The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the agent-spec binary if it is not already present on the system.
Evidence: command -v agent-spec || cargo install agent-spec found in SKILL.md.
This uses the official Rust package registry (crates.io) to fetch the tool.
[COMMAND_EXECUTION]: The skill's primary functionality is built around executing the agent-spec CLI tool and version control commands.
Evidence: Execution of various subcommands including agent-spec contract, agent-spec lifecycle, agent-spec guard, and agent-spec resolve-ai.
Evidence: Interaction with Git and Jujutsu (jj) version control systems for checking file status and applying git trailers.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of .spec files, which define the 'Contract' the agent must follow.
Ingestion points: Specification files (e.g., specs/*.spec) are loaded from the project directory.
Boundary markers: The workflow uses structured sections such as Intent, Decisions, Boundaries, and Completion Criteria to define constraints.
Capability inventory: The agent can run CLI commands, read/write files in the workspace, and interact with the repository's version control.
Sanitization: The skill implements a quality gate using agent-spec lint and agent-spec parse to validate the structure and clarity of the input specifications before they are acted upon.

agent-spec-tool-first