Skills ad-test-designer Audit History
📦

Audit History

ad-test-designer - 9 audits

Version comparison

Capability and finding changes across audited versions, newest first.

VersionDateResultReview itemsChange vs previous
v9 LatestJul 26, 2026, 09:54 AM No confirmed findings0No capability change
v8 Jul 26, 2026, 09:54 AM No confirmed findings0No capability change
v7 Jul 13, 2026, 11:58 AM No confirmed findings0No capability change
v6 Jul 13, 2026, 11:58 AM No confirmed findings0No capability change
v5 Jul 12, 2026, 10:42 AM 1 confirmed1No capability change
v4 Jul 6, 2026, 03:07 PM No confirmed findings0No capability change
v3 Jul 9, 2026, 12:06 PM No confirmed findings3No capability change
v2 Jul 6, 2026, 03:07 PM No confirmed findings0No capability change
v1 Jul 4, 2026, 03:52 PM 1 confirmed0Baseline

Jul 26, 2026, 09:54 AM

All 35 static alerts are false positives caused by Markdown backticks, relative documentation links, metadata, or the decimal alpha notation. The sole documented command runs a fixed local statistical helper and contains no evidence of network access, arbitrary code execution, data exfiltration, or prompt injection. The skill also explicitly treats exported CSV content as untrusted data and requires consent before saving results.

2
Files scanned
167
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude

Jul 26, 2026, 09:54 AM

All 35 static alerts are false positives caused by Markdown backticks, relative documentation links, metadata, or the decimal alpha notation. The sole documented command runs a fixed local statistical helper and contains no evidence of network access, arbitrary code execution, data exfiltration, or prompt injection. The skill also explicitly treats exported CSV content as untrusted data and requires consent before saving results.

2
Files scanned
167
Lines analyzed
3
Review items
0
False positives ignored
Audited by: claude

Jul 13, 2026, 11:58 AM

All 35 static alerts are false positives caused by Markdown syntax, fixed repository links, decimal notation, and homepage metadata. The command example invokes a fixed local statistical helper with numeric inputs; no injection, arbitrary traversal, credential use, or exfiltration intent appears.

2
Files scanned
167
Lines analyzed
3
Review items
0
False positives ignored
Audited by: codex

Jul 13, 2026, 11:58 AM

All 35 static alerts are false positives caused by Markdown syntax, fixed repository links, decimal notation, and homepage metadata. The command example invokes a fixed local statistical helper with numeric inputs; no injection, arbitrary traversal, credential use, or exfiltration intent appears.

2
Files scanned
167
Lines analyzed
3
Review items
0
False positives ignored
Audited by: codex

Jul 12, 2026, 10:42 AM

Most static alerts are false positives caused by Markdown code formatting and relative links to repository documentation. The skill also recommends an unaudited Python helper and can persist test summaries after consent, so those behaviors remain disclosed.

2
Files scanned
167
Lines analyzed
5
Review items
0
False positives ignored

Confirmed security concerns (1)

Low
Consent-Gated Persistence of Test Data
The skill can save campaign hypotheses, results, guardrails, ownership details, and approved actions to persistent memory after user confirmation.
Line 78 explicitly instructs a dated memory write and names the stored fields. The same line requires asking the user first, which limits the risk.
Capability review items (1)

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

Medium
Ruby/shell backtick execution
> **Statistical facts (keyless):** `python3 "${CLAUDE_PLUGIN_ROOT}/scripts/connectors/experiment.py"
The line explicitly recommends invoking Python on experiment.py through CLAUDE_PLUGIN_ROOT. That helper is outside the two audited files, so its behavior and argument handling cannot be verified here.
Audited by: codex

Jul 6, 2026, 03:07 PM

The command and network findings are false positives caused by Markdown formatting and repository metadata. The path traversal findings are fixed documentation links, and I found no command execution, data exfiltration, arbitrary file access, or prompt injection intent.

2
Files scanned
164
Lines analyzed
3
Review items
0
False positives ignored
Audited by: codex

Jul 9, 2026, 12:06 PM

Most static alerts are false positives from markdown code fences, relative repository links, and homepage metadata. I confirmed three medium-risk command-invocation instructions for a local experiment.py helper; no prompt injection, data exfiltration intent, or malicious network behavior was found.

2
Files scanned
166
Lines analyzed
6
Review items
0
False positives ignored
Capability review items (3)

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

Medium
Ruby/shell backtick execution
> **Significance (keyless — closes the design→measure loop):** once the variant results are in, `pyt
This line instructs the agent to run a local python3 experiment.py helper with user-derived count values. It appears intended, but it is real external command execution that needs strict argument validation.
Medium
Ruby/shell backtick execution
5. **Sample size, duration, power.** From the stated baseline and minimum detectable effect, size ea
This line tells the agent to use experiment.py samplesize when available. The helper is local and purpose-specific, but it is still intentional external command execution.
Medium
Ruby/shell backtick execution
- Apply **p<0.05 AND a minimum practical lift** (e.g. ≥ 10–15%, set at design time) — statistical si
This line again prefers experiment.py on the user exported counts. It is not malicious by itself, but it relies on an external helper with user-derived values.
Audited by: codex

Jul 6, 2026, 03:07 PM

The command and network findings are false positives caused by Markdown formatting and repository metadata. The path traversal findings are fixed documentation links, and I found no command execution, data exfiltration, arbitrary file access, or prompt injection intent.

2
Files scanned
164
Lines analyzed
3
Review items
0
False positives ignored
Audited by: codex

Jul 4, 2026, 03:52 PM

Static alerts are false positives from Markdown links, fenced examples, inline placeholders, and repository metadata; I found no shell execution or network call. One semantic issue remains: the optional memory filename includes a topic value without sanitization guidance, which could create unsafe paths in permissive hosts.

2
Files scanned
164
Lines analyzed
4
Review items
0
False positives ignored

Confirmed security concerns (1)

Medium
User-Controlled Memory Filename Lacks Sanitization Guidance
SKILL.md line 76 asks the agent to write a summary to memory/ad/ad-test-designer/YYYY-MM-DD-<topic>.md after consent, but it does not say to sanitize the topic. A crafted topic could influence the output path in hosts that follow the template literally.
The template includes a user-controlled topic inside a file path, but the final risk depends on the host filesystem sandbox and any filename normalization it applies.
Audited by: codex