agent-editing-safety
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFE
Full Analysis
- [CREDENTIALS_UNSAFE]: The guidance document (
references/current-guidance.md) explicitly instructs the agent to protect secrets and private data, forbidding the printing, copying, or summarization of credentials, including tokens, private keys, and auth cookies. - [COMMAND_EXECUTION]: The skill defines safe command execution practices, including the use of repo-provided wrappers, bounding long-running checks, and requiring approval for commands that mutate external state or git history.
- [INDIRECT_PROMPT_INJECTION]: The skill provides mitigations for processing external data by enforcing append-only updates to records, using dated sections, and requiring provenance links. It also instructs the agent to summarize facts only and exclude bot or system noise to maintain data integrity.
- [SAFE]: Verification procedures are established using standard search tools (
/usr/bin/grep,ripgrep) to ensure thorough sweeps for references before performing deletions or migrations.
Audit Metadata