The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The db_query tool defined in references/tools.md includes a connection_string parameter. Connection strings commonly contain sensitive credentials (username/password), posing a risk of exposure if hardcoded in scripts or captured in execution logs.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection as it processes data from untrusted external sources.
Ingestion points: Tools like aiohttp_race (returns bodies), state_checker (polls endpoints), and db_query (fetches state) ingest data from external targets into the agent context.
Boundary markers: None identified; the skill does not specify delimiters or instructions to ignore embedded commands within the fetched data.
Capability inventory: The skill has capabilities for parallel network requests (turbo_intruder_race, aiohttp_race), shell command execution via curl wrappers (gnu_parallel_curl), and database interactions (db_query).
Sanitization: There is no mention of sanitization, validation, or escaping of the ingested content before it is processed by the agent.
[COMMAND_EXECUTION]: Several tools such as gnu_parallel_curl, xargs_curl, and race_the_web described in references/tools.md appear to be wrappers for shell-based utilities. If user-supplied or externally-sourced inputs (like headers or data) are interpolated into these commands without rigorous sanitization, it could lead to command injection.
[DATA_EXFILTRATION]: The skill utilizes tools that handle sensitive session identifiers (cookies and authorization headers) during parallel request execution. While intended for testing, this pattern involves the handling and transmission of sensitive authentication data across the network.

pentest-race-conditions