The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted files from manufacturing codebases (Phases 1-8).
Ingestion points: The agent is instructed to read various configuration files (e.g., package.json, requirements.txt, pyproject.toml) and search the entire codebase for sensor data connectors, ML model logic, and reliability calculations.
Boundary markers: The instructions do not provide any boundary markers or "ignore embedded instructions" warnings for the agent when processing the content of the files it reads.
Capability inventory: The agent uses its file-reading capabilities to extract information from the codebase to generate a detailed report.
Sanitization: There are no instructions to sanitize, escape, or validate the content of the files being audited before they are processed by the agent.
[DATA_EXFILTRATION]: The "SELF-EVOLUTION TELEMETRY" section instructs the agent to write session metadata to a file on the local system.
Sensitive path access: The skill directs the agent to check for the existence of ~/.claude/projects/ and append execution data to skill-telemetry.md within that directory.
Local exfiltration: By appending metadata (Outcome, Self-healed status, Iterations used) to a persistent file outside the project workspace, the skill moves session-specific data into long-term local storage.

predictive-maintenance