ktl-website

Pass

Audited by Gen Agent Trust Hub on Oct 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted Markdown data from the repository's README.md, docs/, and .lokf/knowledge/ directories (as seen in templates/website/scripts/sync.mjs). This data is processed and rendered into the final website using Astro components like ConceptCard.astro and ConceptPreviews.astro. While these components perform basic HTML entity escaping via an inline helper function, the skill lacks explicit boundary markers or warnings for the agent when processing this external content, creating a surface where malicious instructions in the repository could influence agent behavior.
  • [COMMAND_EXECUTION]: The skill's operational logic depends on the execution of several local CLI tools. scripts/website-preflight.sh runs shell environment checks; templates/website/scripts/sync.mjs executes uv run and bash commands to interact with the LOKF toolkit and sidecar scripts; and templates/website/scripts/config.mjs uses git rev-parse to retrieve repository metadata.
  • [DYNAMIC_EXECUTION]: The skill performs dynamic code and content generation at runtime. In Step 1 of SKILL.md, a Python one-liner is used to perform literal string substitution for project placeholders across the website/ directory. Furthermore, sync.mjs dynamically generates a significant number of Markdown source files and JSON metadata files in src/content/docs/ and src/generated/ based on the state of the host repository.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the download of external dependencies from public registries. SKILL.md and the GitHub Actions workflow templates/github/knowledge-website.yaml invoke npm ci and uv sync to install Node.js and Python-based packages required for building and running the site.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 18, 2026, 12:09 AM