Most static findings are Markdown fences, relative documentation links, placeholders, or SEO checklist text. Five findings are confirmed: four executable command recipes and one outbound-network use at SKILL.md lines 98-104; no prompt injection or malicious intent was found.
9
Files scanned
980
Lines analyzed
8
Review items
0
False positives ignored
Capability review items (5)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
**Zero-dependency local helpers** (no tool needed, run yourself): `python3 "${CLAUDE_PLUGIN_ROOT}/sc
Line 98 explicitly tells the agent to run local Python helpers with a user-supplied URL and includes a shell pipeline. This is legitimate SEO tooling, but command execution and unvalidated input create real execution risk.
**JS-rendering fallback (keyless)**: when `crawl.py`/`onpage.py` return an empty or thin body on a c
Line 100 directs the agent to execute a Firecrawl helper against a supplied URL and permits an own-site bypass. The command is relevant, but it performs active network retrieval through executable code.
Line 102 provides an executable curl recipe that interpolates a supplied domain and sends a request to crt.sh. Shell interpolation and outbound execution are genuine risks even though the audit purpose is legitimate.
**Index-push after fixes (write channel, gated)**: once crawl/indexing fixes ship, `python3 "${CLAUD
Line 104 directs execution of an index-push helper that can submit URLs to external search services when --live is used. The dry-run gate reduces risk, but the command has an intentional external side effect.
Line 102 intentionally sends audit queries to crt.sh and identifies the W3C validator as another external endpoint. These are relevant public services, but they disclose the queried domain or URL to third parties.
Most static findings are Markdown fences, relative documentation links, placeholders, or SEO checklist text. Five findings are confirmed: four executable command recipes and one outbound-network use at SKILL.md lines 98-104; no prompt injection or malicious intent was found.
9
Files scanned
980
Lines analyzed
8
Review items
0
False positives ignored
Capability review items (5)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
**Zero-dependency local helpers** (no tool needed, run yourself): `python3 "${CLAUDE_PLUGIN_ROOT}/sc
Line 98 explicitly tells the agent to run local Python helpers with a user-supplied URL and includes a shell pipeline. This is legitimate SEO tooling, but command execution and unvalidated input create real execution risk.
**JS-rendering fallback (keyless)**: when `crawl.py`/`onpage.py` return an empty or thin body on a c
Line 100 directs the agent to execute a Firecrawl helper against a supplied URL and permits an own-site bypass. The command is relevant, but it performs active network retrieval through executable code.
Line 102 provides an executable curl recipe that interpolates a supplied domain and sends a request to crt.sh. Shell interpolation and outbound execution are genuine risks even though the audit purpose is legitimate.
**Index-push after fixes (write channel, gated)**: once crawl/indexing fixes ship, `python3 "${CLAUD
Line 104 directs execution of an index-push helper that can submit URLs to external search services when --live is used. The dry-run gate reduces risk, but the command has an intentional external side effect.
Line 102 intentionally sends audit queries to crt.sh and identifies the W3C validator as another external endpoint. These are relevant public services, but they disclose the queried domain or URL to third parties.
Reviewed 61 static findings across the skill and reference files. The findings are markdown prompt examples, relative documentation links, SEO placeholders, or transparent helper recipes; no malicious execution, data exfiltration, or prompt injection evidence was found.
All 61 static findings are false positives caused by Markdown code fences, inline code, example URLs, relative documentation links, and SEO terminology. No evidence found of prompt injection, credential exfiltration, hidden executable code, or malicious network behavior in the reviewed files. The skill mentions optional audit helpers and write-channel index submission, but they are documented, user-directed, and gated by dry-run or live flags.