The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool and executes locally bundled Node.js scripts (located in references/scripts/) to manage internal state, transition between workflow phases, and record checkpoints. These operations are restricted to the project's runtime directory (.hex-skills/) and are consistent with the skill's operational purpose.
[REMOTE_CODE_EXECUTION]: All executable logic resides within the skill's own directory structure. There is no evidence of the skill downloading external scripts or piping remote content into a shell. The scripts use standard Node.js built-in modules (node:fs, node:path, node:crypto) to perform their tasks.
[DATA_EXFILTRATION]: Data interactions are strictly limited to local file system operations and authorized task tracking providers (Linear, GitHub, or local files) as defined in the configuration. No unauthorized network calls or hardcoded credentials were found.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from task descriptions and external documentation. However, it incorporates several safety layers to mitigate this surface, including a mandatory 'Goal Articulation Gate' to enforce objective reasoning and an 'Implementation Blueprint' phase that requires the agent to plan and document all changes before execution.
[PRIVILEGE_ESCALATION]: The skill does not use administrative commands or attempt to modify system-level configurations. It includes a dedicated reference for 'Destructive Operation Safety' which instructs the agent to seek user confirmation for high-risk operations like bulk file deletion or database schema changes.

ln-401-task-executor