practice-english-writing
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input, including chat messages, PR descriptions, and commit messages, to provide corrections and lessons. This ingestion of external data creates a potential surface for instructions embedded in the text to influence agent behavior.
- Ingestion points: User-provided text passed through commands or conversational triggers (e.g., "/practice-english-writing", commit messages, or chat replies) as described in the frontmatter and Step 1.
- Boundary markers: The skill defines 'protected spans' including text inside backticks, fenced code blocks, and quotes, which are intended to be copied through without modification (Step 2).
- Capability inventory: The skill writes user mistake profiles to the local filesystem at
~/.claude/practice-english-writing/profile.md(Step 6). - Sanitization: There is no explicit sanitization or filtering mentioned for the user-supplied content before it is stored or processed for corrections.
- [PERSISTENCE]: The skill maintains a persistent user profile across sessions to track error frequency, sample mistakes, and progress signals.
- Evidence: It creates and updates a Markdown file at
~/.claude/practice-english-writing/profile.mdto store aggregate error counts, sample before/after pairs, and register usage history (Step 6).
Audit Metadata