alterlab-tubitak-proposal

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes user-supplied proposal drafts via the scaffold_proposal.py --check command. An attacker could potentially embed malicious instructions within a draft file to influence the agent's behavior. However, this ingestion is central to the skill's primary document-checking function, and the processing is performed by a local Python script using string matching and regex-based parsing rather than direct prompt interpolation.
  • Ingestion points: Reads proposal draft files via the --check argument in scripts/scaffold_proposal.py.
  • Boundary markers: The script identifies sections using official Markdown heading structures as delimiters.
  • Capability inventory: File system access (Read, Write), network access (WebFetch), and local script execution (Bash).
  • Sanitization: The script strips Markdown comments and designated placeholders before performing word count analysis.
  • [EXTERNAL_DOWNLOADS]: The skill references and fetches content from official government and academic domains, specifically tubitak.gov.tr. These operations are performed to retrieve the most recent budget limits, application guidelines, and program forms. These downloads target authoritative sources and are necessary for the skill's stated purpose of ensuring proposal compliance.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 02:37 PM