The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to execute curl -fsSL https://raw.githubusercontent.com/AnandChowdhary/continuous-claude/HEAD/install.sh | bash. This pattern involves downloading an unverified shell script from a third-party GitHub repository and piping it directly into a bash shell, which allows for arbitrary code execution with the user's current privileges.
[EXTERNAL_DOWNLOADS]: The skill references and downloads scripts from an external GitHub account (AnandChowdhary) which is not a verified organization or a recognized trusted vendor for this platform. This introduces supply chain risk as the remote script's integrity is not verified before execution.
[COMMAND_EXECUTION]: The patterns described in the skill rely heavily on shell automation and direct execution of CLI tools like claude -p and GitHub CLI (gh). While functional for the task, this creates a high-privilege environment where an autonomous agent can perform destructive repository actions or expose metadata if compromised by malicious input.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its ingestion of external data.
Ingestion points: Files like specs/component-spec.md, docs/auth-spec.md, and general RFC/PRD documents are read and processed as instructions for agent sub-tasks (SKILL.md).
Boundary markers: The provided patterns do not include explicit boundary markers or system instructions to help the agent distinguish between the developer's core instructions and potentially malicious commands contained within those external files.
Capability inventory: The agent is granted extensive capabilities, including the ability to write code, execute bash commands, create branches, and push to remote repositories (SKILL.md).
Sanitization: No validation or sanitization mechanisms are described for the content of the specification files before they are interpolated into the agent's context.

autonomous-loops