The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an autonomous researcher loop that processes output from a local script (run.log) and existing source files to guide its next actions, including code modifications.\n
Ingestion points: The agent is instructed to read run.log (the output of layer-eval.sh), configuration files (.star), and specification files to evaluate experiment results.\n
Boundary markers: No explicit instructions are provided to use delimiters or ignore potential instructions embedded within the read files or log output.\n
Capability inventory: The skill allows the agent to execute shell commands (layer-eval.sh) and modify the local git repository (git commit, git reset).\n
Sanitization: There is no evidence of sanitization or filtering of the ingested data before it is used to influence the agent's decision-making process.\n- [COMMAND_EXECUTION]: The skill requires the execution of multiple shell commands and local scripts to perform its optimization tasks.\n
Evidence: The instructions call for running scripts/autoresearch/layer-eval.sh and utilizing git for branch management, commits, and resets within the tdk-cli-core project context.

layer-autoresearch