The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Bash to manage a local calibration log at .claude/token_calibration/log.tsv. It uses mkdir -p to create the directory and printf to append tab-separated values (task type, estimated cost, actual cost, multiplier, and date) to the log file. This is a standard and safe implementation for local data persistence.
[DATA_EXPOSURE_AND_EXFILTRATION]: While the skill reads and writes to a local .claude subdirectory, this is a common practice for agent configuration and state. There are no network operations or attempts to exfiltrate this data to external servers.
[DYNAMIC_EXECUTION]: The skill calculates a 'Final formula' for token estimation (base × file_multiplier × agent_multiplier × iteration_multiplier). This calculation is performed by the LLM based on instructions and does not involve the execution of dynamically generated code or unsafe deserialization.

token-budget-gate