The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest research hypotheses and benchmark metrics, which may originate from untrusted external sources or user-provided data.
Ingestion points: The workflow processes experimental data recorded in session files such as autoresearch.md and autoresearch.jsonl.
Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from treating data within these files as instructions.
Capability inventory: The skill uses shell tooling and specialized experiment tools (init_experiment, run_experiment) to execute the research loop.
Sanitization: No sanitization or validation of the input hypotheses or benchmark data is specified.
[DYNAMIC_EXECUTION]: The skill involves the generation and subsequent execution of shell scripts as part of the research benchmarking process.
Evidence: The instructions explicitly mention the use of a session file named autoresearch.sh for running benchmarks through the available shell/tooling.

autoresearch