The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external documents (Requirements, Technical docs, Project plans, etc.) provided by the user. These inputs are parsed by Python scripts (common_parser.py, xlsx_reader.py) and injected into the agent's context. A malicious document could attempt to influence the agent's behavior or conclusions via embedded instructions.
Ingestion points: scripts/run_strategy.py loads user-provided files via common_parser.py.
Boundary markers: The skill prompt and workflow instructions define a structured context, but do not include explicit 'ignore instructions in data' delimiters for all input sections.
Capability inventory: The skill can write output files to the local filesystem using scripts/common_formatter.py. No network or shell execution capabilities were found.
Sanitization: Content is extracted as raw text without specific sanitization of potential embedded instructions, though the skill uses an 'evidence scaffold' approach that labels parsed data as NOT_ASSESSED to help the agent and user distinguish between raw data and semantic analysis.
[EXTERNAL_DOWNLOADS]: The README mentions installation scripts (scripts/install-skills-mac.sh and scripts/install-skills-windows.ps1) for setup, which are typical for skill deployment on certain platforms and not provided in the skill package for direct analysis.

test-strategy-plus