graff — a fast agentic coding harness in Zig: multi-provider, MCP, workflows, DGM evolution loop, TS/Python SDKs
Python
ACP released-fixes test: a successful edit keeps its diff; tier-1 baseline 2855
Rach Pradhan8eba9610b128
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 4c5c6954..81bbb8c0 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -26,6 +26,7 @@ current is part of cutting a release.- Tool calls whose arguments a server lost (empty, `{}` or cut off) are rebuilt from the call markup in the same reply and the markup is removed from the message. The affected model family gets plain parameter types instead of nullable unions, and a turn whose last three tool batches all failed on unusable arguments ends instead of looping. (#1276)- Over-cap tool output keeps its head and tail around a truncation marker, so test summaries and final errors survive. (#1271)- Images read with `read_file` are downscaled to the provider-safe size, or refused with a reason, instead of failing the next request. (#1272)+- When a reasoning model spends its whole reply reasoning without answering, the retry asks for brief thinking and a direct answer instead of repeating the identical request. (#1293)- Retry backoff adds bounded jitter so parallel workers don't retry in lockstep; a server's Retry-After is never shortened. (#1274)### ACPdiff --git a/scripts/eval/tier1-manifest.json b/scripts/eval/tier1-manifest.jsonindex 26af8314..e0ed8c00 100644--- a/scripts/eval/tier1-manifest.json+++ b/scripts/eval/tier1-manifest.json@@ -15,7 +15,7 @@"src/repl.zig","TUI/root.zig"],- "test_count_baseline": 2823,+ "test_count_baseline": 2855,"test_count_slack": 25,"required_invariants": [{diff --git a/scripts/test-acp-released-fixes.py b/scripts/test-acp-released-fixes.pyindex 0231c836..522584f9 100644--- a/scripts/test-acp-released-fixes.py+++ b/scripts/test-acp-released-fixes.py@@ -147,11 +147,18 @@ def main():assert result.get('stopReason') == 'end_turn', resultassert model.error is None, model.errorreturn updates- edits = terminal_updates(prompt('Apply the scripted batch, correct its invalid span, then finish.', 3))- assert len(edits) == 2, edits- assert [u['status'] for u in edits] == ['failed', 'completed'], edits+ turn = prompt('Apply the scripted batch, correct its invalid span, then finish.', 3)+ edits = terminal_updates(turn)+ assert len(edits) == 1 and edits[0]['status'] == 'failed', editsassert 'no batch changes written' in text(edits[0])- assert '2 edit span(s)' in text(edits[1])+ # #1287: a successful native edit keeps its announced diff, so+ # its completion carries status only (content would replace it).+ updates = [e['params']['update'] for e in turn if e.get('method') == 'session/update']+ done = [u for u in updates if u.get('sessionUpdate') == 'tool_call_update' and u.get('status') == 'completed' and not u.get('content')]+ assert done, updates+ announced = {u['toolCallId']: u for u in updates if u.get('sessionUpdate') == 'tool_call'}+ diffs = [c for c in announced[done[-1]['toolCallId']].get('content') or [] if c.get('type') == 'diff']+ assert len(diffs) == 2, announced[done[-1]['toolCallId']]shell = terminal_updates(prompt('Start a job; reject legacy output and kill handles, then read and stop the valid handle.', 4))assert len(shell) == 5, shellassert shell[1]['status'] == shell[2]['status'] == 'failed', shell