From 892c5e4d95e153775874618605c54f49931f7f68 Mon Sep 17 00:00:00 2001 From: Harshul jain Date: Sun, 11 Oct 2026 18:33:32 -0400 Subject: [PATCH] chore: pin the framework at 0.0.4, and install voice from PyPI MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The pins were at 0.0.2, two releases behind, so a clean install of this repo got none of the voice work it demonstrates. agentship-voice is now installed rather than noted as absent. The note said it was unpublished and had to come from a sibling checkout; it has been on PyPI since 0.0.3, and leaving that standing meant the voice demo and its tests silently SKIPPED on a clean install — `pytest.importorskip` found nothing to import and said so quietly, which reads identically to passing. The `[pipecat]` extra is what pulls Pipecat, and with it the local smart-turn model that decides whether you have finished a sentence or only paused. Verified by installing requirements-dev.txt into a clean Python 3.13 venv from PyPI: all six resolve at 0.0.4 and Pipecat 1.12 comes with them. --- requirements-dev.txt | 26 +++++++++++++++----------- 1 file changed, 15 insertions(+), 11 deletions(-) diff --git a/requirements-dev.txt b/requirements-dev.txt index ce08bd3..7e65a6b 100644 --- a/requirements-dev.txt +++ b/requirements-dev.txt @@ -1,29 +1,33 @@ # The AgentShip framework, from PyPI. # # This used to install the framework editable from a sibling checkout, because it was not -# published. It is now — 0.0.2 — and the sibling paths only ever resolved on a machine with +# published. It is now — 0.0.4 — and the sibling paths only ever resolved on a machine with # both repositories side by side, so CI could not install this repo at all: pip reported # "../agentship/packages/agentship-core is not a valid editable requirement" and stopped. # # python -m venv .venv && source .venv/bin/activate # pip install -r requirements-dev.txt # or `make install` # -# Pinned exactly. The six ship in lockstep and are only ever tested together, so a mixed set +# Pinned exactly. The seven ship in lockstep and are only ever tested together, so a mixed set # is a combination nobody has run. -agentship-core==0.0.2 -agentship-langgraph[mcp]==0.0.2 -agentship-cli==0.0.2 +agentship-core==0.0.4 +agentship-langgraph[mcp]==0.0.4 +agentship-cli==0.0.4 # The OTel span-tree + exporters (console/Opik/LangFuse/LangSmith). -agentship-observability==0.0.2 +agentship-observability==0.0.4 # `[serve]` adds uvicorn: the service demo boots a real server on a loopback port rather than # relying on TestClient, because "the ASGI app starts" is part of what it claims to show. -agentship-service[serve]==0.0.2 +agentship-service[serve]==0.0.4 -# NOT here: agentship-voice. It is unpublished while the voice phase lands, so the voice demo -# and its tests skip on a clean install rather than failing — `pytest.importorskip` in -# tests/test_voice.py. Install it from a sibling checkout to run them: -# pip install -e ../agentship/packages/agentship-voice[pipecat] +# Voice, with Pipecat. This said agentship-voice was unpublished and had to come from a sibling +# checkout; it has been on PyPI since 0.0.3, and leaving the note meant the voice demo and its +# tests silently skipped on a clean install — `pytest.importorskip` in tests/test_voice.py found +# nothing to import and said so quietly. +# +# The extra is what pulls Pipecat itself, and with it the local smart-turn model that decides +# whether you have finished a sentence or only paused. +agentship-voice[pipecat]==0.0.4 pytest>=8 pytest-asyncio>=0.24