The Agent Skills Directory

[PROMPT_INJECTION]:\n
Heuristic detectors identified override instructions (e.g., 'Ignore previous instructions') in the file tests/v3.2-evals.md.\n
Verification confirms these strings are deliberate test cases within an evaluation suite (e.g., Test R.2 and R.3) designed to ensure the skill's defensive modules are functioning correctly.\n- [INDIRECT_PROMPT_INJECTION]:\n
The skill processes untrusted external content such as transcripts, job descriptions, and recruiter emails.\n
It utilizes a centralized 'External Text Validation Module' that scans inputs for embedded directives and requires user confirmation before processing if suspicious text is detected.\n
It enforces a 'No Number Without A Source' rule to prevent the inclusion of fabricated quantitative data from external data sources.\n- [COMMAND_EXECUTION]:\n
Uses standard shell commands including date, grep, ls, cat, sed, and awk for legitimate local file manipulation and state management.\n
The skill mandates the use of Bash for all date computations to ensure interview schedule accuracy and prevent arithmetic errors.\n- [EXTERNAL_DOWNLOADS]:\n
Installation is performed via npx skills add dbhat93/job-search-os, a standard method for skill distribution.\n
The skill integrates with the Exa neural search tool and the Minutes open-source meeting recorder to gather company intelligence and capture interview data.\n- [DYNAMIC_CONTEXT_INJECTION]:\n
The SKILL.md file uses the !cat command to dynamically load instructions from COACH.md during activation, which is a standard implementation for maintaining modular instructions.

coach