The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an auditing mechanism that reads and displays free-text content from a local JSON ledger, creating a surface for indirect instructions.
Ingestion points: The scripts/kill_ledger.py script parses the ledger.json file which contains multiple prose fields such as measured_cause, note, and failing_prong.
Boundary markers: The diagnostic output printed by the script does not utilize specific delimiters or boundary markers to differentiate user-provided data from agent instructions.
Capability inventory: The skill enables file system read access and the execution of a diagnostic Python script to audit the ledger.
Sanitization: Prose content within the ledger is displayed to the agent without specific sanitization or escaping mechanisms.

negative-result-taxonomy