The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external files (such as threat reports and CSV feeds), creating an ingestion surface where malicious actors could embed prompt injection strings inside indicators or reports.
Ingestion points: Found in scripts/agent.py (extract_iocs and ingest_ioc_feed functions) and scripts/process.py (--import-iocs option), which read unstructured text and CSV files.
Boundary markers: Absent; the extracted data is not wrapped in specific delimiters or data structures that instruct the underlying LLM to ignore potential instructions embedded within the text.
Capability inventory: There are no highly privileged capabilities (such as network requests, subprocess spawning, or dynamic execution) present in the scripts; they are confined to local file read/write and text parsing.
Sanitization: Absent; the scripts extract indicators via regular expressions or row reading without sanitizing or escaping content before it enters the agent's context.
[METADATA_POISONING]: A metadata inconsistency exists where the SKILL.md frontmatter defines the author as mahipal, whereas the LICENSE file attributes the copyright to mukul975.

performing-indicator-lifecycle-management