The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the Wilma school system, including message bodies, news bulletins, and downloaded document attachments (PDFs/docs).
Ingestion points: The workflow fetches summary data, reads messages/news, and downloads attachments from the Wilma API as described in Step 1 & 2 of the workflow.
Boundary markers: Absent. The skill does not define delimiters or specific instructions for the agent to ignore potentially malicious content within the messages or files.
Capability inventory: The skill uses the wilma CLI for fetching/downloading and the gog CLI for modifying Google Calendar events (write access). It also performs file-write operations to the local workspace.
Sanitization: None. There is no evidence of filtering or sanitizing the content of messages or attachments before they are processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill depends on external CLI tools installed from an unverified source.
Evidence: The skill requires wilma and gog tools to be installed via clawhub, which is not a standard platform registry or a well-known service.
[CREDENTIALS_UNSAFE]: The skill accesses local configuration files containing credentials for the Wilma school system and Google OAuth tokens.
Evidence: The skill frontmatter and setup instructions explicitly reference ~/.config/wilmai/config.json and ~/Library/Application Support/gogcli/ for accessing user account data.

wilma-triage