The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the Bash tool to autonomously execute shell commands for measurement and version control management, such as running python tests/bench.py, git commit, and git revert. This is a core component of the intended autonomous improvement loop.\n- [DYNAMIC_EXECUTION]: The skill's protocol involves the agent programmatically modifying source files (Write/Edit tools) and then executing that code via benchmarks or test suites to verify metrics. This runtime modification and execution of code is an intended feature of the workflow.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the project environment while possessing significant execution capabilities. Ingestion points: Reads git log, results log (TSV), and project source files via the Read tool as described in the Core Protocol. Boundary markers: No explicit delimitation or instructions to ignore embedded prompts in the processed data are present. Capability inventory: Includes the Bash, Write, Edit, and Agent tools, allowing for file modification and command execution. Sanitization: There is no evidence of sanitization or validation of the ingested content before it is processed by the agent.

autoresearch