dialectic

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime LLM context can include outsider-authored free text from public web content fetched at runtime: Phase 4 Stage B/C pulls Wikipedia extracts via curl (random donors + “Outline of academic disciplines” wikitext) and then decomposes that readable text into atomic parts that are fed into the orchestrator/LLM context for synthesis.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs running runtime fetches (curl/wget) to pull remote content that is then injected into the orchestration and prompt pipeline — e.g. Wikipedia API calls (https://en.wikipedia.org/w/api.php?action=query&list=random&rnnamespace=0&rnlimit=50&format=json and https://en.wikipedia.org/w/api.php?action=query&titles=ARTICLE_TITLE&prop=extracts&exintro=true&explaintext=true&format=json and https://en.wikipedia.org/w/api.php?action=parse&page=Outline_of_academic_disciplines&prop=wikitext&format=json) and fetching the skill file from GitHub (https://raw.githubusercontent.com/KyleAMathews/hegelian-dialectic-skill/refs/heads/main/SKILL.md), which are used at runtime to select donor domains and shape agent prompts — so these are runtime external dependencies that directly control prompts.