Propose a patch

Evidence

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

diff --git a/references/video-prompting.md b/references/video-prompting.md
--- a/references/video-prompting.md
+++ b/references/video-prompting.md
@@ -127,1 +127,0 @@
-Generated by npm run sync:prompt-guides from @sogni-ai/sogni-intelligence-client 4.25.3; edit the rules there, not here.
@@ -182,1 +181,0 @@
-Generated by npm run sync:prompt-guides from @sogni-ai/sogni-intelligence-client 4.25.3; edit the rules there, not here.
@@ -976,1 +974,0 @@
-Generated by npm run sync:prompt-guides from @sogni-ai/sogni-intelligence-client 4.25.3; edit the rules there, not here.
diff --git a/references/image-prompting.md b/references/image-prompting.md
--- a/references/image-prompting.md
+++ b/references/image-prompting.md
@@ -66,1 +66,0 @@
-Generated by npm run sync:prompt-guides from @sogni-ai/sogni-intelligence-client 4.25.3; edit the rules there, not here.
diff --git a/references/seamless-tiling.md b/references/seamless-tiling.md
--- a/references/seamless-tiling.md
+++ b/references/seamless-tiling.md
@@ -78,1 +78,0 @@
-sogni-agent -m krea2_turbo_fp8_scaled -w 1024 -h 1024 -n 4 -o tile.png "<prompt>"
diff --git a/references/models.md b/references/models.md
--- a/references/models.md
+++ b/references/models.md
@@ -1373,1 +1373,0 @@
-sogni-agent --json --video -m minimax-h3-fasth3-i2v-turbo --ref first.png --keyframe turn.png@3 --keyframe wave.png@6 --duration 8 --estimate-video-cost "<prompt>"

Zero-context hunks: apply with git apply --unidiff-zero fix.patch on the tagged source of 3.65.1. 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: sogni-creative-agent-skill: HTML comment carrying imperative instructions to the agent (KillerSkills verdict caution 77)

## What the honeypot observed
KillerSkills ran **[email protected]** in an isolated sandbox (decoy credentials, recording proxy, syscall trace; engine 1.7) and scored it **CAUTION 77/100**. Summary: no network, touches nothing outside the project, writes no memory.

- See the findings below.

Full trace and findings: https://killerskills.ai/skills/clawhub/sogni-creative-agent-skill/3.65.1

## Findings with a location
- `references/video-prompting.md:127` · HTML comment carrying imperative instructions to the agent — references/video-prompting.md:127 · HTML comment carrying imperative instructions to the agent
- `references/image-prompting.md:66` · HTML comment carrying imperative instructions to the agent — references/image-prompting.md:66 · HTML comment carrying imperative instructions to the agent
- `references/video-prompting.md:976` · HTML comment carrying imperative instructions to the agent — references/video-prompting.md:976 · HTML comment carrying imperative instructions to the agent
- `references/video-prompting.md:182` · HTML comment carrying imperative instructions to the agent — references/video-prompting.md:182 · HTML comment carrying imperative instructions to the agent
- `references/seamless-tiling.md:78` · Fake instruction boundary tag — references/seamless-tiling.md:78 · Fake instruction boundary tag · informational: not an instruction file
- `references/models.md:1373` · Fake instruction boundary tag — references/models.md:1373 · Fake instruction boundary tag · informational: not an instruction file

## 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._