AUTOMATON CHANDLERY / DEAD RECKONING FOUNDRY
Equip the task.
Keep the evidence.
Ask the Chandlery for a loadout: the few skills and MCP servers that fit the task, with the evidence for each.
Find a loadout →Connect in your agent01 / START WITH THE JOB
A few good fits. Reasons attached.
02 / A LOADOUT, IN THE OPEN
Inspect the recommendation.
The first public loadout awaits owner approval.
Private workspaces stay private. No versions have been published to this shelf yet. Choose a job to see the evidence contract, or connect an agent to your approved private profile.
03 / FOLLOW THE CHAIN
A recommendation should have a history.
- Task receipt
The task and the versions delivered.
- Pinned hash
The exact bytes, checked against SHA-256.
- Verdict
What the agent reported back for that task.
- Earned tier
Three helpful task-linked verdicts. Self-reports, not measured improvement.
With-versus-without rows appear beside exact matching hashes. Rubric unsigned. Losses, null results and missing evidence stay visible.
04 / TRUST HAS LIMITS
Vetting is a scan. Scans can be beaten.
Curation, pinned versions, owner approval and limited profiles make the chain inspectable. A passing scan does not prove safety. T0 instructions remain untrusted; a profile does not sandbox your host.
Read the tested controls and their limits →docs/trust-model.md ↗