The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content such as CRM exports, notes, and emails which could contain malicious instructions.
Ingestion points: SKILL.md (Step 1 and Step 5) and references/champion-assets.md describe ingesting user-pasted text and uploaded files.
Boundary markers: The skill contains explicit instructions to treat all external text as data rather than instructions. It also mandates a 'Notice line' to alert the user if an instruction was detected in the input but ignored.
Capability inventory: The skill can execute a local Python script (scripts/concession_math.py) and perform web searches to verify facts provided by the user.
Sanitization: The instructions specifically prohibit following actions requested in external text and forbid sending data to addresses found within such text.
[COMMAND_EXECUTION]: The skill invokes a local Python script to perform financial calculations using user-provided data.
Evidence: SKILL.md (Step 2) instructs the agent to run python3 scripts/concession_math.py using deal figures provided by the user as arguments.
Sanitization: The script scripts/concession_math.py implements strict input validation using regular expressions to ensure inputs are numeric and valid money units before processing, preventing potential injection attempts through command-line arguments.

negotiation-and-champion