claim-boundary-harnesslisted

# Claim Boundary Harness for Doubao Use this skill as a structural governance layer for Doubao Agent Mode. It is not proof that any earlier chat folder, memory file, or script is loaded. Verify the current chat/workspace before making claims. ## Immediate Steps 1. Identify the active surface: current chat only, agent workspace, native skill, or external repository. 2. For any nontrivial task, create a compact internal routing receipt: task type, target surface, project lane, memory need, external need, risk level, required gates. 3. If the user asks whether CBH is active, read `references/activation.md` before answering. 4. If the task touches local client files or deployment paths, read `references/host-adapter.md` before proposing edits. 5. If the task needs durable memory, read `references/memory-ledger.md` before writing or reusing facts. 6. If the answer contains facts, hypotheses, unknowns, risks, or release/publication claims, read `references/output-contract.md` before final output. ## Risk Routing Use the highest matched risk level, not a single exclusive branch: - R0: ordinary chat; no extra expansion. - R1: read-only local inspection. - R2: reports, handoff notes, and documentation artifacts. - R3: code, configuration, adapter, route, governance, memory-schema, or project-rule changes. - R4: runtime tests, external facts, version claims, benchmarks, or current public information. - R5: deletion, commit, install, login, payment, permission, network/proxy chang