The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured and unstructured data from the protocols.io API, which may contain instructions embedded by external authors.\n
Ingestion points: Protocol metadata, step text, and discussion comments are ingested through scripts/protocols_read.py and processed via scripts/validate_protocol_json.py.\n
Boundary markers: The skill's instructions in SKILL.md and references/discussions.md explicitly mandate treating all remote content as untrusted data and provide guidelines for neutralizing links and mentions.\n
Capability inventory: The skill performs idempotent network GET requests and writes bounded JSON and PDF data to the local filesystem. It lacks capabilities for dynamic code evaluation or arbitrary shell command execution.\n
Sanitization: All remote data is processed through scripts/_common.py using sanitize_untrusted to redact sensitive keys and clean_text to remove control characters and enforce length limits.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates data retrieval from the official protocols.io domain.\n
Network access is restricted to HTTPS requests targeting www.protocols.io and validated organization subdomains on port 443.\n
The implementation uses a custom transport that explicitly disables HTTP redirects and proxy discovery to prevent credential leakage.\n
All network operations are gated behind an explicit --execute flag and capped by response size limits.\n- [COMMAND_EXECUTION]: The skill utilizes the Python interpreter to run its bundled helper scripts for data validation and request planning.\n
Script execution is limited to the skill's own source code using the standard library, with no invocation of external binaries or shell environments.

protocolsio-integration