The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell scripts stored within the plugin environment (${CLAUDE_PLUGIN_ROOT}/scripts/), specifically init-writing-dir.sh for project initialization and archive-search.sh for retrieving historical context.
[COMMAND_EXECUTION]: Standard Git commands such as git add and git commit are integrated into the task templates to ensure incremental progress is saved for each section and artifact generated.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes user-provided data from .writing/outline.md and .writing/metadata.yaml to populate task descriptions in the generated plan.
Ingestion points: .writing/outline.md, .writing/metadata.yaml, and claims/section_*.md files.
Boundary markers: None present for input file interpolation.
Capability inventory: Execution of shell scripts, file system write operations, and delegation to subsequent drafting and reviewing skills.
Sanitization: None performed on the content of the outline or claims before inclusion in the plan.
[EXTERNAL_DOWNLOADS]: References to external research data are managed through the Zotero-MCP server, which is an established service for bibliographic management.

writing-plans