The Agent Skills Directory

[DATA_EXFILTRATION]: The skill records application data in an unredacted format and exports it to several locations, including Vitest metadata, DevRunner API/CLI output, and a local extension socket.
Evidence: references/export-and-debugging.md explicitly states, "Naite does not redact, sample, or filter recorded values." It further details that NaiteReporter.reportTestResult() attempts "local extension-socket delivery whenever it runs."
Risk: Sensitive data like credentials, tokens, and personal information can be exposed if not manually masked by the user before recording, as warned in the documentation.
[INDIRECT_PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection because the agent is instructed to ingest and analyze traces generated by the code under test, which could include untrusted external input.
Ingestion Points: references/querying-traces.md describes methods such as Naite.get() and .result() used to fetch recorded data into the agent's context.
Capability Inventory: The agent uses this data to diagnose failures, check framework internals (Puri/UpsertBuilder), and verify output serialization.
Boundary Markers: There are no specific delimiters or instructions provided to the agent to treat trace values as untrusted content.
Sanitization: Documentation confirms that no redaction or filtering is performed on the ingested data.
[COMMAND_EXECUTION]: The skill instructions involve executing CLI commands to view test traces.
Evidence: references/export-and-debugging.md provides commands like pnpm sonamu test invoice.model --pattern "save" --traces for debugging.

sonamu-naite