workflow-onboard
Warn
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: MEDIUMCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill instructs the agent to read sensitive local environment files such as
.env.local. While the instructions include a guardrail advising the agent to only report variable names and never values, the agent's internal context will still be populated with the full content of these files, including any secrets or credentials they contain, which increases the risk of data exposure or exfiltration.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes a wide variety of untrusted data from the repository, creating an attack surface for indirect prompt injection.\n - Ingestion points:
SKILL.md(Steps 1, 2, and 3) directs the agent to read arbitrary files includingpackage.json,README.md, source code, SQL migrations, ORM schemas, and git history.\n - Boundary markers: Absent. There are no instructions for the agent to use delimiters or to disregard instructions embedded within the processed files to prevent them from influencing the agent's behavior.\n
- Capability inventory: The skill utilizes file system reading and shell command execution (
git log,git diff).\n - Sanitization: Absent. The skill does not mention sanitizing or escaping the content retrieved from the repository files before processing.
Audit Metadata