The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content in multiple modes, creating a surface for indirect prompt injection.
Ingestion points: The Repo Value Analysis mode performs git clone on external repositories and dispatches agents to read every file. The Codebase Overview mode reads all files in a local project.
Boundary markers: The agent prompts defined in references/adr-consultation/agent-prompts.md and references/multi-persona-critique/personas.md do not include explicit delimiters or instructions to ignore embedded commands within the analyzed data.
Capability inventory: The skill has access to powerful tools including Bash, Task, Agent, Read, and Write across its various scripts and instruction sets.
Sanitization: There is no evidence of sanitization or escaping of the ingested file content before it is interpolated into the context of the dispatched analysis agents.
[EXTERNAL_DOWNLOADS]: The Repo Value Analysis mode uses git clone --depth 1 to download content from external Git repositories for the purpose of analysis. This is a primary function of the skill and targets standard development platforms.
[COMMAND_EXECUTION]: The skill utilizes several internal Python scripts (scripts/cartographer.py, scripts/cartographer_omni.py, scripts/cartographer_ultimate.py) to perform statistical analysis on Go codebases. These are executed via the Bash tool.
[DYNAMIC_EXECUTION]: The skill uses the Task and Agent tools to launch parallel sub-agents for deep-dive analysis and architectural critiques. While this follows the platform's orchestration patterns, it relies on the instructions and prompts provided in the reference files being executed correctly by the sub-agents.

assessment