Skip to content

graff — a fast agentic coding harness in Zig: multi-provider, MCP, workflows, DGM evolution loop, TS/Python SDKs

Python

Prepare v0.302.42 with current main and release fixes

justrach7b61d4f3d00e
diff --cc CHANGELOG.md
index 7c978fcc,9e295551..8f48107d
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@@ -10,32 -10,6 +10,40 @@@ The release workflow uses a tag's secti
hand-written `docs/releases/<tag>.md` wins if present), so keeping this file
current is part of cutting a release.
++## v0.302.42
++
++- Keeps reasoning-only output-limit replies out of subsequent requests and saved history while preserving partial answers and valid tool-call reasoning.
++- Shows each subagent's resolved worker model in ACP activity, including fallback selections, without breaking older activity records.
++- Retains large-session loading, bounded MCP task deadlines, streamed progress, oversized tool results and parked-command outcomes.
++- Reports compaction failures without ending the session and bounds repeated runaway replies.
++- Integrates current main with the existing release fixes; release builds verify cached artifact restoration before packaging.
++
+## v0.302.41
+
+- Adds native subscription browser sign-in with shared CLI, REPL and ACP credential recovery; metered-key access remains separate.
+- Preserves independent MCP structured results, displays resource links and embedded text, and answers inbound pings during initialization and tool waits.
+- Gives loopback inference request-local deadline allowances without extending unrelated remote requests.
+- Adds repeatable isolated native-CLI fixture comparisons with explicit routing checks and separately labeled task subsets.
+- Retains all changes from v0.302.40 and current main.
+
+## v0.302.40
+
+- Integrates the pending stream recovery, claim inventory, workspace skills, stored conversation, MCP configuration, companion update and typed ACP sign-in recovery changes.
+- Preserves partial replies and unread tool results, secret-free configuration listings, and atomic installer replacement.
+- Hosted compaction remains explicitly opt-in; existing local-summary and provider-owned compaction defaults are unchanged.
+- Includes the unattended-run recovery and one-shot result contracts from the unpublished v0.302.39 draft.
+
+See `docs/releases/v0.302.40.md` for the release summary.
+
+## v0.302.39
+
+### Unattended runs
+
+- Output limits continue with bounded recovery and incremental file writes; incomplete tool calls are never executed.
+- Empty replies retry with cancellable, jittered backoff. Account-limit and sign-in failures stop with distinct exit codes instead of wasting retries.
+- `-p --output-format stream-json` emits a session handle at startup and a final result with status, usage, cost completeness and checkpoint status. Forced budget landings report partial work, not completion.
+- Buffered generation gets request-local silence allowances, and raw MCP startup exceptions are debug-only. Recovery still uses session checkpoints, not exactly-once journal replay.
+
## v0.302.38
### Windows
diff --cc docs/releases/v0.302.42.md
index 00000000,00000000..fb739e37
new file mode 100644
--- /dev/null
+++ b/docs/releases/v0.302.42.md
@@@ -1,0 -1,0 +1,8 @@@
++# v0.302.42
++
++- Reasoning-only replies that reach the output limit no longer leave empty assistant messages in later requests or saved history. Valid partial answers and tool-call reasoning remain intact.
++- ACP subagent activity shows the resolved worker model before child activity starts, including fallback selections. Older activity records remain readable.
++- Large saved sessions load correctly, MCP task deadlines match bounded execution, and streamed progress and oversized tool results are preserved.
++- Parked commands retain their outcomes across continuation, compaction failures are reported without ending the session, and repeated runaway replies have bounded recovery.
++
++This release integrates current main with the existing release fixes. Release builds verify cached artifact restoration before packaging.
diff --cc scripts/eval/tier1-manifest.json
index 77167433,3f17b5d5..00d086a8
--- a/scripts/eval/tier1-manifest.json
+++ b/scripts/eval/tier1-manifest.json
@@@ -14,7 -14,7 +14,7 @@@
"src/main.zig",
"src/repl.zig"
],
- "test_count_baseline": 3373,
- "test_count_baseline": 3254,
++ "test_count_baseline": 3389,
"test_count_slack": 25,
"required_invariants": [
{
diff --cc src/agent_length_cutoff.zig
index 16e131b3,4573c2d2..c6c8b7d6
--- a/src/agent_length_cutoff.zig
+++ b/src/agent_length_cutoff.zig
@@@ -251,3 -285,51 +285,54 @@@ test "ReplySize: text past the limit st
var raw: ReplySize = .{};
try std.testing.expect(!raw.feed(&agent, reason_ev, 0));
}
+
+ test "empty length reply cleanup keeps recovery notes and valid prior tool reasoning on both chat routes" {
+ var arena_state = std.heap.ArenaAllocator.init(std.testing.allocator);
+ defer arena_state.deinit();
+ const a = arena_state.allocator();
+ for ([_][]const u8{ "deepseek", "codegraff" }) |provider| {
+ for ([_][]const u8{ "null", "\"\"", "\" \\n\"" }) |content| {
+ for ([_]bool{ false, true }) |retries_spent| {
+ var agent: Agent = .{
+ .gpa = std.testing.allocator,
+ .arena = a,
+ .io = std.testing.io,
+ .client = undefined,
+ .provider = .{ .id = provider, .kind = .openai, .auth = .bearer, .url = "", .api_key = "", .model = "deepseek-flash", .context = 100_000 },
+ .messages = .init(a),
+ .sub = true,
+ .label = "test",
+ .out = null,
+ };
+ try agent.messages.append(try messages.textMessage(a, "user", "Write the artifact"));
+ const prior = try parse(a, "{\"role\":\"assistant\",\"content\":null,\"reasoning_content\":\"keep tool reasoning\",\"tool_calls\":[{\"id\":\"c\",\"type\":\"function\",\"function\":{\"name\":\"read_file\",\"arguments\":\"{}\"}}]}");
+ try agent.messages.append(prior);
+ try agent.messages.append(try messages.toolResultMessage(a, .openai, "c", "saved work", false));
+ const start = agent.messages.items.len;
+ const root = (try parse(a, try std.fmt.allocPrint(a, "{{\"choices\":[{{\"finish_reason\":\"length\",\"message\":{{\"role\":\"assistant\",\"content\":{s},\"reasoning_content\":\"unfinished thought\",\"tool_calls\":[]}}}}]}}", .{content}))).object;
- _ = try agent.stepOpenAI(root);
++ // Exercise cleanup independently of the stepper's early exclusion.
++ try agent.messages.append(root.get("choices").?.array.items[0].object.get("message").?);
+ // Exhaustion used to leave an invalid assistant in saved history.
+ if (retries_spent) agent.empty_completion_retries = @import("agent_empty_completion.zig").max_consecutive;
- try std.testing.expectEqual(!retries_spent, try @import("agent_empty_completion.zig").handle(&agent, "", start));
++ if (retries_spent) {
++ try std.testing.expectError(error.EmptyCompletion, @import("agent_empty_completion.zig").handle(&agent, "", start));
++ } else try std.testing.expect(try @import("agent_empty_completion.zig").handle(&agent, "", start));
+ discardEmptyReply(&agent, root, start);
+ try std.testing.expectEqual(start + @as(usize, if (retries_spent) 0 else 1), agent.messages.items.len);
+ const body = try agent.buildBody(null, false, false, false);
+ defer std.testing.allocator.free(body);
+ try std.testing.expect(std.mem.indexOf(u8, body, "unfinished thought") == null);
+ try std.testing.expect(std.mem.indexOf(u8, body, "keep tool reasoning") != null);
+ try std.testing.expect(std.mem.indexOf(u8, body, "\"tool_call_id\":\"c\"") != null);
+ if (!retries_spent) try std.testing.expect(std.mem.indexOf(u8, body, "Think briefly") != null);
+ // Cleanup is not a blanket reasoning filter or a partial-answer rewind.
+ try agent.messages.append(prior);
+ discardEmptyReply(&agent, root, agent.messages.items.len - 1);
+ try std.testing.expectEqualStrings("keep tool reasoning", agent.messages.items[agent.messages.items.len - 1].object.get("reasoning_content").?.string);
+ try agent.messages.append(try messages.textMessage(a, "assistant", "partial document"));
+ discardEmptyReply(&agent, root, agent.messages.items.len - 1);
+ try std.testing.expectEqualStrings("partial document", agent.messages.items[agent.messages.items.len - 1].object.get("content").?.string);
+ }
+ }
+ }
+ }
diff --cc src/agent_turn.zig
index 7ad45209,31df5ed1..09666286
--- a/src/agent_turn.zig
+++ b/src/agent_turn.zig
@@@ -16,9 -16,7 +16,9 @@@ pub fn run(self: *Agent, root_turn_prep
defer if (!self.sub) @import("peer_idle.zig").noteTurnEnd();
var pending_work = empty_completion.PendingWork.begin(self);
var bounced_answer: empty_completion.BounceAnswer = .{};
+ var length: @import("agent_length.zig").State = .{};
+ self.empty_completion_retries = 0;
- self.length_cutoff_used = false; // agent_length_cutoff.zig: once per turn
+ self.length_cutoffs = 0; // agent_length_cutoff.zig: rewinds this turn
self.completed = null;
self.yielded = null;
self.read_miss.reset();
@@@ -81,7 -79,12 +81,12 @@@
if (try @import("agent_length_cutoff.zig").recoverRunaway(self, err, hist_len)) continue;
return review_deadline.finish(try @import("agent_model_loop.zig").finishError(self, err));
};
+ // request() may compact history; capture this reply's actual boundary.
+ // Keep reasoning available to recovery, then exclude non-replayable
+ // empty length replies on every exit before another call or save.
+ const reply_start = self.messages.items.len;
+ defer @import("agent_length_cutoff.zig").discardEmptyReply(self, root, reply_start);
- const done = try @import("agent_steps.zig").stepForWire(self, root);
+ const done = try length.step(self, root);
if (done) |final_text| {
if (self.yielded != null) return review_deadline.finish(final_text);
if (@import("tool_call_repair.zig").endsTurn(self, final_text)) return review_deadline.finish(final_text); // final: no retry/nudge
diff --cc src/cli.zig
index 2510a2e8,1c1b0d93..c27d3860
--- a/src/cli.zig
+++ b/src/cli.zig
@@@ -24,22 -24,6 +24,27 @@@ const update_target = @import("update_t
pub const changelog_text =
\\What's new
\\──────────
++ \\0.302.42
++ \\ • output-limit recovery keeps later requests and saved history valid
++ \\ • ACP subagent activity shows the resolved worker model
++ \\ • current main plus session, MCP, compaction and parked-command fixes
++ \\
+ \\0.302.41
+ \\ • native subscription sign-in across terminal and ACP clients
+ \\ • MCP structured results, resource content and inbound ping handling
+ \\ • request-local loopback deadlines and repeatable fixture comparisons
+ \\
+ \\0.302.40
+ \\ • stream recovery preserves partial replies across ACP and resume
+ \\ • workspace skills in headless runs; actionable claim inventory
+ \\ • MCP off-lists and JSON configuration listing; atomic companion updates
+ \\ • stored conversation continuation and typed sign-in recovery
+ \\
+ \\0.302.39
+ \\ • unattended runs recover from output limits and empty replies
+ \\ • one-shot stream-json reports a resume handle, outcome and usage
+ \\ • exhausted budgets and account limits stop with distinct exit codes
+ \\
\\0.302.38
\\ • Windows: faster start, and the companion is found on PATH
\\