The Agent Skills Directory

[COMMAND_EXECUTION]: The skill heavily relies on shell commands for searching and archive maintenance. In the search command, user-provided arguments are interpolated directly into rg (ripgrep) calls (e.g., rg -i "$ARGUMENTS"). Without explicit sanitization or escaping, this pattern could lead to command injection if the underlying agent platform's shell execution bridge is vulnerable to argument splitting.
[EXTERNAL_DOWNLOADS]: The ingest command enables the retrieval of content from arbitrary external URLs using WebFetch or curl -sL. This functionality is used to populate the local knowledge base with remote data.
[PROMPT_INJECTION]: The ingest and 'ripple pass' workflow represents an indirect prompt injection surface. The skill fetches untrusted data from URLs or files and uses it to update a persistent knowledge archive. Malicious instructions or false claims embedded in these sources could influence the agent's long-term knowledge base or trigger unintended 'ripple' updates to other notes.
Ingestion points: ingest <url-or-file> function.
Boundary markers: None specified for isolating ingested content during processing.
Capability inventory: Subprocess execution (rg, bun), file-system writes, and network operations (curl).
Sanitization: No sanitization or validation of the fetched content is mentioned before it is integrated into the archive.

Knowledge