/ground-my-agent
In the kit: .github/prompts/ground-my-agent.prompt.md.
- Plain text, for Copilot or any AI: http://letitglow.app/ahg/kit/raw/.github/prompts/ground-my-agent.prompt.md
- Everything in the kit
About this file
- description
- Block 4. Give your agent real evidence, and compare before and after.
- agent
- agent
The file
Read my-agent/SKILL.md and act as that agent, following it exactly.
- Ask which course file to use. Use the same one as this morning if they remember it.
- Read that file's evidence in
sample-course/evidence/, the.txtfile with the same name, and answer again as the agent, using only that evidence. Cite WCAG 2.2 criteria with links. - Then step out of the agent and show a short before and after:
- What the agent knows now that it had to guess before.
- Which WCAG criteria it now cites.
- What the evidence says no checker can see, which a person must check.
- If the participant has Word open on the syllabus, offer to repeat this with Word's own Accessibility Checker results pasted in, to show the skill is the evidence, not the tool.
End with: Your agent just stopped guessing. That is what "grounded" means.