The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to autonomously execute shell commands defined by the user as 'Run command' and 'Measure command'. This is a core functionality required for quantitative experimentation and performance tuning.
[DYNAMIC_EXECUTION]: The skill dynamically generates and executes helper scripts within the .lab/bin/ directory (such as run, measure, and data_head). These scripts are used to post-process experiment results and manage data previews during the research lifecycle.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from project source files and experiment logs, which could potentially serve as an attack surface for indirect prompt injection.
Ingestion points: Reads project files within the defined scope, .lab/results.tsv, and .lab/log.md to inform future experiment hypotheses.
Boundary markers: The skill uses structured headings and labeled fields in its iteration logs, but does not employ specific 'ignore embedded instructions' markers when reading project source code.
Capability inventory: The agent possesses extensive capabilities including file system writes, git branch management and resets, spawning of subagent evaluators, and arbitrary command execution.
Sanitization: There is no explicit logic for sanitizing or validating external content before it is processed by the agent or passed to subagent evaluators.

researcher