/memory-management-cross-agent

What it is

Cross-agent memory extensions -- read-only access to another agent's sidecar (for example, the validator reading the architect's decision log) and shared budget monitoring utilities. Split from memory-management to keep each skill under the size limit.

Called by

• /gaia-val-validate (cross-referencing ground truth)
• Any agent that declares cross-references in its persona definition

What it does

Provides just-in-time cross-reference loading: an agent can read another agent's decision log or ground truth as read-only context.

Budget monitoring utility returns threshold status (ok, warn, alert, archive_needed) used by session-save operations.

Progressive downgrade when budget is tight: full content, then recent entries, then summary, then skip.

When you will see it surface

• You will not directly see this skill surface. It operates behind the scenes when agents need context from other agents' memory.
• If a cross-reference target is missing, a warning appears in the output but the workflow continues.

Technical notes

• Cross-references are strictly read-only. An agent can never write to another agent's sidecar.
• Self-references (an agent pointing at its own sidecar) are rejected at load time.
• Failures are soft -- missing or malformed targets produce warnings, never halts.