The Agent Skills Directory

[SAFE]: The skill implements a 'Team Mode' orchestration pattern, delegating work to specialized roles such as Explorer, Executor, Reviewer, and ExpertAdvisor. This structure promotes task decomposition and independent verification.
[SAFE]: The skill includes diagnostic scripts (scripts/current_model.py, scripts/usage_by_model.py) that access local session traces located in ~/.codex/sessions. These scripts are used for performance and cost tracking, do not perform network operations, and exclusively use standard Python libraries for processing local JSONL data.
[SAFE]: References to external resources, such as model pricing and agent configuration documentation, target official OpenAI/ChatGPT domains and repositories, which are recognized as trusted services.
[INDIRECT_PROMPT_INJECTION]: The diagnostic scripts ingest data from previous session logs. However, this risk is minimal as the scripts only extract metadata (model names, token counts, timestamps) using structured JSON parsing and do not evaluate or execute the content of the dialogue.
Ingestion points: scripts/current_model.py and scripts/usage_by_model.py read local session trace files.
Boundary markers: The scripts utilize json.loads to parse structured event logs and do not interact with raw natural language content fields.
Capability inventory: The skill executes local Python scripts for data aggregation.
Sanitization: Metadata extraction is restricted to specific keys, ensuring that instructions within the logs are not interpreted by the scripts.

team-mode