The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The 'First-Time Installation' workflow directs the agent to execute an installation process using a GitHub link provided by the user. This creates a high-risk vector where an attacker could provide a malicious URL to execute arbitrary code on the host system.\n- [INDIRECT_PROMPT_INJECTION]: The skill regularly fetches and processes data from external Felo LiveDoc workspaces, such as project READMEs, task descriptions, and document content. This data is incorporated into the agent's context without sanitization or boundary markers, which could allow malicious instructions embedded in the workspace to influence the agent's actions.\n
Ingestion points: Untrusted content enters the agent's context through commands that fetch workspace data, specifically $SCRIPT get-readme, $SCRIPT resources, and $SCRIPT content as defined in SKILL.md.\n
Boundary markers: No explicit delimiters or instructions are used to separate external project data from system instructions or to warn the agent about potentially embedded commands.\n
Capability inventory: The agent possesses the ability to execute shell commands via a local Node.js script, read and write to sensitive local configuration files in the ~/.claude/ directory, and manage environment variables.\n
Sanitization: The skill lacks validation or filtering mechanisms to inspect the content retrieved from external workspaces before it is processed by the agent.\n- [COMMAND_EXECUTION]: The skill's operational workflows rely on executing a local Node.js script (felo-livedoc/scripts/run_livedoc.mjs) to manage project states, synchronize tasks, and interact with the vendor's API.

claw-deck