The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill includes a command to download and execute a shell script from a remote URL using a pipe to bash. \n
Evidence: curl -fsSL https://raw.githubusercontent.com/AnandChowdhary/continuous-claude/HEAD/install.sh | bash in SKILL.md. \n
This pattern allows for the execution of arbitrary code from an external source without user review. \n- [EXTERNAL_DOWNLOADS]: The skill references external tools and scripts hosted on a third-party GitHub account (AnandChowdhary) that are not from a verified organization. \n- [COMMAND_EXECUTION]: The skill provides several bash scripts designed to automate development workflows, including file system modifications, Git commits, and interacting with GitHub via the gh CLI. \n- [PROMPT_INJECTION]: The skill architecture facilitates indirect prompt injection by design through ingestion of untrusted files. \n
Ingestion points: The agent is instructed to read specification files (e.g., docs/auth-spec.md) and persistent note files (e.g., SHARED_TASK_NOTES.md) to guide its actions. \n
Boundary markers: There are no explicit instructions or delimiters to distinguish between task data and potential malicious instructions within these files. \n
Capability inventory: The agent has access to claude -p, bash, and Git tools, allowing it to modify the codebase based on ingested data. \n
Sanitization: No sanitization or validation of the external content is performed before it is processed by the agent.

autonomous-loops