What changed

handle_request in the JSONL server now includes LifecycleError in the caller-visible domain-error tuple, next to ProposalError, so lifecycle guard failures map to {"code": "invalid_request"} instead of falling through to the generic internal_error branch. one import, one tuple entry, a regression test, and a changelog line.

Why

the envelope treated ValueError / ProposalError / ArtifactNotFoundError as bad requests, but LifecycleError — a RuntimeError subclass exactly like ProposalError — fell through to the catch-all. on current test:

{"id": "r1", "method": "kb.supersede", "params": {"old_claim_id": "c1", "new_claim_id": "c1"}}
→ {"id": "r1", "ok": false, "error": {"code": "internal_error", "message": "a claim cannot supersede itself"}}

plus _log.exception dumping a full server-side traceback for what is a caller mistake. the http /rpc transport routes through the same handle_request, so both wire surfaces misreported it — while the cli already translated the identical error into a clean one-liner via _cli_errors (which lists LifecycleError explicitly for this exact reason). an agent retrying on internal_error (a server fault, plausibly transient) will loop on a request that can never succeed; invalid_request tells it the request itself is wrong.

any future lifecycle guard raising LifecycleError gets the correct mapping for free.

Testing

• tests/test_jsonl_server.py::test_jsonl_lifecycle_guard_maps_to_invalid_request — fails on current test (envelope carries internal_error), passes with the fix; also asserts the guard fired before any write, leaving the claim untouched.

ran locally:

python -m ruff check src tests   # all checks passed
python -m mypy src               # success: no issues found in 91 source files
python -m pytest tests/ -q --ignore=tests/embeddings   # 1151 passed, 28 skipped