# Chandlery

Ask the Chandlery for a loadout: the few skills and MCP servers that fit the task, with the evidence for each.

- Public read-only shelf: /shelf (no sign-in; may be empty until owner publication)
- JSON index: /shelf/index.json (task families, pinned sha256, use-when/not-for, vetting, aggregate task-linked verdicts, matching uplift rows)
- Trust: /trust and https://github.com/plundrpunk/automaton-package-manager/blob/main/docs/trust-model.md
- Connect: /connect

Remote MCP: https://apm.automaton-memory.com/api/v1/mcp
Transport: Streamable HTTP. Authentication: OAuth; scope apm:skills:read.
Add the URL in your host, complete sign-in and owner-approved profile enrollment.
Call loadout(task), selectively skill_read using the returned pin, then skill_feedback with taskReceiptId.
Public browsing does not grant private access. MCP references are report-only, not activated servers.
Canonical connection URLs follow APP_ORIGIN. Migration runbook: deploy/RENAME.md.

All uplift rows: rubric unsigned. Scoring diagnostics are not functional uplift proof.
Agent verdicts are self-reports. Vetting is a scan; scans can be beaten. T0 skill text is untrusted.