The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes a bridge script (scripts/invoke.py) that dynamically determines the path to a central management engine (kntnt) by searching various agent-specific configuration directories (e.g., ~/.claude/skills, ~/.cursor/skills, ~/.codeium/windsurf/skills). It then executes this engine using uv run to process the skill's logic.
[COMMAND_EXECUTION]: The skill performs file system and repository operations using git and uv. It executes commands to plan commits, apply changes, and inspect repository status via git diff and logs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the working tree (such as CHANGELOG.md entries and git diff output) and accepts user-provided instructions via the -- <instruction> syntax, which are then passed to the manager engine.
Ingestion points: User input from stdin, CHANGELOG.md content, and git diff output.
Boundary markers: The skill documentation defines a reserved -- separator to distinguish between formal arguments and natural language instructions.
Capability inventory: The skill has the ability to write to the repository (.gitignore, CHANGELOG.md) and execute shell commands via the uv and git binaries.
Sanitization: Content is processed by a central manager engine designed for these specific workflows, which manages the interpolation of data into commit messages and changelogs.

commit