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

Inline comments:
In `@bin/lib/onboard.js`:
- Around line 637-641: The pidsLimit environment value (NEMOCLAW_PIDS_LIMIT) is
injected directly into the shell command used by runCapture which allows command
injection; validate and sanitize it before use by parsing it as an integer and
falling back to a safe default if invalid, then pass that numeric value (e.g.,
parsedPidLimit) into the docker update invocation instead of the raw string;
update code references where pidsLimit is read and used (NEMOCLAW_PIDS_LIMIT,
pidsLimit, runCapture, and the docker update command that uses sandboxName) to
enforce an integer-only range (e.g., positive integer and reasonable upper
bound) and handle errors by logging and using the default.

In `@docs/security/process-limits.md`:
- Around line 62-64: Change the CLI code block to use the console language tag
and add a $ prompt: replace the ```bash fence containing
"NEMOCLAW_PIDS_LIMIT=1024 nemoclaw onboard" with a ```console fence and prefix
the command with "$ " so the block reads "$ NEMOCLAW_PIDS_LIMIT=1024 nemoclaw
onboard"; update the code fence start/end accordingly around the
"NEMOCLAW_PIDS_LIMIT" environment variable and "nemoclaw onboard" command.
- Around line 1-4: Add YAML frontmatter to the top of the page containing title,
description, keywords, topics, tags, content_type (or content type), difficulty,
audience, and status fields, then immediately after the frontmatter add the
required SPDX license header (e.g., "SPDX-License-Identifier: <license>");
ensure the frontmatter values are concise and accurate for "Process Limits" /
"Fork Bomb Protection" and include relevant keywords and topics so the page
meets the documentation guidelines.

In `@docs/security/sandbox-hardening.md`:
- Around line 1-6: The document lacks the required SPDX license header and
frontmatter and uses passive voice; add the standard SPDX header line at the top
and a YAML frontmatter block containing required fields (title, description,
author/date or other repo-required keys) before the content, and rewrite passive
sentences like "The NemoClaw sandbox image is hardened..." to active voice
(e.g., "NemoClaw hardens the sandbox image...")—update the title "Sandbox Image
Hardening" and the sentence mentioning "NemoClaw sandbox image" to reflect
active voice and ensure frontmatter values match those names.

---

Nitpick comments:
In `@docs/security/process-limits.md`:
- Around line 33-34: Split the two sentences in the paragraph so each sentence
occupies its own line in the source: move "The default process limit is 512 per
sandbox." and "This is sufficient for normal agent operation (typically < 50
processes) whilst preventing fork bombs." onto separate lines (i.e., one
sentence per line) to follow the documentation guideline for readable diffs.

In `@docs/security/sandbox-hardening.md`:
- Around line 41-43: The paragraph "Even if one layer is bypassed, the others
provide protection. Removing tools from the image means an attacker who escapes
the proxy still cannot compile custom tooling or open raw TCP connections."
should be split so each sentence is on its own line; edit the
docs/security/sandbox-hardening.md content replacing that single-line paragraph
with two lines, one containing "Even if one layer is bypassed, the others
provide protection." and the next containing "Removing tools from the image
means an attacker who escapes the proxy still cannot compile custom tooling or
open raw TCP connections." to satisfy the one-sentence-per-line guideline.

In `@test/process-limits.test.js`:
- Around line 66-69: Update the assertion in the test "uses NEMOCLAW_PIDS_LIMIT
env var with 512 default" to accept both logical OR and nullish coalescing
forms; modify the expect(onboardJs).toMatch(...) regex to allow either "||" or
"??" (with optional whitespace) before the default string "512" so it still
passes if code switches from || "512" to ?? "512".

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

Inline comments:
In `@Dockerfile`:
- Around line 31-35: The Dockerfile RUN step that does "apt-get update &&
apt-get purge -y --auto-remove ... && rm -rf /var/lib/apt/lists/* || true" is
masking failures; remove the trailing "|| true" so the build fails if apt-get
update/purge/cleanup fails, ensuring the purge of packages like gcc/g++/make and
netcat completes or the build stops; keep the existing chained commands and
error propagation semantics (no additional try/catch), so failures in that RUN
line fail the image build.

---

Duplicate comments:
In `@bin/lib/onboard.js`:
- Around line 1511-1515: pidsLimit is taken directly from process.env and
interpolated into the shell command, allowing command injection; change the
logic around the pidsLimit assignment so you parse and validate it as an integer
(e.g., const parsed = parseInt(process.env.NEMOCLAW_PIDS_LIMIT, 10)), ensure it
is a finite number within an acceptable range (fallback to 512 if
NaN/out-of-range), and then use that validated integer when calling runCapture
for the `docker update --pids-limit ${pidsLimit}` invocation so only a numeric
value is injected into the command.