The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data from Cloudflare traffic logs, creating a surface for indirect prompt injection.
Ingestion points: Data from firewallEventsAdaptiveGroups (including userAgent, clientRequestPath, and clientRequestQuery) is ingested in steps/step-05-analyze-traffic.md and steps/step-07-extract-users.md.
Boundary markers: There are no explicit delimiters or instructions telling the agent to treat the log data as untrusted or to ignore instructions embedded within it.
Capability inventory: The skill has access to Bash, Read, Write, and Edit tools across its scripts.
Sanitization: Log data is incorporated directly into an incident report in steps/step-08-synthesize.md without filtering or escaping.
[EXTERNAL_DOWNLOADS]: The skill automatically installs the cloudflare-mcp-cli package globally via NPM if it is not present.
Evidence: steps/step-01-get-parameters.md contains the command npm install -g cloudflare-mcp-cli.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute cloudflare-mcp-cli with arguments inferred from user-supplied inputs (e.g., ZONE_ID, START_UTC).
Evidence: Multiple steps, including steps/step-02-confirm-spike.md and steps/step-04-identify-ja4.md, use the Bash tool to run CLI commands with parameters derived from the $ARGUMENTS input.

cloudflare-traffic-investigator