The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to perform web research (Step 2 in SKILL.md) to gather facts and figures for the slides. This represents an attack surface where malicious content on external websites could attempt to influence the agent's behavior via indirect prompt injection.\n
Ingestion points: SKILL.md (Step 2: 調べる
Investigation from Web).\n
Boundary markers: Absent in the instructions.\n
Capability inventory: The skill possesses significant local execution capabilities via Python and Node.js scripts.\n
Sanitization: No explicit sanitization or filtering of web content is mentioned before use in slide generation.\n- [DYNAMIC_EXECUTION]: The skill utilizes dynamic execution for calculations and layout analysis:\n
Math Calculations: SKILL.md (Step 5) instructs the agent to use python3 -c to calculate values for the slides. This is a form of dynamic code execution based on agent-generated strings.\n
CSS Evaluation: scripts/html_dump.mjs evaluates arithmetic expressions inside CSS calc() functions using the Function constructor. While it uses a basic character allowlist, it remains a dynamic evaluation point.\n- [COMMAND_EXECUTION]: The skill relies on executing multiple local scripts and tools via subprocess.run and spawn. This includes check_deck.py, check_layout.mjs, and the Chrome browser. These operations are intended for the skill's primary purpose of slide validation and rendering but represent an active local execution capability.

consulting-pptx-skill