The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external API responses and local filesystem content to perform validation checks.\n
Ingestion points: The validation scripts in SKILL.md and references/launch-verification.md read and process responses from openrouter.ai (including model lists and usage data) and output from grep commands scanning local source code.\n
Boundary markers: Structural boundaries are established by parsing API data via jq and python3, though no natural language instructions specifically delimit the external data for the agent.\n
Capability inventory: The skill is granted access to Read, Write, Edit, Grep, Bash(python3), Bash(curl), and Bash(jq) in its SKILL.md frontmatter.\n
Sanitization: The skill relies on standard JSON parsing and regex pattern matching to extract data, which mitigates simple command injection via API responses.\n- [COMMAND_EXECUTION]: The skill utilizes bash to execute curl, jq, and grep. These tools are used for their intended diagnostic purposes: checking service health and ensuring no sensitive OpenRouter API keys (sk-or-v1-) are hardcoded in the codebase.\n- [EXTERNAL_DOWNLOADS]: The skill makes network requests to https://openrouter.ai, a well-known AI model aggregation service, to verify account status and model availability. These requests are legitimate interactions with the target service of the checklist.

openrouter-prod-checklist