📦

Audit History

seo-new-keywords-research - 3 audits

Version comparison

Capability and finding changes across audited versions, newest first.

VersionDateResultReview itemsChange vs previous
v3 LatestOct 28, 2026, 12:35 PM No confirmed findings0No capability change
v2 Oct 27, 2026, 09:14 AM No confirmed findings7No capability change
v1 Oct 20, 2026, 02:49 PM 2 confirmed0Baseline

Oct 28, 2026, 12:35 PM

All 34 static findings are false positives involving Markdown syntax, legitimate CLI operations, provider links, and documented authentication. The skill requires paid lookup approval and explicitly treats retrieved instructions as untrusted data. No evidence of malicious behavior was found in SKILL.md; external package implementation and runtime behavior were not inspected.

1
Files scanned
73
Lines analyzed
4
Review items
0
False positives ignored
Audited by: codex

Oct 27, 2026, 09:14 AM

Most detections are Markdown formatting, documentation links, or keyword terminology rather than malicious commands. The workflow executes an external package with paid API credentials, but its implementation was not included in the review.

1
Files scanned
69
Lines analyzed
11
Review items
0
False positives ignored
Capability review items (7)

These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.

High
Hidden file in home directory
3. An AIsa key must be available, as `AISA_API_KEY` or from the AIsa CLI (`aisa login` stores it in
The instructions allow the external CLI to use a credential stored in ~/.aisa; the reviewed file does not establish how the package handles it.
High
Generic API/secret keys
3. An AIsa key must be available, as `AISA_API_KEY` or from the AIsa CLI (`aisa login` stores it in
The third-party CLI is configured with an API credential; its handling and network destinations cannot be verified from this file.
High
Generic API/secret keys
4. `GOANYAPI_API_KEY` is optional (a key comes from https://goanyapi.com/r/ga-95708898fb). With it t
The third-party CLI is configured with an API credential; its handling and network destinations cannot be verified from this file.
Medium
Ruby/shell backtick execution
1. Run the pinned release, never a clone of the repository: `uvx new-term-radar==0.6.0 <command> ...
The instructions execute a pinned third-party PyPI package; its implementation was not included in the reviewed files.
Medium
Ruby/shell backtick execution
```bash
The instructions execute a pinned third-party PyPI package; its implementation was not included in the reviewed files.
Medium
Ruby/shell backtick execution
```
The instructions execute a pinned third-party PyPI package; its implementation was not included in the reviewed files.
Medium
Hidden file access
3. An AIsa key must be available, as `AISA_API_KEY` or from the AIsa CLI (`aisa login` stores it in
The instructions allow the external CLI to use a credential stored in ~/.aisa; the reviewed file does not establish how the package handles it.
Audited by: codex

Oct 20, 2026, 02:49 PM

All 29 static findings reflect Markdown formatting, documented provider authentication, relevant links, or ordinary keyword analysis rather than the alleged threats. No evidence found of credential exfiltration or prompt injection in SKILL.md; the external CLI implementation was not provided for review. Separate workflow risks include executing an unpinned repository and initiating paid lookups after cost disclosure without explicit approval.

1
Files scanned
61
Lines analyzed
6
Review items
0
False positives ignored

Confirmed security concerns (2)

Medium
Unpinned External Repository Execution
Setup says to clone the repository 'if missing', then runs 'python -m radar keywords'. No reviewed commit or integrity check is required. This allows changing upstream code to execute with local credentials and filesystem access.
The instructions directly connect an unpinned clone with module execution. The external implementation is unavailable, so this identifies supply-chain exposure, not proven malicious code.
Medium
Paid Lookups Without Explicit Spending Approval
Every lookup is paid, but the workflow says 'State the estimate, then' run the command. It does not require approval or a budget ceiling. An agent following these instructions could spend existing account credit without a separate spending authorization.
The text explicitly describes paid calls and proceeds after disclosure, without an approval step. Existing user authorization could mitigate this workflow risk.
Audited by: codex