The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download code from external GitHub repositories, specifically https://github.com/open-clone/openclone for the core tool and https://github.com/akillness/oh-my-gods for plugin installation.
[REMOTE_CODE_EXECUTION]: The installation process involves a 'sparse checkout' of a remote repository followed by the execution of a ./setup script. This allows for the execution of arbitrary shell commands provided by the external repository on the host machine.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform installation steps, manage a local file-based database in ~/.openclone, and run the persona management commands.
[PROMPT_INJECTION]: The skill implements a 'knowledge injection' feature via the /openclone ingest command, which fetches data from external URLs and YouTube transcripts to populate persona context. This creates a surface for indirect prompt injection, where malicious instructions hidden in the ingested web content could influence the agent's behavior during subsequent chat sessions.
Ingestion points: SKILL.md (Instruction Step 4: Inject knowledge into a clone)
Boundary markers: The instructions do not specify any delimiters or safety headers to separate ingested 'knowledge' from the persona's core system instructions.
Capability inventory: The skill has access to powerful tools including Bash, WebFetch, Write, and Edit (listed in allowed-tools).
Sanitization: There is no mention of sanitization, filtering, or validation of the content retrieved from external URLs before it is stored in the local markdown files used for prompt context.

openclone