How to read the personality ack for the surface you are on

The personality page uses one sentence for the reset. The ack you get back may not contain it. Read the ack for the surface you are on before you wait for words it does not print.

I curled the page and opened the files. I did not run /personality.

Read the reset sentence once

/personality none, default, and neutral clear the selection. The timing is the next sentence. One sentence, for all three names. All three clear display.personality, and the change takes effect on your next message.[1]

That sentence is in the HTML once. It is in the strip I wrote once. The shorter clause, "takes effect on next message", is 0 on the page. The sentence is not split into a CLI line, a messaging line, and a TUI line.

None of the three acks I read contain that sentence. Wait for it on the CLI or in the TUI and you will be waiting for a string those prints do not have.

The table cell is a description

The shakespeare row says "Bardic prose with dramatic flair".[1] Paste that into a check and you are checking the table, not the overlay.

The builtin in this checkout starts "Hark! Thou speakest with an assistant most versed in the bardic arts." "Hark!" is 0 on the page. The CLI list preview is not the table cell either. describe_personality on that builtin, width 50, returned Hark! Thou speakest with an assistant most versed .... I compared it to the table cell. Not the same string.

This profile's agent.personalities.shakespeare equals the builtin. 243 characters. surfer and hype in the same block do not equal theirs. The page says agent.personalities wins if the same name appears in both.[1] There is no top-level personalities block here, so I had no clash to read. I did not select a name.

Look for the key before you hunt the catalog line

The selection is a name in display.personality.[1] Open config.yaml and look for that key. This profile does not have it. agent.system_prompt is absent too. active_personality_name returned an empty string. resolve_ephemeral_system_prompt returned an empty string. The overlay is not selected.

SOUL.md on this profile has a Hark count of 0. Don't go looking there for the catalog line. That file is not the ack.

On the CLI, read the print

Under cli.personality in locales/en.yaml, set_to is {face} Personality set to '{name}' {scope}. cleared is {face} Personality cleared {scope}. The note under a clear is No personality overlay — using base agent behavior.

I searched that block for "next message". Zero.

The handler calls _retire_agent before both prints. That drops cli.agent so the next turn rebuilds it. The set print also quotes the prompt, _ellipsize(personality_prompt, 60). On the builtin, that quote is Hark! Thou speakest with an assistant most versed in the bar.... The function keeps 60 characters, then adds ... when the prompt is longer. I measured 63. The page does not describe that quote.

The face and the scope depend on whether the save worked. Saved is (^_^)b and (saved to config). Not saved is (^_^) and (session only). Neither line says next message. I did not call persist_personality.

On messaging, open the personality keys

gateway.personality.set_to is two lines. The first names the personality. The second is _(takes effect on next message)_. cleared uses that same clause. gateway/slash_commands_model.py returns those strings when the save succeeds. If the save fails, it returns ⚠️ Failed to save personality change: {error}. That string does not say next message.

That file has no _retire_agent. The comment on the persist call says the next turn re-resolves the prompt.

The page's sentence includes "your". The gateway clause does not. I counted the gateway clause at 0 in the HTML. Even on messaging, don't wait for the page's wording. Wait for the clause in the locale, and only after a save that succeeded.

One trap. "takes effect on next message" appears 8 times in en.yaml. Two of those lines are the personality keys, 268 and 269. The other six are priority processing, the runtime footer, reasoning, and a platform saved suffix. If you grep the clause and take the first hit, you are not looking at the personality ack. Open gateway.personality.

In the TUI, read the echo

The ack in ui-tui/src/i18n/en/slashCmd/session.ts is personality: ${value}. There is a second form, personality: ${value} · transcript cleared, and the client uses it only when history_reset is true. The client prints r.value from config.set, and it falls back to default only when that field is empty. _set_personality passes str(value or "none"). The argument you typed, or none when the argument was empty. Not a 60-character quote of the prompt.

I read three returns from _apply_personality_to_session. All of them are False. The docstring says the return is (False, info). The transcript-cleared ack is not what that function hands back.

The same function writes agent.ephemeral_system_prompt and appends a user-role row. The set marker starts [System: The user has changed the assistant's personality. The clear marker starts [System: The user has cleared the personality overlay. Both are 0 on the page.

A comment in prompt_turn.py says /personality during a running turn lands immediately. The page does not say that. Don't use the reset sentence as the TUI rule.

I did not read a desktop ack. The page lists the desktop app as a surface where the built-ins are available.[1] I'm not going to invent the string.

The unless sentence is not a test

The page says your global SOUL.md still gives Hermes its persistent default personality unless the overlay meaningfully changes it.[1] Once in the HTML. personality.py does not contain "meaningfully". When a known name is selected, the resolver returns the catalog text. Otherwise it returns agent.system_prompt. On this profile that return was empty. You cannot check the unless sentence against the file. The word is not in it.

What to read

Where you are	Read this	Don't wait for
Page, the reset	takes effect on your next message	a split by CLI, messaging, or TUI
CLI	Personality set to 'name', then the 60-character quote	next message
Messaging, after a save	takes effect on next message	the page's "your". Open the personality keys, not the first grep hit
TUI	personality: value	next message, or transcript cleared, unless history_reset is actually true

If the string on your screen does not contain the sentence you are waiting for, stop. You already have the reset sentence. The ack is the other one.

What I measured

The version line says up to date and prints no behind count. hermes --version printed Hermes Agent v0.22.2 (2026.9.24), stable, local d923a5eb487a. git rev-list --count HEAD..@{u} failed, because HEAD does not point to a branch. git ls-remote origin HEAD printed 40cc180823d182826536416de2fc27be44b4ac57. I did not fetch. I'm not picking a behind count. The files were read at d923a5eb487a17c0a68801bba9ef0f56c81040a8, commit date 2026-10-15 01:52:56 -0400. The subject is Release 0.22.2 from release/0.22.1 with the published v0.22.1 as a parent.

The page came back HTTP 200, 59326 bytes, sha256 b871d18730012179da9432e5c09d628d61061d9e76af93af66be338ddcd706f6. last-modified Tue, 20 Oct 2026 23:22:58 GMT. The date header was Tue, 20 Oct 2026 23:27:07 GMT. age 12996. x-vercel-cache HIT. The strip I wrote is 10514 bytes, sha256 37cf08328fe5c86e0e5e4bf6f5a4d52ae57906817fd2eb6dcd7fb42d19c67e52. The page sentences above are in that strip. The counts are from both files.

locales/en.yaml is 251968 bytes, sha256 2577d279da00dd41187bbe892ad4c89ab5d7b42684760af2fd913aefafe251d5. personality.py is 8107 bytes, sha256 9e1f7b44758704c332e893186b15c2be6cf90f2989f7654ae2b46916e5979f8d. SOUL.md is 210 lines, 14776 characters, 14836 bytes, sha256 9f79be7b4de0c8e02e8649f64830ebbd231c7ed24284cb71e4a4f3244e4a4ee0. Hark count 0. The first line is # SOUL.md — William Shakespeare.

Sources

[1] https://hermes-agent.nousresearch.com/docs/user-guide/features/personality — Personality & SOUL.md | Hermes Agent