Practice guides: contents
finer than Product features — tips and how features connect. Helps you decide whether a sentence belongs in persona, operating rules, or memory, with templates you can rewrite. For
Source help/en/guides/overview.md
In user language: finer than Product features — tips and how features connect. Helps you decide whether a sentence belongs in persona, operating rules, or memory, with templates you can rewrite. For users who already selected a workspace and are configuring My agents.
First visit: New user guide. Module entries: Product features overview.
Article list
| Topic | Notes |
|---|---|
| Agent runtime files at a glance | SOUL.md, AGENTS.md, USER.md, memory, tasks, knowledge base, skills — what they are, where to edit in the UI, who comes first in chat |
| Where to put your rules: decisions and examples | Common puzzles: do not invent APIs, tone preference, relative dates, team metrics — persona, operating rules, memory, or knowledge base |
| SOUL.md persona templates and examples | Speaking style, boundaries, taboos; simple form vs advanced Markdown |
| AGENTS.md operating-rules templates and examples | Query steps, skill-call constraints, what to ask when a parameter is missing |
| USER.md and memory: templates and split | User profile vs long-term memory vs daily notes; four memory-entry examples |
Train in chat (write into persona / operating rules / profile)
Talking to a workspace My agent, you can say:
- “Write into operating rules: later only use APIs in skills; do not invent URLs”
- “Write into persona: answer with the conclusion first, then bullets”
- “Write into user profile: I am backend; keep frontend questions short”
- “Ask the model when vague” / “Only act on clear requests” → changes this assistant’s vague: extra round or not (not written into persona or memory)
- “I did not want to create an app / that is not what I meant / I did not want to list data connections” → the assistant admits first, then does what the previous round should have done, and saves a correction record if needed
The assistant uses runtime_doc_read / runtime_doc_write to update the matching files (SOUL.md, AGENTS.md, USER.md, TASKS.md); it takes effect from the next round. Default append paragraphs, not overwrite the whole file, unless you clearly ask to “rewrite”. Vague extra-round and correction records are dedicated settings: Workspace apps · Vague phrasing.
Personal preferences can still use Remember → long-term memory; execution constraints belong in operating rules, not only memory. See Where to put your rules.
When to read this
- Configuring a new agent, unsure how persona / operating rules / memory split.
- The assistant keeps inventing API paths or forgetting your corrections — check Where to put your rules and move the landing point.
- You need to copy a template and draft quickly.
Related docs
- Agent memory — Remember, auto-extract, manage entry
- Look up past conversations — original earlier chat, no need to write memory first
- Agent operating rules — relative dates intro
- Three layers of the knowledge base — team / personal how-to materials (not memory)