The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill derives its merge strategy and file modification plan from potentially untrusted data in roadmap files and source skill directories.
Ingestion points: Data is ingested from roadmap JSON files (e.g., docs/plans/roadmap-<wave>.json) and the contents of the source skill directories being merged.
Boundary markers: The instructions do not specify any delimiters or safety markers to differentiate external data from the skill's instructions.
Capability inventory: The skill has extensive write access to the project filesystem, can delete directories via scripts/zj-git rm, perform git renames, and execute local Python scripts.
Sanitization: A mandatory human-in-the-loop checkpoint is included where the agent must show the file-level plan to the user for confirmation before proceeding.
[EXTERNAL_DOWNLOADS]: The skill logic explicitly handles source components identified by external URLs.
Evidence: The pre-condition section references roadmap decisions that may contain a "URL -> scratch path" mapping for source components.
[COMMAND_EXECUTION]: The skill executes shell commands and Python scripts to perform git operations and metadata updates.
Evidence: The process includes executing python3 <zj-roadmap-driven-skill-dir>/roadmap_cli.py to manage roadmap status.
Evidence: Uses a local script scripts/zj-git rm for directory deletions.

zj-merge-skill-pair