The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to locate and inspect sensitive files such as "environment variables," "configuration files," "secrets," and "billing exports" to identify model usage and costs.
Evidence in SKILL.md: Workflows involve inspecting "billing exports" and "pricing configuration."
Evidence in references/model-inventory.md: Directs the agent to look for "environment variables" and "configuration files" to resolve runtime values.
[COMMAND_EXECUTION]: The reference materials suggest using shell tools to search the repository for model usage patterns.
Evidence in references/model-inventory.md: Instructs the agent to "Search with rg" to find provider SDK constructors, aliases, and logging implementations.
[EXTERNAL_DOWNLOADS]: The skill fetches live pricing data from external sources to perform its primary function.
Evidence in SKILL.md: Rule 5 mandates fetching current prices from official provider sources when pricing affects the results.
[PROMPT_INJECTION]: The skill processes untrusted content from the repository and external logs, creating a surface for indirect prompt injection.
Ingestion points: Source code, telemetry data, billing exports, and provider pricing pages.
Boundary markers: Includes a "Safety and evidence rules" section forbidding secret exposure, but lacks strict structural delimiters for data processing.
Capability inventory: File system read access, shell search (rg), and network access for pricing retrieval.
Sanitization: No explicit sanitization or validation logic is defined for processing search results or billing data.

ai-cost-audit