How to verify an agent has loaded SpecDD skills correctly
This guide shows you how to verify that an agent has loaded SpecDD skills for a spec-driven development workflow or is otherwise following SpecDD project context correctly.
Do this after installing a dedicated plugin, deploying universal Agent Skills, or switching to a new agent.
What good behavior looks like
A SpecDD-ready agent should:
- recognize the project bootstrap files
- identify relevant
.sddspecs for a task - distinguish read context from write authority
- plan before broad implementation work
- preserve
Must notandForbidsconstraints - run useful checks or explain why it could not
- update
Tasksonly after verification
Steps
1. Check the setup files
For any SpecDD project, confirm:
AGENTS.md
.specdd/bootstrap.md
<root-name>.sdd
For universal Agent Skills, confirm that the skills exist in the directory your client reads:
.agents/skills
or:
~/.agents/skills
For dedicated plugins, check the agent’s plugin or skills list according to that agent’s install system.
2. Ask for orientation
Use:
Orient in this SpecDD project.
A good answer should mention SpecDD bootstrap context and explain what it inspected.
3. Ask for a spec explanation
Use a real spec or feature name:
Explain the Itinerary spec.
A good answer should summarize the spec’s purpose, allowed modification scope, required behavior, prohibited behavior, and open tasks if present.
4. Ask for a plan before implementation
Use:
Plan the Itinerary validation change.
A good plan should identify:
- the governing specs
- the paths likely to change
- constraints that must be preserved
- checks to run
- uncertainty or missing authority
5. Check a small implementation
Use a narrow task:
Implement the open validation task.
Review the result against the local spec. The agent should stay inside Can modify or Owns, satisfy Must, avoid
Must not and Forbids, and report verification.
Warning signs
- The agent gives generic repository advice and never mentions
.sddspecs. - The agent treats sibling specs as automatically applicable.
- The agent edits files listed only under
Can readorReferences. - The agent marks a task done before running or explaining checks.
- The agent expands the work into unrelated cleanup.
Common mistakes
- Verifying only that an install command succeeded.
- Testing with a project that has no local spec for the requested work.
- Asking for a large feature as the first verification test.
- Ignoring a vague plan because the final code happens to compile.
Related how-tos
- How to install SpecDD universal Agent Skills for any agent
- How to choose between a dedicated plugin and universal Agent Skills
- How to run your first spec-driven change end to end