tsmc-research-notes
Pass
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: Hardcoded local directory path.
- Evidence: The file
SKILL.mdspecifies a fixed local path for note storage: `/Users/kc/Obsidian Vaults/obsidian_kc_main/_AIpoint 臥龍/Project - TSMC/CuCMP AI 加藥建議 — 研究筆記/`.
- Impact: This exposes the author's local username (
kc) and internal directory structure. While consistent with the author's name (kcchien), hardcoding absolute paths is a security anti-pattern that can reveal information about the host environment. - [PROMPT_INJECTION]: Indirect prompt injection surface.
- Ingestion points: The skill instructions in
SKILL.mddirect the agent to read analysis sources such as notebooks, reports, and discussion records. - Boundary markers: No explicit delimiters or instructions are provided to the agent to distinguish between its instructions and the content of the processed files.
- Capability inventory: The skill allows the agent to write files to the local Obsidian vault.
- Sanitization: There is no evidence of sanitization or filtering of the ingested external content before it is incorporated into the notes.
- Risk: Maliciously crafted analysis files could include instructions that manipulate the agent's behavior during the knowledge extraction process.
Audit Metadata