The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies heavily on executing local binaries such as fift, tolk, lite-client, and validator-engine to manage the local test network and process blockchain artifacts. Scripts including run_liteclient.py, wallet_send.py, and run_fift_script.py use subprocess.run to invoke these tools based on CLI arguments.
[DYNAMIC_EXECUTION]: The skill performs dynamic loading and execution of code from the provided repository root. In scripts/ton_triage_lib.py, the add_tontester_to_syspath function modifies sys.path to include source directories from the --repo-root argument, and ensure_tonapi executes a Python script (generate_tl.py) found within that repository. Additionally, the skill provides wrappers like run_fift_script.py and compile_tolk.py to execute Fift and Tolk scripts supplied by the user.
[INDIRECT_PROMPT_INJECTION]: Several scripts, including summarize_run.py and inspect_latest_transaction.py, ingest and process external data such as node logs, account states, and transaction bodies. This creates a surface for indirect prompt injection where malicious content embedded in the processed blockchain data could attempt to influence the agent's follow-up analysis or actions.
[DATA_EXPOSURE]: The skill manages sensitive artifacts such as private keys (.pk files) and wallet environment configurations (wallet-env.txt) within its working directory. While standard for blockchain development tools, the handling of these files requires careful management of the skill's output directory.

ton-bug-triage