The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untracked markdown files, which are untrusted inputs that could contain malicious instructions designed to bypass agent logic or corrupt documentation.
Ingestion point: Candidate files are identified in the repository via modules/candidate-detection.md.
Boundary markers: The skill implements a two-phase workflow where users must approve a consolidation plan before any actions are taken.
Capability inventory: The skill can write to files and delete source files, providing a potential impact surface if instructions are ignored.
Sanitization: No content-level sanitization is performed on the extracted text.
[COMMAND_EXECUTION]: The skill uses shell commands to interact with the local git repository status.
Evidence: modules/candidate-detection.md uses git status --porcelain to find untracked files.
[DYNAMIC_EXECUTION]: The skill relies on external scripts and detailed logic to perform its content merging and management tasks.
Evidence: SKILL.md references plugins/sanctum/scripts/consolidation_planner.py for planning operations.
Evidence: modules/merge-execution.md provides detailed Python-based logic for file operations and rollback support.

doc-consolidation