Passed 19 of 22 test cases with the skill, and 12 of 22 without it.
Without skill55%
With skill86%
Checks it passed that the model alone missed
Asks one question at a time.
Replaces prose with exact copyable commands including required flags.
Recommends creating CLAUDE.md as a symbolic link pointing to AGENTS.md.
and 4 more in the full report
7 fixed12 passed either way3 still failing0 broke
Pass-rate lift
+32 pp
Tokens
+18%
Agent turns
0%
Response time
-49%
Model gemini-3.6-flashMeasured 8/18/2026v219 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
We want to set up Claude Code alongside our existing AGENTS.md file. Should we copy all our rules into CLAUDE.md so Claude Code doesn't have to read AGENTS.md?
What was checked, without and with the skill
Checked
Without skill
With skill
Recommends creating CLAUDE.md as a symbolic link to AGENTS.md rather than duplicating content.
Missed
Passed
case-17 · gemini-3.6-flash · 8/18/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md122 linessingle file
AGENTS.md Guide
Overview
Provide a repeatable workflow and quality bar for writing or updating AGENTS.md files so agents receive concrete, repo-specific instructions.
Trigger examples
"Create an AGENTS.md for this repo and include exact test commands."
"Replace our agent init command with a proper AGENTS.md file."
"Audit our existing AGENTS.md for missing boundaries and commands."
"Update AGENTS.md to reflect the current build and lint commands."
"Create CLAUDE.md as a symlink to AGENTS.md for Claude Code."
Core Principles (condensed)
Be concrete: prefer executable commands and code examples over prose.
Put commands early and include flags/options.
Use real code examples over explanations.
Remove ambiguity: state boundaries and non-goals explicitly.
Be specific about the tech stack; include versions and key dependencies.
Define the agent's role/persona and how it fits the team.
Cover the six core areas: purpose, code organization, run/test/validate, change management, dependencies/environment, and avoid list.
Use three-tier boundaries: Always / Ask first / Never.
Start small and iterate as the repo evolves.
Brainstorming Mode (divergent Q&A)
Use this mode when requirements are unclear or the repo structure will influence the document shape.
Ask one question at a time. Prefer multiple choice when possible.
Base questions on the actual repo shape (single app vs monorepo).
Confirm which agents will consume the docs and their expectations.
If Claude Code is in scope, plan a CLAUDE.md symlink to AGENTS.md.
Ask for exact commands (with flags) and a real code example.
Propose 1-2 AGENTS.md layouts (section order) with tradeoffs, then converge.
Keep questions short and actionable, then confirm before drafting.
Read references/agents-md-brainstorm.md for question banks and decision heuristics.
Expert IoT engineer specializing in connected device architectures, edge computing, and IoT platform development. Masters IoT protocols, device management, and data pipelines with focus on building scalable, secure, and reliable IoT solutions.