The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements a 'Spike' process where it generates and executes code snippets to answer focused technical questions. These scripts are written to a temporary scratch directory (./scratch/spike-<name>/) and executed using the Bash tool. This represents runtime script generation and execution.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it retrieves untrusted data from the web using WebSearch and WebFetch tools. This external data, along with user-supplied hypotheses, is processed to synthesize research logs and technical essays. There are no explicit instructions for the agent to sanitize this external content or treat it as data distinct from instructions.
Ingestion points: WebSearch, WebFetch, and user input (SKILL.md)
Boundary markers: Absent; no explicit delimiters or 'ignore' instructions for fetched content.
Capability inventory: Bash (code execution), Write/Edit (filesystem), WebFetch (network) (SKILL.md)
Sanitization: Absent; no escaping or filtering mentioned.
[COMMAND_EXECUTION]: The skill uses the Bash tool for managing the file system, performing technical spikes, and executing git commands for library management as described in the research documentation.
[EXTERNAL_DOWNLOADS]: The research log documentation describes a workflow involving cloning a repository from the author's GitHub account (github.com/mrilikecoding/llm-orchestra-library.git). This is documented as a mechanism for contributing to a community library and is identified as a vendor-owned resource.

rdd-research