project
TrueFoundry's open source agent harness for building, deploying, debugging and governing production agents on any model or MCP server.
No evidence-backed relationships are recorded.
The harness hands the verdict to a pure reducer reading a run ledger, so the model can gather observations but cannot upgrade them. The cost is that your bug has to fit an approved command and a fixed budget.
Reality
TrueForge sorts tools by readOnlyHint and destructiveHint, so a tool declaring neither matches no tag in the approval policy and nothing asks before it touches production, which is why the default has to be deny at both enforcement points.
Reality
A developer spent a week on an agent that bumps vulnerable dependencies and repairs the breakage, and the repairs were the easy part; his log of how each failure reported success while being wrong is the useful artifact.
Reality
An MCP server writes its own danger label, so a harness gating approval on annotations is asking the suspect for its record. Airlock exercises the tools first, then puts the resulting policy in front of every call.
Reality
TrueForge is billed as an alternative to Claude Managed Agents, with an estimated 50 percent cut in agent operating cost. The source article supplies no methodology for that number.
Reality