HACP v0.3-candidate
Accountable agent work, explained from the evidence
Eleven problem-first articles for leaders and implementers. Start with the evidence boundary, or jump directly to the problem you are solving.
The canonical maturity and non-claims boundary.
Interworking evidenceTwo Implementations, One Candidate ContractWITS production, HACP validation, and separate who-decides semantics.
Authority boundaryA Receipt Is Evidence, Not ApprovalWhy approval, execution, reports, and receipts answer different questions.
One-use admissionWhy Idempotency Does Not Prove One-Use Human ApprovalConcurrent claimants, restart readback, and stale replay.
Decision gatesHuman-in-the-Loop Without Approval TheaterA practical checklist for exact, bounded human choices.
Fail closedWhen Stop Is a Successful Agent OutcomeTyped outcomes for missing authority, stale context, and scope conflict.
Accountable handoffsFrom Clipboard Relay to Accountable ContinuationPreserving identity, scope, authority, and causation across contexts.
Layer boundariesMCP Moves Tool Calls; HACP Governs Accountable WorkHow capability transport and human-approved coordination compose.
Executable conformanceHow We Test an Agent ProtocolValid branches, exact negative diagnostics, digests, and external bundles.
The next transitionYour Agent Said Done. Who Authorized the Next Step?Why a report cannot silently become permission.
Construction provenanceWho Built This Packet?Why integrity cannot identify the builder or prove correct assembly.
Early implementers
Challenge the candidate against a real workflow
Implement the candidate, run the conformance package, and tell us where the contract fails against your workflow.