The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data in the form of knowledge candidates that are processed by an LLM for semantic verification, representing a surface for indirect prompt injection. \n
Ingestion points: Knowledge candidate objects provided in the input array to the stage B-3 step. \n
Boundary markers: No specific boundary markers or 'ignore' instructions are used to separate candidate text from the verification prompt. \n
Capability inventory: The skill possesses the ability to read local files via Read, Grep, and Glob, and execute vault-related commands via the knowledge-gate CLI. \n
Sanitization: The skill does not perform sanitization of candidate content before semantic evaluation. \n- [COMMAND_EXECUTION]: The skill executes subcommands of the knowledge-gate CLI tool for querying the knowledge vault and resolving file paths. These operations are essential to its function as a knowledge validator and are restricted to specific, non-arbitrary subcommands. \n- [DATA_EXFILTRATION]: The skill accesses repository artifacts using Read, Grep, and Glob to verify claim derivability (Rule R7). While this involves reading local file content based on candidate-provided domains and evidence, it is constrained to the repository environment and intended verification scope.

quality-gate