mixdog 0.9.41 → 0.9.44
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/package.json +3 -2
- package/scripts/agent-tag-reuse-smoke.mjs +124 -10
- package/scripts/ansi-color-capability-test.mjs +90 -0
- package/scripts/anthropic-oauth-refresh-race-test.mjs +59 -0
- package/scripts/arg-guard-test.mjs +16 -0
- package/scripts/compact-pressure-test.mjs +256 -1
- package/scripts/generate-runtime-manifest.mjs +39 -22
- package/scripts/grok-oauth-refresh-race-test.mjs +198 -0
- package/scripts/internal-comms-bench.mjs +0 -1
- package/scripts/internal-comms-smoke.mjs +38 -39
- package/scripts/internal-tools-normalization-test.mjs +52 -0
- package/scripts/maintenance-default-routes-test.mjs +164 -0
- package/scripts/max-output-recovery-test.mjs +49 -0
- package/scripts/mcp-client-normalization-test.mjs +45 -0
- package/scripts/openai-oauth-ws-1006-retry-test.mjs +123 -0
- package/scripts/provider-toolcall-test.mjs +23 -0
- package/scripts/result-classification-test.mjs +75 -0
- package/scripts/routing-corpus.mjs +1 -1
- package/scripts/shell-jobs-windows-hide-test.mjs +164 -0
- package/scripts/smoke.mjs +1 -1
- package/scripts/submit-commandbusy-race-test.mjs +99 -7
- package/scripts/tui-transcript-jitter-harness-entry.jsx +302 -0
- package/scripts/tui-transcript-jitter-harness.mjs +58 -0
- package/scripts/tui-transcript-perf-test.mjs +56 -1
- package/src/agents/reviewer/AGENT.md +5 -3
- package/src/rules/lead/lead-brief.md +5 -4
- package/src/runtime/agent/orchestrator/agent-runtime/agent-dispatch.mjs +40 -3
- package/src/runtime/agent/orchestrator/config.mjs +29 -14
- package/src/runtime/agent/orchestrator/internal-tools.mjs +16 -3
- package/src/runtime/agent/orchestrator/mcp/client.mjs +17 -1
- package/src/runtime/agent/orchestrator/providers/anthropic-oauth-credentials.mjs +49 -6
- package/src/runtime/agent/orchestrator/providers/anthropic-sse.mjs +14 -0
- package/src/runtime/agent/orchestrator/providers/grok-oauth.mjs +46 -21
- package/src/runtime/agent/orchestrator/providers/media-normalization.mjs +59 -2
- package/src/runtime/agent/orchestrator/providers/retry-classifier.mjs +5 -1
- package/src/runtime/agent/orchestrator/session/agent-loop.mjs +18 -0
- package/src/runtime/agent/orchestrator/session/context-utils.mjs +140 -81
- package/src/runtime/agent/orchestrator/session/loop/compact-policy.mjs +23 -5
- package/src/runtime/agent/orchestrator/session/loop/termination.mjs +3 -0
- package/src/runtime/agent/orchestrator/session/loop/tool-exec.mjs +12 -3
- package/src/runtime/agent/orchestrator/session/manager/ask-session.mjs +4 -2
- package/src/runtime/agent/orchestrator/session/result-classification.mjs +4 -1
- package/src/runtime/agent/orchestrator/session/tool-batch.mjs +7 -2
- package/src/runtime/agent/orchestrator/session/tool-envelope.mjs +8 -6
- package/src/runtime/agent/orchestrator/tools/builtin/arg-guard.mjs +15 -3
- package/src/runtime/agent/orchestrator/tools/builtin/shell-analysis.mjs +1 -1
- package/src/runtime/agent/orchestrator/tools/builtin/shell-jobs.mjs +7 -5
- package/src/runtime/memory/data/runtime-manifest.json +11 -11
- package/src/runtime/memory/lib/runtime-fetcher.mjs +1 -2
- package/src/runtime/memory/tool-defs.mjs +4 -3
- package/src/runtime/shared/tool-surface.mjs +1 -2
- package/src/session-runtime/context-status.mjs +8 -2
- package/src/standalone/agent-tool/render.mjs +2 -0
- package/src/standalone/agent-tool.mjs +267 -72
- package/src/standalone/explore-tool.mjs +17 -16
- package/src/tui/app/provider-setup-picker.mjs +1 -0
- package/src/tui/app/use-transcript-window.mjs +5 -1
- package/src/tui/components/StatusLine.jsx +11 -32
- package/src/tui/dist/index.mjs +257 -106
- package/src/tui/engine/session-api.mjs +3 -0
- package/src/tui/engine/session-flow.mjs +19 -8
- package/src/tui/engine/turn.mjs +24 -2
- package/src/tui/engine.mjs +0 -1
- package/src/tui/index.jsx +3 -2
- package/src/tui/markdown/streaming-markdown.mjs +35 -9
- package/src/tui/statusline-ansi-bridge.mjs +17 -7
- package/src/ui/ansi.mjs +85 -5
- package/src/ui/statusline-agents.mjs +0 -8
- package/src/ui/statusline-format.mjs +7 -7
- package/src/ui/statusline.mjs +2 -4
- package/src/workflows/default/WORKFLOW.md +29 -20
- package/src/workflows/solo-review/WORKFLOW.md +47 -0
- package/src/workflows/bench/WORKFLOW.md +0 -60
|
@@ -1,60 +0,0 @@
|
|
|
1
|
-
---
|
|
2
|
-
id: bench
|
|
3
|
-
name: Bench
|
|
4
|
-
description: "Autonomous headless benchmark workflow with directed execution and parallel cross-verification."
|
|
5
|
-
hidden: true
|
|
6
|
-
agents: worker, heavy-worker, reviewer, debugger, maintainer
|
|
7
|
-
---
|
|
8
|
-
|
|
9
|
-
# Bench
|
|
10
|
-
|
|
11
|
-
Autonomous headless run: no user exists. Never ask questions, propose plans
|
|
12
|
-
for approval, or end the turn waiting — decide and proceed. Standing
|
|
13
|
-
pre-approval covers every action: edits, state mutation, delegation, builds.
|
|
14
|
-
Loop until the task is verified complete or provably blocked.
|
|
15
|
-
|
|
16
|
-
The pipeline has exactly three stages. Lead directs and verifies it, but NEVER
|
|
17
|
-
executes the work itself: no edits, answers, or artifact generation, regardless
|
|
18
|
-
of task size, perceived simplicity, or number of steps.
|
|
19
|
-
|
|
20
|
-
## 1. Lead analysis and direction
|
|
21
|
-
Lead performs real analysis first on EVERY task: task structure, required
|
|
22
|
-
deliverables, boundaries, the task's own verification means, and solution
|
|
23
|
-
direction. Convert it into clear execution direction and delegation briefs
|
|
24
|
-
following the Lead brief contract.
|
|
25
|
-
|
|
26
|
-
Hand the direction and briefs to executors. Analysis is not permission for
|
|
27
|
-
Lead to implement any portion of the result: all edits, answer production, and
|
|
28
|
-
artifact generation belong to executor sessions.
|
|
29
|
-
|
|
30
|
-
## 2. Executors perform the work
|
|
31
|
-
Use Worker sessions, or Heavy Worker sessions for broad scopes, to execute the
|
|
32
|
-
directed work. Multiple workers MAY fan out in parallel, one per independent
|
|
33
|
-
scope. Keep ownership clear for each assigned deliverable.
|
|
34
|
-
|
|
35
|
-
Executors build the answer or artifact and perform basic does-it-run checks.
|
|
36
|
-
They are not a verification role and do not own acceptance, requirements
|
|
37
|
-
judgment, or final correctness. Each executor reports its result to Lead.
|
|
38
|
-
|
|
39
|
-
## 3. Lead and Reviewer verify
|
|
40
|
-
|
|
41
|
-
When executors report, Lead and Reviewer start cross-verification IN PARALLEL,
|
|
42
|
-
never serially. Lead personally runs the task's own verification means, such
|
|
43
|
-
as its test suite, grader script, simulator, or self-checkable output.
|
|
44
|
-
Reviewer independently judges the result against every task requirement,
|
|
45
|
-
including intent, correctness, boundaries, and deliverables.
|
|
46
|
-
|
|
47
|
-
Judgment-type answers with no direct ground truth still require parallel
|
|
48
|
-
cross-verification, even for a single small deliverable. Lead evaluates against
|
|
49
|
-
available criteria while Reviewer independently judges requirements;
|
|
50
|
-
perceived simplicity is no exemption.
|
|
51
|
-
|
|
52
|
-
Merge any finding into a clear fix delta and return it to the SAME executor
|
|
53
|
-
session owning the affected scope. The executor applies it and reports again.
|
|
54
|
-
Lead and Reviewer repeat the same parallel cross-verification until both legs
|
|
55
|
-
are clean; Lead never fixes or completes the work directly.
|
|
56
|
-
|
|
57
|
-
Before declaring completion, Lead personally runs the task-required
|
|
58
|
-
verification one final time and checks that every deliverable is present.
|
|
59
|
-
Never end with a question, approval request, or proposed next step. Continue
|
|
60
|
-
until verified complete or stop only when completion is provably blocked.
|