The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool for read-only repository analysis, including commands like find, grep, wc, and ls. It explicitly prohibits any commands that could modify the audited repository, such as git or formatters, and limits its file-writing capabilities to a specific reports directory.- [DATA_EXFILTRATION]: The skill audits sensitive files like .env and settings.json but includes a mandatory redaction phase. This phase automatically scans for and masks AWS keys, private headers, and tokens using the report-format-enforcer reference, ensuring that no sensitive data is leaked in the final artifacts.- [PROMPT_INJECTION]: The skill uses a modular subagent architecture with clearly defined roles and boundaries. Its instructions focus strictly on the auditing task and do not contain any patterns aimed at bypassing safety filters or manipulating the underlying agent's core behavior.- [EXTERNAL_DOWNLOADS]: One external URL to the author's GitHub repository is included for metadata purposes. This reference is informational and does not trigger automated downloads, package installations, or remote script execution.

harness-audit