The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system-level shell commands to manage database state. The fixture init command shells out to mysqldump and mysql. The fixture sync command utilizes pg_dump and issues DROP DATABASE commands to reset the test environment.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted or user-defined metadata that is passed to an LLM to influence data generation.\n
Ingestion points: Metadata is ingested from cone.note fields in SKILL.md and referenced files like references/cone.md.\n
Boundary markers: There are no documented boundary markers used to separate notes from instructions in the prompt sent to the LLM.\n
Capability inventory: The skill has capabilities to execute shell commands (mysqldump, pg_dump), perform destructive database writes (DROP DATABASE), and write files to the file system.\n
Sanitization: The skill claims to parse fixtureGenerator expressions rather than using eval, which reduces the risk of direct code injection through these fields.\n- [DATA_EXFILTRATION]: The skill contains commands like fixture fetch and fixture import that move data from production or current environments into fixture databases. While this is the intended purpose, it represents a data movement path that could be used to expose sensitive information if target databases are misconfigured.

sonamu-fixture