The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes an external validator (defaulting to c2patool) to perform cryptographic integrity checks on assets.
Evidence: The run_tool function in scripts/provenance_core.py uses subprocess.Popen to execute the validator.
Observation: This implementation is secure; it explicitly sets shell=False to prevent shell injection and utilizes a custom _drain function to enforce a strict memory ceiling on captured output, preventing resource exhaustion attacks from malformed tool responses.
[DATA_EXPOSURE]: The skill reads local files to analyze their metadata and structure.
Evidence: scripts/map_survival.py accepts and processes file paths for original and derivative assets.
Observation: This is a read-only operation required for the skill's primary function. The tool does not transmit asset content externally unless the optional --allow-network flag is explicitly enabled for remote manifest fetching.
[PROMPT_INJECTION]: The skill includes proactive defenses against indirect prompt injection by scanning incoming data for hidden instructions.
Evidence: scripts/provenance_core.py implements scan_covert_channels, which detects zero-width characters, Unicode tags, and homoglyphs in processed text.
Observation: These mitigations help prevent an agent from being influenced by hidden payloads embedded in the media files or text documents being analyzed.

map-provenance-survival