The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill facilitates the creation and execution of modified code snippets ('seeds') to test the recall of checking agents.\n
Evidence: The 'Protocol' section describes generating 'N seeded variants' and running checkers against them (SKILL.md).\n
Evidence: The worked example demonstrates the execution of a bash script (check-model-versions.sh) to verify a seeded defect.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external artifacts, creating a potential surface for indirect prompt injection.\n
Ingestion points: The skill reads artifacts such as manuscripts, code, and replication packages (SKILL.md, Protocol Step 2).\n
Boundary markers: The instructions recommend using a 'fresh context' and the Agent tool with context: fork to isolate the checking process (SKILL.md, Protocol Step 3).\n
Capability inventory: The skill utilizes the Read, Write, Bash, and Agent tools to modify and evaluate artifacts.\n
Sanitization: No explicit sanitization or filtering of artifact content is mentioned, relying instead on environment isolation.

vaccinate