peach-setup-harness
Warn
Audited by Gen Agent Trust Hub on May 12, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to delete specific directories and files (
.cursor/rules/,.cursorrules) using destructive shell commands during the cleanup process. It also recommends direct execution of database CLI tools (psql,mysql) if the specialized query skill is unavailable. - [DATA_EXFILTRATION]: The skill accesses sensitive local configuration files, specifically
api/env.local.yml, to extract database connection details and environment metadata. This exposure of potential credentials to the agent context is a documented data security concern. - [PROMPT_INJECTION]: An indirect prompt injection surface exists because the skill reads and processes the contents of
CLAUDE.mdandAGENTS.mdto guide its setup workflow without implementing security measures: - Ingestion points: Reads project-level documentation files in Step 1.
- Boundary markers: Absent; there are no instructions to delimit or ignore instructions contained within the read files.
- Capability inventory: Includes file system deletion (
rm -rf), file creation, and database query execution. - Sanitization: No validation or filtering is performed on the ingested content before it is used to determine subsequent agent actions.
Audit Metadata