The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill is designed to clone external Git repositories using the BASE_REPO parameter and execute code derived from them. It also generates new experiment scripts based on research plans and executes them via the /run-experiment command.
[COMMAND_EXECUTION]: The skill requests Bash(*) tool access and utilizes it to perform file system operations, repository cloning, and script execution. It also suggests using pip install to resolve missing dependencies during automated debugging phases.
[EXTERNAL_DOWNLOADS]: The workflow involves fetching external codebases through git clone and potentially downloading Python packages from public registries.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) due to its dependency on externally provided research documents.
Ingestion points: The skill reads EXPERIMENT_PLAN.md, FINAL_PROPOSAL.md, and other markdown files to determine experiment logic and parameters.
Boundary markers: There are no explicit instructions or delimiters used to prevent the agent from obeying malicious instructions embedded within the research plans.
Capability inventory: The agent has broad capabilities including arbitrary file writing (Write, Edit) and shell command execution (Bash, /run-experiment).
Sanitization: The skill does not describe any validation or sanitization logic for the content parsed from input files before it is used to generate executable code.

experiment-bridge