The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the uv run researchkit advise command to interface with multiple local AI CLI tools such as claude, codex, antigravity, and grok.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and displays verbatim responses from multiple external AI models, creating a surface where malicious instructions in a model's output could influence the agent's behavior.
Ingestion points: Verbatim responses retrieved from external CLI harnesses (Claude, Codex, Antigravity, Grok).
Boundary markers: The output is delimited by ## <harness-spec> headers for each member.
Capability inventory: Command execution via uv run and subprocess calls to external CLI tools.
Sanitization: No sanitization, filtering, or instruction-ignore wrappers are applied to the retrieved content before it is relayed to the agent context.
[EXTERNAL_DOWNLOADS]: The use of uv run to execute the researchkit tool may involve the automated download or update of unverified Python packages or dependencies from public registries.

advise