The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to execute shell commands for verifying iterations and shipping artifacts. It implements a mandatory safety screen that checks for high-risk patterns such as arbitrary file deletion or fetch-and-execute remote scripts before running any dry-runs or verification steps.
[DATA_EXFILTRATION]: The skill performs legitimate network operations via the GitHub CLI and email service APIs (e.g., SendGrid) as part of the shipping and promotional workflows. The promotion feature (starring a repository) is explicitly gated by a request for user consent.
[PROMPT_INJECTION]: A static analysis hint in references/predict-workflow.md was identified as a false positive. The strings flagged as potential injection attempts are actually part of a security deny-list defined within the skill's instructions to protect the agent from malicious external content.
[DATA_EXPOSURE]: The skill contains specific protocols for handling sensitive data, including a requirement to mask credentials (e.g., API keys, passwords, and JWTs) in findings, proofs of concept, and logs. It uses standard placeholders like <REDACTED_TOKEN> to ensure secrets are not stored in the iteration history.
[DYNAMIC_EXECUTION]: The skill generates and modifies code files as part of its iterative loop. This behavior is the intended purpose of the 'autoresearch' functionality and is monitored by verification and guard commands to ensure correctness and safety.

autoresearch