The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses behavioral constraints to enforce a 'telegraphic' and 'ultra-dense' communication style. These instructions are focused on performance optimization and token efficiency, rather than attempting to bypass safety filters or override system-level instructions.
[COMMAND_EXECUTION]: The skill specifies the use of grep for filtering log tracebacks and identifying code structures (AST-based inspection). These are routine file-system operations intended for navigating large files and reducing the amount of data loaded into the prompt context.
[DATA_EXFILTRATION]: No network operations, credential harvesting, or unauthorized data access patterns were detected. The instructions for 'JSON/YAML Crushing' and 'Log Compression' are localized data processing steps for token optimization.
[REMOTE_CODE_EXECUTION]: There are no patterns involving the download of external scripts or the execution of untrusted remote code. All code manipulation is performed via surgical local edits (str_replace).
[INDIRECT_PROMPT_INJECTION]: The skill defines a clear attack surface by ingesting and processing external data such as error logs, VCS output, and structured payloads. However, the logic specifically calls for 'pruning' and 'stripping' information, which acts as a rudimentary filter for external content.
Ingestion points: Processes logs, source files, JSON/YAML payloads, and tool results (SKILL.md).
Boundary markers: None explicitly defined, though minification/pruning is applied.
Capability inventory: Includes file reading (view_file), surgical editing (str_replace), and filtering (grep).
Sanitization: Relies on regex-based pruning to isolate specific technical content.

zipai-optimizer