Troubleshooting

Updated SOUL.md But Agent Still Acts the Same?

A fast fix for the common “I changed personality instructions, but nothing changed” loop.

If you edited SOUL.md and your agent still sounds like the old version, you’re not crazy. This is a common setup issue, especially for new users testing across dashboard + Discord.

Key idea: Most failures come from editing the wrong file, testing in the wrong conversation surface, or not verifying that the file actually saved.

2-minute diagnosis flow

1) Verify the file was actually written

Ask your agent to show proof:

Read SOUL.md and quote line 1 plus one unique phrase I added.

If it can’t quote your new text, your edit didn’t land where you think it did.

2) Confirm you are in the same workspace/account

A frequent failure mode is editing SOUL.md in one account/workspace and testing in another. Verify account + workspace before retesting.

3) Test in one surface only

Dashboard chat, Discord DM, and Discord channels are separate conversation contexts. Do one clean test in a single place first.

4) Run a tiny personality smoke test

Use an obvious instruction and see if it appears immediately:

For this message only, answer in exactly 2 short bullet points and start with the phrase "Neon check:"

Most common root causes

Wrong file path: edited a different markdown file, not SOUL.md.
Unsaved change: text looked edited in chat, but file update didn’t persist.
Surface mismatch: expecting dashboard edit to instantly affect an old Discord thread’s behavior.
Instruction conflict: strong old rules in AGENTS.md or recent chat prompts override your style tweak.
Ambiguous SOUL text: personality request is too vague to produce a visible shift.

Don’t chase ghosts: changing SOUL.md won’t magically rewrite prior responses in an existing long chat. Test with a fresh, tiny prompt and visible style markers.

Known-good rewrite pattern

If your SOUL file is fluffy, rewrite key lines with measurable behavior:

Voice rules: - Keep replies under 120 words unless asked for detail. - Use plain language for non-technical users. - Start with direct answer first, then optional context. - Use one emoji max when tone is casual. - Never claim actions without proof (file path, command output, or link).

Advanced conflict checks (high-friction cases)

Competing rules in AGENTS.md: If AGENTS.md has strict voice/format rules, they can overpower softer tone tweaks in SOUL.md.
Old thread inertia: Long-running chats can keep behavioral momentum. Start a fresh session for a clean read of current files.
Dashboard edit confusion: If you asked chat to "change personality" but never verified file lines, you may have a planned edit rather than a persisted write.
Invisible test prompt: Broad prompts like "be better" are too vague. Use a measurable marker (exact phrase, word cap, output format).

Known-good verification prompt (copy/paste)

Before answering, read SOUL.md and quote the exact line that says "Never claim actions without proof". Then answer this in exactly 2 bullets, under 20 words each, starting with "Neon check:".

If it still doesn’t change

Escalate with evidence so support can reproduce quickly:

Issue: SOUL.md updated but behavior unchanged Where edited: [dashboard/files/path] Proof of file content: [quoted line from SOUL.md] Where tested: [dashboard chat | Discord DM | channel] Exact test prompt: [paste] Observed behavior: [what stayed the same] When started: [time + timezone]

Bottom line: verify file write + test one surface + use a measurable prompt. That resolves most “SOUL.md didn’t work” cases in minutes.