📦

Audit History

embedding-strategies - 9 audits

Version comparison

Capability and finding changes across audited versions, newest first.

VersionDateResultReview itemsChange vs previous
v9 LatestJul 7, 2026, 07:47 AM No confirmed findings0No capability change
v8 Jul 7, 2026, 07:47 AM No confirmed findings0External commands
v7 Jul 1, 2026, 12:25 AM No confirmed findings2No capability change
v6 Jul 1, 2026, 12:25 AM No confirmed findings2Network access
v5 Jan 21, 2026, 07:19 PM No confirmed findings0 Network accessExternal commands
v4 Jan 17, 2026, 08:42 AM No confirmed findings0No capability change
v3 Jan 17, 2026, 08:42 AM No confirmed findings0Network accessExternal commands
v2 Jan 4, 2026, 05:13 PM No confirmed findings0No capability change
v1 Jan 4, 2026, 05:13 PM No confirmed findings0Baseline

Jul 7, 2026, 07:47 AM

All static detections are false positives caused by Markdown code fences, documentation links, and ordinary variable text. No prompt injection attempt, covert exfiltration intent, command execution, or system reconnaissance behavior was found in SKILL.md.

1
Files scanned
480
Lines analyzed
2
Review items
0
False positives ignored
Audited by: codex

Jul 7, 2026, 07:47 AM

All static detections are false positives caused by Markdown code fences, documentation links, and ordinary variable text. No prompt injection attempt, covert exfiltration intent, command execution, or system reconnaissance behavior was found in SKILL.md.

1
Files scanned
480
Lines analyzed
2
Review items
0
False positives ignored
Audited by: codex

Jul 1, 2026, 12:25 AM

Static analysis flagged markdown code fences as Ruby shell backticks, but these are documentation delimiters and not executable shell commands. The hardcoded URLs are reference links to vendor documentation and benchmarks, while sample embedding code may contact model providers only if a user runs it with their own environment.

1
Files scanned
480
Lines analyzed
3
Review items
1
False positives ignored
Capability review items (2)

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

Low
Runnable Examples May Contact External Embedding Services
The OpenAI embedding example calls client.embeddings.create, and local sentence-transformer examples can download named models when run. This is expected for an embedding strategy guide, but users should review provider terms and avoid sending sensitive text without approval.
The cited lines contain direct model or API calls that can use network access when executed. The behavior is clear, but it is presented as user-run example code rather than hidden execution.
Low
External Documentation Links
The resources section links to OpenAI, Sentence Transformers, and the MTEB leaderboard. These are normal reference links and do not execute automatically, so the static network findings are low risk.
The URLs appear only as markdown reference links in a resources section. I found no evidence that the skill fetches these URLs automatically.
Static false positives ignored (1)

These static matches were dismissed by semantic review or matched schema-only tokens, so they are shown for transparency but do not drive the quality score.

Low
False Positive Shell Backtick Detection
Static analysis reported shell backtick execution at multiple locations, but the marked text is markdown code fencing around Python examples and diagrams. I found no shell command execution, no subprocess use, and no instruction to run hidden commands.
The exact locations are markdown fences surrounding non-shell content. Semantic review found documentation examples, not executable backtick interpolation or shell invocation.
Audited by: codex

Jul 1, 2026, 12:25 AM

Static analysis flagged markdown code fences as Ruby shell backticks, but these are documentation delimiters and not executable shell commands. The hardcoded URLs are reference links to vendor documentation and benchmarks, while sample embedding code may contact model providers only if a user runs it with their own environment.

1
Files scanned
480
Lines analyzed
3
Review items
1
False positives ignored
Capability review items (2)

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

Low
Runnable Examples May Contact External Embedding Services
The OpenAI embedding example calls client.embeddings.create, and local sentence-transformer examples can download named models when run. This is expected for an embedding strategy guide, but users should review provider terms and avoid sending sensitive text without approval.
The cited lines contain direct model or API calls that can use network access when executed. The behavior is clear, but it is presented as user-run example code rather than hidden execution.
Low
External Documentation Links
The resources section links to OpenAI, Sentence Transformers, and the MTEB leaderboard. These are normal reference links and do not execute automatically, so the static network findings are low risk.
The URLs appear only as markdown reference links in a resources section. I found no evidence that the skill fetches these URLs automatically.
Static false positives ignored (1)

These static matches were dismissed by semantic review or matched schema-only tokens, so they are shown for transparency but do not drive the quality score.

Low
False Positive Shell Backtick Detection
Static analysis reported shell backtick execution at multiple locations, but the marked text is markdown code fencing around Python examples and diagrams. I found no shell command execution, no subprocess use, and no instruction to run hidden commands.
The exact locations are markdown fences surrounding non-shell content. Semantic review found documentation examples, not executable backtick interpolation or shell invocation.
Audited by: codex

Jan 21, 2026, 07:19 PM

All static findings are false positives. C2 keyword alerts triggered by hash hex strings. Weak crypto alerts from hash substrings. External command alerts from ASCII flow diagrams using arrows. Hardcoded URL alerts are legitimate documentation links. No malicious code, command execution, or data exfiltration patterns found.

2
Files scanned
818
Lines analyzed
0
Review items
0
False positives ignored
No confirmed security findings were recorded for this completed audit.
Audited by: claude

Jan 17, 2026, 08:42 AM

AI analysis failed after multiple attempts - MANUAL REVIEW REQUIRED before publishing. This skill cannot be auto-published until reviewed by a human.

2
Files scanned
656
Lines analyzed
2
Review items
0
False positives ignored

Detected Patterns

Hardcoded URLWeak cryptographic algorithmRuby/shell backtick executionSystem reconnaissance
Audited by: claude

Jan 17, 2026, 08:42 AM

AI analysis failed after multiple attempts - MANUAL REVIEW REQUIRED before publishing. This skill cannot be auto-published until reviewed by a human.

2
Files scanned
656
Lines analyzed
2
Review items
0
False positives ignored

Detected Patterns

Hardcoded URLWeak cryptographic algorithmRuby/shell backtick executionSystem reconnaissance
Audited by: claude

Jan 4, 2026, 05:13 PM

Pure documentation skill with code templates. No network, file system, or credential access. Contains only educational content about embedding models and strategies.

4
Files scanned
480
Lines analyzed
0
Review items
0
False positives ignored
No confirmed security findings were recorded for this completed audit.
Audited by: claude

Jan 4, 2026, 05:13 PM

Pure documentation skill with code templates. No network, file system, or credential access. Contains only educational content about embedding models and strategies.

4
Files scanned
480
Lines analyzed
0
Review items
0
False positives ignored
No confirmed security findings were recorded for this completed audit.
Audited by: claude