Personalised

Documentation Index

Fetch the complete documentation index at: https://docs.xysq.ai/llms.txt

Use this file to discover all available pages before exploring further.

A personalised skill is your baseline skill plus an overlay generated from your own memory. The baseline tells the agent how to use xysq. The overlay tells the agent who you are — your working style, your active context, your stated preferences, and the behaviour rules you’ve laid down. The overlay grows sharper as your memory grows. The agent does not learn from you live in one session; instead, your memory consolidates into structured signals, and the next time the skill is regenerated those signals get baked in.

​

What’s in the overlay

The personalised section injected under ## About you in the skill file is composed of five distinct inputs:

​

1. Tier-1 mental models (always-on)

Four standard models, computed from your full memory bank:

• User identity — who you are, what you do, your domain
• Communication style — how you prefer to be addressed and responded to
• Preferences — concrete likes, dislikes, defaults you’ve expressed
• Context — what you’re currently working on

These refresh after every memory consolidation cycle.

​

2. Tier-2 tag-promoted mental models

Domain-specific models derived from your tag taxonomy. When you tag enough memories with a custom tag (e.g. project:xysq or domain:fintech), xysq promotes that tag into its own mental model — a structured summary of everything memory-side that touches that tag.

​

3. User directives

Verbatim behaviour rules you’ve explicitly told xysq to enforce. Added via the skill_directive_add MCP tool. Directives are not derived from memory — they’re rules you state once and want applied every session. Example directive: “Always use TypeScript strict mode in new files. Reason: caught three null-check bugs last quarter.” Directives can be deactivated (soft) or deleted (hard). Deactivation triggers a regen so the rule disappears from the next generated skill.

​

4. Tag taxonomy

Your custom tag definitions get included so the agent knows what each tag means and uses them correctly when retaining new memories.

​

5. Trigger index

Always-on instructions that tell the agent how to detect decisions, corrections, and preferences in conversation — so capture happens automatically without you asking.

​

How the overlay is formed

Generation flow when you click Download personalised skill (or when xysq auto-regenerates):

1. Read all mental models from your memory bank (Tier-1 + Tier-2).
2. Fetch active directives.
3. Pull your tag taxonomy.
4. Render each section into Markdown.
5. Compute a SHA-256 hash of the rendered content.
6. Inject the block into the baseline skill under ## About you.
7. Write a versioned file per supported platform.

If the hash matches the previous version, no new file is written — the version number stays put. This means re-clicking generate is cheap; you only get a new version when something actually changed.

​

What feeds the overlay

The overlay is only as good as the signals feeding it. Three things shape it: Memories retained as significant. When the agent calls memory_retain with significance="high" — usually after a decision, correction, or strong preference — that memory becomes a candidate for mental model refresh. Routine facts are stored too, but high-significance items are weighted heavily by consolidation. Tag classification. Memories tagged with consistent custom tags get promoted into Tier-2 mental models. This is the lever you have for shaping domain-specific personalisation: tag your fintech memories domain:fintech, and a few weeks later “fintech” becomes its own structured section in your skill. Directives. Direct rules you set yourself via skill_directive_add. These bypass memory and go straight into the skill verbatim. Use them for behaviour that should never drift: testing rules, framework choices, communication style hard-limits.

​

Lifecycle

All versions are kept. Nothing is destroyed by a regen.

​

What it is not

To set expectations honestly:

• Not real-time. The personalised skill is a generated artifact, not live retrieval. New memories don’t appear in your agent’s skill until the next regen.
• Not template substitution. The overlay isn’t {{user_name}} placeholders being filled in. It’s structured outputs from memory consolidation being rendered as prose.
• Not per-conversation. One personalised skill per user, used across all your agents. Scope is your personal memory bank.
• Not shared. Only the authenticated user can read or regenerate their own personalised skill. Team-scoped skills are a separate surface.

​

When to regenerate manually

You usually don’t need to — the dashboard handles it. But trigger a manual regen if:

• You’ve just added a batch of directives and want them all in the next file you download.
• You’ve imported a large batch of memories (e.g. from another tool) and want them reflected immediately.
• You changed your tag taxonomy meaningfully (renamed tags, added a domain).

For anything else, the daily auto-refresh and post-directive regen will keep your skill current.

​

Related

• Skills overview — what skills are and how to install
• Vault — where memories live
• Knowledge base — separate from memory, also feeds the agent