The Agent Skills Directory

[COMMAND_EXECUTION]: The script evidence-gate.mjs executes the local git binary using spawnSync to retrieve repository revisions and verify that the implementation source is clean. This is used strictly for integrity verification and artifact manifest binding.
[DATA_EXFILTRATION]: No data exfiltration patterns were identified. The skill does not perform network operations, and its file system access is restricted to reading project configuration and source files. The containedRegularFile function explicitly checks that paths do not escape the project root and rejects symlinks.
[INDIRECT_PROMPT_INJECTION]: The skill processes source code files which are untrusted data. It mitigates potential prompt injection by using a deterministic check boundary and validating all input manifests and hashes before analysis, ensuring that the agent's context is only influenced by relevant design tokens and code structure rather than arbitrary instructions.
[DYNAMIC_EXECUTION]: The skill performs runtime analysis of design tokens using local scripts. It ensures safety by using strict JSON parsing and schema validation for all reports and manifests, avoiding the use of dangerous functions like eval or exec on untrusted content.

ss-score