The Agent Skills Directory

[COMMAND_EXECUTION]: Subprocess calls are exclusively located in the test files (tests/test_skill_boundary.py, tests/test_workflow_integration.py) to verify the CLI functionality of the skill's own scripts. No arbitrary or untrusted command execution is present in the core logic.
[DATA_EXFILTRATION]: The skill operates entirely on local files within a user-defined Obsidian vault. There are no network requests or external data transmission patterns present in the source code.
[SAFE]: The tool implements a 'preview and wait' workflow that prevents any automated or unauthorized file modifications. The agent must display the proposed changes and wait for explicit user selection before proceeding to the write phase.
[SAFE]: Extensive validation of file boundaries is performed to ensure that both the glossary directory and the target Markdown articles are located within the same Obsidian vault. This prevents the tool from operating on sensitive files outside the intended workspace.
[SAFE]: A sophisticated provenance system uses SHA-256 hashes to bind every step of the workflow—from initial proposal to final link application—to the original source content. This cryptographic chain ensures data integrity and prevents injection or tampering during the processing pipeline.
[SAFE]: File modifications use an atomic write pattern (writing to a temporary file and replacing the original) combined with pre-write hash verification. This protects against data loss or corruption if files are modified by other processes concurrently.

paper-glossary