The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs shell-level version control operations such as git push, commit, and branch management via the jj.execute method and JjWrapper API.
[EXTERNAL_DOWNLOADS]: Fetches the agentic-jujutsu package and its dependencies (e.g., @qudag/napi-core) from the NPM registry, which is a well-known package service.
[REMOTE_CODE_EXECUTION]: Utilizes npx agentic-jujutsu to download and execute the package logic directly from the remote NPM registry during setup.
[PROMPT_INJECTION]: Implements an indirect prompt injection surface through its 'ReasoningBank' self-learning loop.
Ingestion points: Untrusted task descriptions enter the context via jj.startTrajectory(task) and jj.getSuggestion(task) in SKILL.md and references/capabilities-and-api.md.
Boundary markers: None identified in the provided documentation or code snippets.
Capability inventory: The skill can execute shell commands (jj.execute), create branches, and manage commits.
Sanitization: No explicit sanitization or validation of the input task descriptions is shown before they influence the getSuggestion output used for subsequent operations.

agentic-jujutsu