From 59d1466c232f54a52e19dbfd31573bec6c619c51 Mon Sep 17 00:00:00 2001 From: DevMando Date: Thu, 10 Sep 2026 20:24:29 -0700 Subject: [PATCH 1/2] Docs: record the 0.15.0 work missing from the changelog An audit of the 0.15.0 section against the pull requests actually merged for this generation found four capabilities shipped but never written down, and a Validation paragraph that described the suite without stating its result. Adds entries for image delivery to vision-capable models, image accounting in the context estimate, host-supplied agent tools, and on-demand compaction. Replaces the vague Validation opening with the real figure and records what the new command-output-sink coverage actually pins. Documentation only; no code or behaviour changes. --- docs/CHANGELOG.md | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-) diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md index 094cf31..b3443d2 100644 --- a/docs/CHANGELOG.md +++ b/docs/CHANGELOG.md @@ -34,6 +34,23 @@ ships .NET 10 and .NET 8 builds in the same package, with nothing for users to c remains in front of every revised plan and file change. ### Added +- **Images can be delivered to models that can actually see them.** A host can hand a captured + image to the assistant, and whether a model accepts image input is now detected rather than + assumed. A text-only model is told plainly that a picture could not be examined, instead of being + handed bytes it would silently discard and then answer around. Screenshots taken during a plan are + delivered to the step that asked for them, rather than arriving with no context attached. +- **Images are counted toward the context estimate.** A picture occupies real space in a model's + window, and leaving it out of the running total meant the estimate drifted low exactly when it + mattered. Image evidence gathered for a retry is dropped when it no longer fits, so a retry cannot + push a conversation over the limit by carrying attachments it can no longer afford. +- **Hosts can supply their own agent tools.** An application embedding MandoCode can register + functions of its own alongside the built-in ones, so capabilities that only the host can provide — + its own windows, panes, or other sessions — are available to the model without changing the + engine. Tool registration survives model switches and settings changes, so a host attaches its + tools once. +- **Conversations can be compacted on demand.** A long session can be condensed deliberately rather + than waiting for the context window to force the issue, keeping the thread of the work while + recovering the room to continue it. - **Hosts can watch shell commands run.** A host may attach an optional command output sink and receive each `execute_command` as it happens — the command and its working directory when it starts, every line of output as it arrives, and the exit code or the reason it was killed. The @@ -97,11 +114,15 @@ ships .NET 10 and .NET 8 builds in the same package, with nothing for users to c and navigation state, not settled page content. ### Validation -The engine test suite runs on both .NET 8 and .NET 10. Coverage includes workflow topology, +733/733 tests passing on both .NET 8 and .NET 10. Coverage includes workflow topology, checkpoint storage, resume context, retry behavior, plan revision, semantic step outcomes, proposal handoff, input handling, scoped directory references, bounded tool results, and version labels. Desktop adds host-level coverage, and the planner was also exercised through real CLI and Desktop -sessions, including forced process termination between steps. +sessions, including forced process termination between steps. New coverage for the command output +sink pins the properties a host display depends on: the full lifecycle is reported, a non-zero exit +reads as a failure rather than a kill, a bare `cd` produces no report at all, output past the +model's truncation cap still reaches the sink, and a sink that throws on every call leaves the +command running and reporting normally. ### Internal - **Moved AI orchestration from Semantic Kernel to Microsoft Agent Framework.** The chat layer now From 41e1c58fbe21c3496a1243527966398afc341739 Mon Sep 17 00:00:00 2001 From: DevMando Date: Thu, 10 Sep 2026 20:33:19 -0700 Subject: [PATCH 2/2] Docs: record the planner execution changes from #87, #88 and #89 A second pass over the 0.15.0 section against the merged pull requests found three more gaps, all in the planner. The biggest is that per-step verification by a second model was removed: #88 moved completion onto the executor's own evidence with acceptance criteria per step and a closing test-and-repair phase, and #89 deleted the strict path instead of leaving it behind a flag. That halves the model calls in a plan run and changes how a step is judged done, and the changelog said neither. #87's fixes were also unrecorded: the stall when a host stops reading progress, and retries reusing gathered evidence rather than rerunning implementation. --- docs/CHANGELOG.md | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md index b3443d2..7c55a00 100644 --- a/docs/CHANGELOG.md +++ b/docs/CHANGELOG.md @@ -69,6 +69,19 @@ ships .NET 10 and .NET 8 builds in the same package, with nothing for users to c skipped, used as the starting point for a replacement suffix, or used to cancel the plan. ### Changed +- **A step no longer needs a second model call to be judged complete.** Every plan step used to be + gated by a separate model that decided whether the step was really done. That doubled the model + calls in a plan run and, in practice, blocked correct work — rejecting a discovery step for + finding no build system in a genuinely empty folder, for example. A step now completes on the + executor's own evidence, measured against acceptance criteria the step carries itself, and the + plan ends with an explicit test-and-repair phase that has to produce real, observable results. + Plans are faster and cheaper to run, and a plan can no longer finish green on work that was + never actually exercised. +- **Strict plan verification is gone rather than kept behind a flag.** The old per-step verifier + survived one release behind an off-by-default `strictPlanVerification` setting. Two execution + modes meant two sets of failure semantics and two resume paths to keep correct, for a flag nobody + was expected to turn on — so executor-owned completion is now the only mode and the setting is no + longer read. - **Automatic planning now follows task shape instead of message length.** MandoCode plans high-confidence work such as three-step checklists, cross-cutting changes, and requests with multiple deliverables. Questions, research, explanations, and narrow edits remain direct, and @@ -96,6 +109,14 @@ ships .NET 10 and .NET 8 builds in the same package, with nothing for users to c exists and when it goes away. ### Fixed +- **Long plans no longer stall when the host stops reading progress.** If the interface exited or + errored while a plan was running, progress reporting could block and the run would hang with no + way forward. Blocked work is now released. +- **A failed verification response no longer discards completed implementation work.** A retry + reuses the evidence already gathered — confirmed still valid by file hash — instead of rerunning + the implementation and re-reading unchanged files, so a plan makes progress toward finishing + rather than repeatedly rechecking itself. File-read results also get more room before clipping, + and checkpoints preserve partial file changes and pending verification. - **Session token totals now use provider-reported usage only.** The previous display added rough character-based estimates for file and tool payloads to prompt counts that already included those payloads. Reads, search, and attachments no longer inflate the visible total or trigger