plan
Pass
Audited by Gen Agent Trust Hub on Aug 10, 2026
Risk Level: SAFEDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The skill implements a sophisticated secret redaction engine (
references/support/lib/redact-patterns.ts) that proactively identifies and masks credentials (e.g., AWS, GitHub, Stripe, OpenAI) before transmitting data to external review models like OpenAI Codex, effectively preventing accidental data exfiltration. - [EXTERNAL_DOWNLOADS]: The
references/support/runtime-bootstrap.mjsscript manages the acquisition of optional runtime components. It follows security best practices by restricting downloads to the author's official GitHub release hosts and enforcing SHA-256 checksum verification. - [PROMPT_INJECTION]: Instructions for handling user-supplied plans and specifications (e.g., in
spec.mdandautoplan.md) use defensive delimiters and clear instruction boundaries to mitigate the risk of indirect prompt injection during multi-agent workflows. - [COMMAND_EXECUTION]: Filesystem and repository operations are conducted using standard CLI tools (git, node, jq) within a controlled environment. The skill explicitly handles credentials via environment variables and secure input methods to prevent exposure in process lists.
Audit Metadata