The Agent Skills Directory

[SAFE]: The skill provides a structured methodology for job offer negotiation within the Japanese market. It uses deterministic Python scripts to calculate financial differences and schedule feasibility based on user-provided JSON data.
[COMMAND_EXECUTION]: The skill uses scripts/plan_negotiation.py to process negotiation data. Analysis of this script and its shared library _common.py shows they exclusively use standard Python libraries (json, re, datetime, decimal) for logic and calculation. There are no calls to os.system, subprocess, or other dangerous execution functions.
[DATA_EXFILTRATION]: The skill explicitly states in its instructions and SKILL.md that it does not send messages to employers or recruiting agents automatically. All sensitive data (salary, current employer, etc.) remains within the agent's context for report generation, and the _common.py script even includes regex patterns to help identify and protect PII such as MyNumber, phone numbers, and emails.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests external data in JSON format, it lacks the high-risk capabilities (network access, shell execution with user input) necessary to exploit this surface. The instructions explicitly forbid the AI from fabricating information or making decisions on behalf of the user.

offer-negotiation