The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/bootstrap.py script downloads the core runtime components (ditto.py and MINING_PROMPT.md) from the author's GitHub repository (github.com/ohad6k/ditto). The download process involves cryptographic verification, comparing the SHA-256 hash of the retrieved files against the values defined in the runtime.json metadata file to prevent tampering.
[REMOTE_CODE_EXECUTION]: The skill is designed to execute the downloaded ditto.py script via the local Python interpreter. This remote code execution pattern is secured through the use of specific version tags (e.g., v0.3.7) and the aforementioned integrity hashes, ensuring the agent only executes verified code from the vendor.
[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands to manage the setup and mining process, including python "$SKILL_DIR/scripts/bootstrap.py" and python "$DITTO_PY" plugin prepare. These commands are standard for the tool's stated purpose of environment setup and local data processing.
[PROMPT_INJECTION]: The skill processes user-authored .jsonl session history files. This introduces a surface for indirect prompt injection, as historical data could contain adversarial instructions. The skill attempts to mitigate this using 'contracts' defined in MINING_PROMPT.md to scope worker and reducer behavior.
Ingestion points: Local .jsonl session files.
Boundary markers: Uses worker and reducer contracts defined in MINING_PROMPT.md to guide model processing.
Capability inventory: Execution of Python subprocesses via ditto.py for mining and profile activation.
Sanitization: Integrity checks are performed on the tool itself; however, content-level sanitization of the processed history depends on the internal logic of the ditto.py runtime.

ditto