The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local Python script (cli.py) to handle file system operations and project metadata. It accepts user-provided paths and identifiers as command arguments.
Evidence: python3 ${CLAUDE_PLUGIN_ROOT}/tools/scrivener/cli.py verify --project "<path.scriv>" in SKILL.md.
[PROMPT_INJECTION]: The skill processes external Scrivener project files, creating an indirect prompt injection surface.
Ingestion points: Binder XML (.scrivx) and text/RTF files in the Data/ directory as described in references/integrity-model.md.
Boundary markers: The instructions do not define specific markers to separate project data from agent instructions.
Capability inventory: Execution of a local CLI tool with permissions to read, write, and archive project files.
Sanitization: The CLI tool is responsible for parsing project files; the agent observes the tool's structured JSON output rather than directly executing content from the project files.
[SAFE]: The skill prioritizes project safety with a preflight check that detects concurrent access via lsof and performs automatic ZIP backups before any data is modified.

scrivener-integrity