The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run a local script, tools/cost-report.mjs, to process session tokens and generate usage reports. This is an intended function for analyzing orchestration costs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by reading session transcripts from ~/.claude/projects/.
Ingestion points: Transcript files (.jsonl) containing past conversation turns are read from the disk.
Boundary markers: The instructions specify filtering for assistant lines and extracting structured message.usage metadata, which reduces the risk of treating log content as instructions.
Capability inventory: The skill has the ability to execute local commands and read session files.
Sanitization: No explicit sanitization of the transcript text is mentioned, but the skill focus is on extracting numerical metadata for arithmetic calculation rather than processing the natural language content.
[DATA_EXPOSURE]: The skill accesses session logs in ~/.claude/projects/. While conversation history is technically sensitive data, accessing it is the primary and necessary purpose of this skill to provide usage metrics to the user.

measuring-orchestration-cost