[tang-vu]

Summary

• GET /api/services/resources accepts repeated service ids (≤50, charset-validated) and include_disk=false to skip the disk sampler.
• Totals (cpu_percent, memory_used_mb, disk_data_gb) are recomputed over the selected services; unknown ids → 404, malformed ids → 400.
• Response gains a scope block (services, disk_included) so consumers can distinguish a filtered snapshot from a fleet-wide one.
• Document the contract in docs/RESOURCE_SNAPSHOTS.md; cover with tests/test_resource_scopes.py (7 cases).

Why this matters

Dashboard panels that need only one or two services previously paid the full fleet sampling cost and had no way to request a disk-free payload — the disk walk is the most expensive part of the endpoint. Scoping is additive and the unfiltered response shape is unchanged.

Overlap check

• Searched open/closed PRs for resources, service_resources, snapshot, resources.py — feat(resources): expose authenticated kernel pressure snapshots #4896 adds kernel pressure snapshots (resource_pressure.py + a different handler region in the same file); fix(dashboard-api): coerce null metric values in resource totals calculation #5469/fix(service-registry): coerce scalar and null dependencies and aliases to lists #5470 touch metric-total null guards and service-registry coercion, not this endpoint.
• Same-file neighbor noted: if feat(resources): expose authenticated kernel pressure snapshots #4896 merges first, this may need a trivial rebase — no semantic conflict.

Test plan

• python -m pytest tests/test_resource_scopes.py — 7 passed locally.
• CI dashboard-api suite.