A Codex skill for explaining agent work clearly, selectively, and with evidence.
It helps an agent communicate:
- what it is trying to accomplish;
- what it discovered or changed;
- why an important decision was made;
- how the result was verified;
- what remains uncertain.
The skill avoids private chain-of-thought and low-value narration of every tool call. Its goal is simple: give the user enough information to understand, evaluate, and control the agent's work.
Clone the repository and copy the skill into your Codex skills directory:
git clone https://github.com/HustleCoding/clear-agent-communication.git
mkdir -p ~/.codex/skills
cp -R clear-agent-communication/skill/clear-agent-communication ~/.codex/skills/Open a new Codex task so the skill list is refreshed.
Invoke the skill explicitly for important multi-step work:
$clear-agent-communication
Investigate why users are redirected to the dashboard after login.
Fix the issue and verify the result.
Update me only when you find the cause, make an important decision,
encounter a blocker, or complete verification.
The skill is most useful for investigations, implementations, research, deployments, data changes, and other work where evidence or user control matters.
Do not use it to make trivial answers longer.
Instead of:
I checked several files and found the issue. It should work now.
Prefer:
The frontend sends
displayName, while the API expectsname. This causes profile creation to fail. I corrected the mapping and added a regression test. The focused tests pass; production behavior remains unverified.
See examples/prompts.md for reusable prompts covering bug investigation, research, deployment, and high-risk changes.
skill/clear-agent-communication/
SKILL.md
agents/openai.yaml
examples/
prompts.md
Contributions should make agent communication clearer without adding operational noise. See CONTRIBUTING.md.