show-me-your-work
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements a decision logging system using a TSV format. The provided
scripts/log.shscript employs secure practices by quoting variables to prevent shell injection and sanitizing data by replacing tabs and newlines. Additionally, it prevents spreadsheet formula injection by prefixing specific characters (=, +, -, @) with a single quote. - [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface where instructions to audit transcripts from
~/.claude/projects/could lead to the ingestion of potentially untrusted content from the project history. - Ingestion points: Transcripts located at
~/.claude/projects/<encoded-cwd>/(defined inSKILL.md) and external evidence pointers like PRs or commit SHAs. - Boundary markers: None explicitly defined for the transcript reading process.
- Capability inventory: File system write access via
scripts/log.shand shell command execution for data gathering. - Sanitization: The
log.shscript includes sanitization to prevent the log file from becoming an execution vector in spreadsheet applications.
Audit Metadata