Oathe Security Badge

oathe / scan / reports / safe

Is cathrynlavery/diagram-design safe?

Yes. cathrynlavery/diagram-design is safe to install. Oathe's behavioral security audit gave the diagram-design skill by cathrynlavery a trust score of 88/100 with 8 findings, none critical or high. Report updated

https://github.com/cathrynlavery/diagram-design

88
SAFE

Is cathrynlavery/diagram-design safe to install?

cathrynlavery/diagram-design is a well-engineered, MIT-licensed diagram generation skill with 44 visual types, comprehensive documentation including 14 ADRs, and mature CI/CD verification gates. The primary security consideration is that SKILL.md normalizes agent-executed Python scripts as part of its standard verification workflow; all such scripts are co-located in the repository and serve their stated purposes, but agents with shell access will invoke code execution without explicit per-run user consent. No malicious prompt injection, data exfiltration instructions, or canary file compromise was detected; the canary file accesses in the monitoring log pre-date the clone and originate from the audit framework's baseline setup.

What security issues were found in cathrynlavery/diagram-design?

Category Scores

Prompt Injection 87/100 · 30%
Data Exfiltration 90/100 · 25%
Code Execution 80/100 · 20%
Clone Behavior 97/100 · 10%
Canary Integrity 100/100 · 10%
Behavioral Reasoning 78/100 · 5%

Findings (8)

MEDIUM SKILL.md instructs agents to execute Python scripts -20 ▶

The SKILL.md §9 pre-output checklist explicitly directs agents with shell access to run multiple Python scripts as part of normal diagram verification: self_check.py for the accessible-SVG contract, verify-geometry.py for label geometry, and export_svg.py for SVG export. While all scripts are co-located in the repository and serve legitimate stated purposes, the pattern of a skill file instructing agents to execute code on the host filesystem is a meaningful code-execution surface regardless of script content.

Evidence: §9: 'From the installed skill directory, did python3 scripts/self_check.py <file> pass?' and 'From a repository checkout, verify with python3 <repo-root>/scripts/verify-geometry.py <file>'

LOW Export workflow depends on Playwright headless Chromium -5 ▶

The export feature requires Playwright with Chromium installed on the host. The doctor command checks for this dependency, and render-canonical-screenshots.py drives headless Chrome for PNG capture. A headless browser is a significant host capability that, while disclosed and scoped to export, represents a broader attack surface than a pure text-manipulation skill.

Evidence: references/export.md and doctor.md reference playwright install; render-canonical-screenshots.py uses Chrome headless with --screenshot flag

LOW Profile system reads files from user home directory -6 ▶

The skill's profile and marker resolution system reads from ~/.diagram-design/profiles/ outside the skill installation directory. The profiles.md reference owns the full resolution logic. While this is disclosed as a feature for multi-project branding, it introduces a path that could be influenced by the .diagram-design marker file in a project root.

Evidence: SKILL.md §0: 'Named profiles are full style-guide snapshots in ~/.diagram-design/profiles/' and marker-first resolution described in profiles.md

LOW References unvetted external skills in documentation -12 ▶

docs/superpowers/plans/2026-08-18-polar-chart.md instructs agentic workers to use superpowers:subagent-driven-development or superpowers:executing-plans as required sub-skills. These external skills are not included in this repository and have not been audited. Skill chaining through unvetted intermediaries can bypass per-skill security assessments.

Evidence: 'REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task'

LOW SKILL.md directs agent code execution as part of standard workflow -8 ▶

By embedding script-execution instructions in the §9 checklist, the SKILL.md normalizes shell invocation as a routine part of diagram generation. An agent following this skill will expand its tool usage to include code execution without explicit user request for each invocation. This is a pattern that, while scoped to verification, blurs the boundary between text-generation and system-level actions.

Evidence: Multiple §9 checklist items beginning 'From the installed skill directory, run python3 scripts/...' and 'From a repository checkout, run python3 <repo-root>/scripts/...'

INFO Generated HTML files include Google Fonts CDN link -4 ▶

SKILL.md §12 discloses that output HTML files embed external Google Fonts CSS via a CDN link. This means every diagram rendered and opened in a browser makes an outbound request to Google's servers, exposing the fact that the diagram was viewed. This is standard web practice and explicitly disclosed, but represents an ongoing information-disclosure channel in all skill outputs.

Evidence: SKILL.md §12: 'Embedded CSS (no external except Google Fonts)'; style-guide.md contains the Google Fonts <link> included in every generated HTML file

INFO Git clone attempted twice by audit runner -3 ▶

The process execution log shows the git clone command was issued twice (timestamps 1791799121.867 and 1791799134.005). The second attempt failed with 'destination path already exists'. This is an artifact of the audit infrastructure, not the skill, and has no security impact.

Evidence: Two EXECVE records for git clone with identical arguments; install output shows 'fatal: destination path already exists and is not an empty directory'

INFO Canary file accesses pre-date skill clone — audit framework artifact 0 ▶

inotifywait events show canary files (.env, id_rsa, .aws/credentials, etc.) were opened at 09:58:36, approximately 5 seconds before the git clone began at 09:58:41. These accesses are from the audit framework establishing its monitoring baseline, not from the skill. All canary files are confirmed intact with no content modifications.

Evidence: Canary access timestamps at epoch offset 1791799116.340; git clone exec at 1791799121.867; canary integrity check: ✅ All canary files intact

Should I install cathrynlavery/diagram-design?

Oathe's verdict for cathrynlavery/diagram-design is SAFE with a trust score of 88/100. Recommendation: Install.