The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation explicitly recommends an installation command that pipes a remote script to the shell: curl -fsSL https://raw.githubusercontent.com/AnandChowdhary/continuous-claude/HEAD/install.sh | bash. This allows unverified remote code execution from a source outside of the trusted vendor list.
[EXTERNAL_DOWNLOADS]: The skill references and downloads implementation scripts from the AnandChowdhary/continuous-claude GitHub repository, which is an external source not associated with the primary vendor.
[COMMAND_EXECUTION]: The skill provides bash script templates that automate the execution of CLI tools like claude, gh, and npm. While functional, these patterns facilitate the automated execution of potentially compromised code paths.
[PROMPT_INJECTION]: The architecture described in the skill is highly susceptible to indirect prompt injection.
Ingestion points: The agent processes external files such as docs/auth-spec.md, specs/component-spec.md, SHARED_TASK_NOTES.md, and CI logs via gh run view.
Boundary markers: No explicit delimiters or boundary markers are utilized to separate instructions from untrusted data in these files.
Capability inventory: The skill loops possess extensive capabilities including file system modification (Read, Write, Edit) and shell command execution (Bash).
Sanitization: The instructions provide no evidence of sanitization or validation of the ingested external content before it influences the agent's behavior.

autonomous-loops