The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted HTTP request and response data from the Caido instance history and displays it to the agent without explicit boundary markers.
Ingestion points: Untrusted data enters the context via cmdSearch and cmdGet in lib/commands/requests.ts and via cmdReplay in lib/commands/replay.ts.
Boundary markers: The skill does not implement delimiters or system-level instructions to ignore embedded commands within the HTTP raw data.
Capability inventory: The skill has the capability to perform network operations (sending modified requests) and modify security tool configurations (scopes, environments) through the Caido SDK.
Sanitization: The skill decodes binary data and applies length-based truncation in lib/output.ts before display, which limits context window flooding but does not filter malicious instructions.

caido-mode