The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted bond data from external files which is then rendered directly in the agent's chat context, creating an attack surface for instructions embedded in data.
Ingestion points: The script scripts/bond_search.py reads user-supplied data in various formats (JSON, CSV, TSV, HTML) via the --input argument or environment variables such as KSKILL_SEIBRO_EXPORT.
Boundary markers: Output is formatted using markdown tables in the chat render mode, but lacks explicit delimiters or instructions to the agent to ignore instructions potentially embedded in the processed data.
Capability inventory: The skill can read local files and execute specific scripts; it lacks network-write or file-write permissions.
Sanitization: While the script normalizes numeric, rate, and date fields, descriptive text fields (e.g., bond name, issuer, bond type) are passed to the chat context without specific markdown escaping or instruction-injection filtering.
[EXTERNAL_DOWNLOADS]: The skill uses npx to fetch and execute the vendor's CLI tool @nomadamas/k-skill from a public registry. This is a vendor-owned resource matching the author's identity and is used for instruction delivery and command execution.
[COMMAND_EXECUTION]: The instructions direct the agent to execute a local Python script scripts/bond_search.py and the @nomadamas/k-skill CLI to perform bond data lookup, normalization, and comparison.

korean-bond-search