The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external corpora, including vector databases, Elasticsearch, SQL databases, and filesystems via search tools like semantic_search and grep_search. This creates a surface for indirect prompt injection if the retrieved content contains malicious instructions.
Ingestion points: Data enters the context through tool result messages appended to the messages array in SKILL.md and references/tool-contracts.md.
Boundary markers: The documentation instructs developers to serialize results as JSON rather than prose to maintain data separation.
Capability inventory: The skill utilizes a Chat Completions loop capable of multi-step tool execution as shown in the implementation examples in SKILL.md.
Sanitization: The 'Limits' section in SKILL.md recommends clipping tool results before appending them to the context to prevent overflow and mitigate large-payload injections.
[DYNAMIC_EXECUTION]: The skill provides example code for a tool execution loop that dynamically resolves and calls functions based on model output.
Evidence: SKILL.md contains a loop using IMPLEMENTATIONS[call.function.name](**json.loads(call.function.arguments)) to execute tools. This is a standard pattern for agents but involves runtime resolution of executable paths.
[EXTERNAL_DOWNLOADS]: The documentation references the installation of official SDKs for the search model.
Evidence: references/mixedbread-tools.md includes instructions to install the mixedbread Python package and @mixedbread/sdk Node.js package, which are verified vendor resources.
[COMMAND_EXECUTION]: The skill includes documentation for setting up the environment and installing dependencies.
Evidence: SKILL.md and references/mixedbread-tools.md contain shell commands for export MXBAI_API_KEY, pip install, and npm install. These are standard configuration and installation steps.

mixedbread-search-agent