}" shields status` can flake because it sleeps
a fixed 15s for a 10s timer; modify the test to avoid the race by replacing the
single sleep+check with a short retry loop: after the initial sleep, poll
`nemoclaw "${SANDBOX_NAME}" shields status` up to a configurable timeout (e.g.,
30s total) with small intervals and break when output contains "Shields: UP";
keep the existing cleanup fallback using `nemoclaw "${SANDBOX_NAME}" shields up
|| true` and ensure any failure still calls `fail "Auto-restore timer did not
restore shields within Xs"` with the total wait reflected.