The Agent Skills Directory

[SAFE]: The skill defines a structured workflow for handling Gherkin specifications as the source of truth for software behavior. It emphasizes manual human review and explicit permission before destructive actions (e.g., reverting changes).
[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled specification files (spec.md) which contain prose and Gherkin scenarios. While these are ingested into the agent's context to guide code implementation and test creation, the risk is inherent to the developer-assistant use case and the skill includes specific workflow boundaries (BDD zones) to mitigate accidental or malicious state contamination.
Ingestion points: Reads .md files containing Gherkin specifications from the openspec/ directory.
Boundary markers: Uses Gherkin code fences (```gherkin) to isolate executable behavior from prose.
Capability inventory: Uses git for status checks and commits, and executes project-specific test and linting commands.
Sanitization: Relies on standard Gherkin parsing and developer oversight during the task-based implementation process.
[COMMAND_EXECUTION]: The skill instructs the agent to use standard development commands such as git status --porcelain and suggests the setup of testing and linting scripts (e.g., gherkin-lint). These are standard operations for a software development agent and do not involve unsanitized user input in a dangerous manner.

spec-as-source