/sandbox/agent-surface

Agent Surface

A control-plane board for deciding whether an agent workflow has enough context, scoped authority, recovery paths, and receipts to become real work instead of a clever demo.

workflow

Ship a repo fix

Patch a failing production route, run the build, push a narrow commit, and leave deployment evidence.

Weakest signalFailure recovery

58/100

Review modecheckpoint

66/100

Evidence modereceipts first

82/100

dials

Context pack78

thin promptfull workspace

Permission scope62

wide openleast privilege

Tool readiness84

manual gapsscripted tools

Human review66

silent runclear gate

Failure recovery58

dead endcheckpointed

Evidence trail82

trust mereceipts

Trust posture64

watch closelydelegate

copyable run brief

Agent Surface run brief: Ship a repo fix
Objective: Patch a failing production route, run the build, push a narrow commit, and leave deployment evidence.
Domain: code operations
Readiness: 71/100 — STAGE
Operating mode: Run in a narrow sandbox and require a human checkpoint before external effects.
Tighten next: Failure recovery, Permission scope
Evidence required: diff/log/output links, explicit files touched, validation command result, and unresolved risks.