The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to install the package using a remote command (npx ai-agent-skills install alirezarezvani/claude-skills/c-level-advisor), which downloads skill content and tools from an external GitHub repository.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a structured context file (company-context.md) that is generated from user responses during the onboarding process. This file is subsequently ingested as context for various advisory roles and Python analysis scripts.
Ingestion points: The company-context.md file is read by the agent to provide grounding for all executive roles (SKILL.md, .codex/instructions.md).
Boundary markers: The skill architecture includes a "Two-Layer Memory" system designed to separate raw transcripts from approved decisions, providing a measure of structural isolation.
Capability inventory: The suite includes 25 Python tools (e.g., burn_rate_calculator.py, tech_debt_analyzer.py) and a multi-agent deliberation framework (/cs:board).
Sanitization: There is no explicit documentation of sanitization or validation for the free-text data ingested during the onboarding questionnaire.
[METADATA_POISONING]: There are discrepancies in the documentation regarding the scope of the skill. The README.md describes a suite of 2 executive roles and 6 tools, whereas SKILL.md and CLAUDE.md describe a significantly larger ecosystem of 10 roles, 28 skills, and 25 Python tools. While likely a result of versioning or documentation updates, these inconsistencies can misrepresent the active capabilities of the skill.

c-level-advisor