The Agent Skills Directory

[SAFE]: The skill consists primarily of instructional text and prompt templates designed to improve agent autonomy and thoroughness. No malicious patterns such as obfuscation, credential theft, or unauthorized persistence were detected.
[COMMAND_EXECUTION]: Includes a Node.js one-liner used as a utility to verify that the generated goal objective stays under the 4,000-character limit. The script (node -e) reads from a local file (goal-objective.txt) and calculates its length. This is a benign development tool and does not involve network access or privilege escalation.
[SAFE]: Analysis of potential indirect prompt injection surfaces:
Ingestion points: The agent is instructed to read comprehensive goals from a local goal.md file.
Boundary markers: Absent; the agent is simply directed to follow the file contents.
Capability inventory: The instructions encourage the agent to explore, implement, and test (including using agent-browser), which implies file system and potential network access depending on the goal provided in goal.md.
Sanitization: No sanitization of the file content is performed. While this creates a surface for instructions to be pulled from a file, it is the primary intended function of the skill to facilitate delegation of broad tasks via structured documentation.

codex-goal-prompting