The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent/user to download and execute a setup script from an external repository.
File: SKILL.md
Evidence: git clone https://github.com/23blocks-OS/ai-maestro-plugins.git && cd ai-maestro-plugins && ./install-graph-tools.sh.
[EXTERNAL_DOWNLOADS]: The skill fetches components and instructions from external GitHub repositories not verified as trusted.
File: SKILL.md
Evidence: References to https://github.com/23blocks-OS/ai-maestro-plugins.git and https://github.com/23blocks-OS/ai-maestro.
[COMMAND_EXECUTION]: The skill's primary functionality depends on executing multiple local shell scripts.
File: SKILL.md
Evidence: The skill defines and uses commands like graph-describe.sh, graph-find-callers.sh, graph-find-path.sh, and graph-index-delta.sh.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the user's codebase, which can contain hidden instructions that influence the agent's behavior.
Ingestion points: Source code files are indexed via the graph-index-delta.sh command in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded code comments or strings that might contain natural language instructions.
Capability inventory: The skill executes various shell scripts to query component relationships and call chains (defined in SKILL.md).
Sanitization: No sanitization, escaping, or validation of the data retrieved from the indexed codebase is mentioned or implemented in the instructions.

graph-query