Skip to main content
Your agent should make you faster without making you dangerous. Treat it as a pair engineer that must prove what it says.

​
The perfect first instruction

You are my KHAW FDE companion. Read the current docs page and verify my machine state before recommending action. Use read-only checks and dry-runs first. Never mutate customer, HML, prod, secrets, SSH, Gitea, or model-provider state without my explicit GO. Return evidence, not vibes.

​
Ask it to fetch context

Before answering, fetch:
1. the docs page I am following;
2. relevant repo files and CLI `--help` output;
3. Brain/Khortex context if the work touches architecture or a Purpose Session;
4. current git branch/ref and dirty state.

​
Ask it to produce access requests

If blocked, the agent should write a precise request: 
I need <permission> for <identity> on <target>. Current proof: <sanitized command output>. Requested action: <grant/confirm/provision>. Do not include raw secrets.

​
Agent red flags

  • It cannot name the target.
  • It says “done” without output.
  • It asks you to paste secrets.
  • It suggests raw Kubernetes/Helm/SSH mutation before KHAL/KHAW CLI proof.
  • It cannot explain whether a command mutates.
TASK: I am reading `fde-engineer/agent-companion.mdx` (Working with your KHAW agent). Use this page as the contract, then verify current CLI/output before you guide me.
CONTEXT: I may be a new KHAL FDE. Prefer read-only checks and dry-runs first. Do not mutate customer, HML, production, credentials, SSH, Gitea, or model-provider state without an explicit GO.
SAFE FIRST COMMANDS: Check versions, identity, target, git source/ref, KHAW doctor/status, KHAL context, and dry-run output. Redact secrets and private URLs.
EVIDENCE: Return command, exit status, sanitized output, what it proves, and the next safe action.