The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute local application commands and HTTP tasks as part of simulation recipes (e.g., firedrill simulate). The instructions explicitly note that commands run as ordinary local processes without a sandbox, advising users to review shared state and security boundaries.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of output and evidence from external target applications for evaluation and reporting purposes. It mitigates risk by instructing the operator to treat downloaded artifacts and reported values as untrusted data rather than instructions.
[EXTERNAL_DOWNLOADS]: The skill recommends installing the @firedrill-run/cli release. This follows the vendor's resource naming pattern and includes a proactive security warning to avoid the unscoped firedrill npm package to prevent dependency confusion attacks.
[CREDENTIALS_SAFE]: Instructions throughout the references emphasize keeping secrets (API keys, Firedrill tokens, and business logins) out of model context, logs, and shared recipes. The provided Python adapter example includes an assertion to verify that sensitive tokens are not leaked into the command environment.

firedrill