feat: claude-brain — управляющий слой, выделен из Лидерры по ADR-020
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Дмитрий
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
File diff suppressed because one or more lines are too long
@@ -0,0 +1,16 @@
|
||||
[batch-reviewer] total in period unreviewed: 67, processing first 67 with concurrency 5
|
||||
[batch-reviewer] progress 5/67 (17.4s)
|
||||
[batch-reviewer] progress 10/67 (28.1s)
|
||||
[batch-reviewer] progress 15/67 (38.2s)
|
||||
[batch-reviewer] progress 20/67 (47.1s)
|
||||
[batch-reviewer] progress 25/67 (56.0s)
|
||||
[batch-reviewer] progress 30/67 (66.1s)
|
||||
[batch-reviewer] progress 35/67 (76.6s)
|
||||
[batch-reviewer] progress 40/67 (86.6s)
|
||||
[batch-reviewer] progress 45/67 (97.0s)
|
||||
[batch-reviewer] progress 50/67 (106.3s)
|
||||
[batch-reviewer] progress 55/67 (116.3s)
|
||||
[batch-reviewer] progress 60/67 (125.8s)
|
||||
[batch-reviewer] progress 65/67 (134.6s)
|
||||
[batch-reviewer] progress 67/67 (140.7s)
|
||||
[batch-reviewer] done: 67 reviewed, 0 errors, 140.7s wall-clock
|
||||
@@ -0,0 +1,200 @@
|
||||
{
|
||||
"path_type": {
|
||||
"improvised": 65,
|
||||
"regulated": 2
|
||||
},
|
||||
"node_chosen_top": [
|
||||
[
|
||||
"direct",
|
||||
64
|
||||
],
|
||||
[
|
||||
"superpowers:using-git-worktrees",
|
||||
1
|
||||
],
|
||||
[
|
||||
"subagent-driven-development",
|
||||
1
|
||||
],
|
||||
[
|
||||
"superpowers:brainstorming",
|
||||
1
|
||||
]
|
||||
],
|
||||
"recommended_node": [
|
||||
[
|
||||
"null",
|
||||
60
|
||||
],
|
||||
[
|
||||
"#37",
|
||||
4
|
||||
],
|
||||
[
|
||||
"#18",
|
||||
1
|
||||
],
|
||||
[
|
||||
"#25",
|
||||
1
|
||||
],
|
||||
[
|
||||
"#11",
|
||||
1
|
||||
]
|
||||
],
|
||||
"sources": {
|
||||
"prefilter": 23,
|
||||
"regex": 10,
|
||||
"prefilter_inherited": 3,
|
||||
"llm": 29,
|
||||
"cache": 2
|
||||
},
|
||||
"perCls": {
|
||||
"other": {
|
||||
"total": 37,
|
||||
"trigger_matched": 4,
|
||||
"via_skill": 3
|
||||
},
|
||||
"release": {
|
||||
"total": 7,
|
||||
"trigger_matched": 0,
|
||||
"via_skill": 0
|
||||
},
|
||||
"question": {
|
||||
"total": 13,
|
||||
"trigger_matched": 0,
|
||||
"via_skill": 0
|
||||
},
|
||||
"monitoring": {
|
||||
"total": 7,
|
||||
"trigger_matched": 0,
|
||||
"via_skill": 0
|
||||
},
|
||||
"planning": {
|
||||
"total": 1,
|
||||
"trigger_matched": 0,
|
||||
"via_skill": 0
|
||||
},
|
||||
"bugfix": {
|
||||
"total": 1,
|
||||
"trigger_matched": 0,
|
||||
"via_skill": 0
|
||||
},
|
||||
"cleanup": {
|
||||
"total": 1,
|
||||
"trigger_matched": 0,
|
||||
"via_skill": 0
|
||||
}
|
||||
},
|
||||
"outcomesReviewed": {
|
||||
"soft_success": 33,
|
||||
"success": 16,
|
||||
"rework": 13,
|
||||
"blocked": 5
|
||||
},
|
||||
"groupSummary": {
|
||||
"skill_used": {
|
||||
"total": 3,
|
||||
"outcomes": {
|
||||
"success": 1,
|
||||
"blocked": 1,
|
||||
"soft_success": 1
|
||||
},
|
||||
"rework_rate": "0.0%"
|
||||
},
|
||||
"direct_no_rec": {
|
||||
"total": 58,
|
||||
"outcomes": {
|
||||
"soft_success": 29,
|
||||
"success": 15,
|
||||
"rework": 10,
|
||||
"blocked": 4
|
||||
},
|
||||
"rework_rate": "17.2%"
|
||||
},
|
||||
"direct_ignored_rec": {
|
||||
"total": 6,
|
||||
"outcomes": {
|
||||
"soft_success": 3,
|
||||
"rework": 3
|
||||
},
|
||||
"rework_rate": "50.0%"
|
||||
}
|
||||
},
|
||||
"reviewerVerdicts": {
|
||||
"node_quality": {
|
||||
"disputable": 31,
|
||||
"correct": 25,
|
||||
"wrong_node": 11
|
||||
},
|
||||
"self_assessment_accuracy": {
|
||||
"accurate": 38,
|
||||
"no_self_assessment": 29
|
||||
}
|
||||
},
|
||||
"gap": {
|
||||
"total": 6,
|
||||
"rework": 3,
|
||||
"cases": [
|
||||
{
|
||||
"time": "2026-05-27T04:05:21.242Z",
|
||||
"task": "b11f6b8d",
|
||||
"rec": "#37",
|
||||
"outcome": "soft_success",
|
||||
"node_quality": "disputable",
|
||||
"reasoning": "Task was a background completion notification with trivial processing (1 Read, 1 TodoWrite). Direct handling is reasonable despite classifier recommending #37 for deploy/release, since no actual deployment work was needed. Agent's self-assessment honestly flags the unexplained divergence."
|
||||
},
|
||||
{
|
||||
"time": "2026-05-27T04:09:31.149Z",
|
||||
"task": "b11f6b8d",
|
||||
"rec": "#37",
|
||||
"outcome": "rework",
|
||||
"node_quality": "wrong_node",
|
||||
"reasoning": "Classifier recommended #37 but agent went direct without override justification. Self-assessment honestly flags this with low confidence (0.15) and identifies the missing override step. Agent should have either invoked #37 or documented an explicit override."
|
||||
},
|
||||
{
|
||||
"time": "2026-05-27T05:32:27.040Z",
|
||||
"task": "b11f6b8d",
|
||||
"rec": "#18",
|
||||
"outcome": "rework",
|
||||
"node_quality": "wrong_node",
|
||||
"reasoning": "Classifier recommended node #18 for the task-notification, but the agent went direct without invoking it, risking loss of background task result handling. The agent's self-assessment honestly acknowledges this deviation and its consequences."
|
||||
},
|
||||
{
|
||||
"time": "2026-05-27T07:16:20.117Z",
|
||||
"task": "0ade4c82",
|
||||
"rec": "#25",
|
||||
"outcome": "rework",
|
||||
"node_quality": "wrong_node",
|
||||
"reasoning": "Agent went direct on an ambiguous short prompt ('долго ждешь проверь') despite classifier recommending #25, then hit a PowerShell error during execution. Self-assessment honestly recognizes the routing mistake and need for clarification."
|
||||
},
|
||||
{
|
||||
"time": "2026-05-27T08:14:25.441Z",
|
||||
"task": "0ade4c82",
|
||||
"rec": "#37",
|
||||
"outcome": "soft_success",
|
||||
"node_quality": "disputable",
|
||||
"reasoning": "Classifier recommended #37 with low confidence (0.5) after parse failure, but agent chose direct handling for a background task notification. Self-assessment honestly flags the deviation and uncertainty. Direct response is plausible for a monitoring-type notification, though #37 may have been more appropriate."
|
||||
},
|
||||
{
|
||||
"time": "2026-05-27T12:31:06.105Z",
|
||||
"task": "0ade4c82",
|
||||
"rec": "#11",
|
||||
"outcome": "soft_success",
|
||||
"node_quality": "disputable",
|
||||
"reasoning": "Classifier recommended #11 (cleanup) with low-ish confidence and a parse_null LLM error, but agent chose 'direct' path. With only a 55-char prompt and no tool calls/files touched, a direct response was reasonable for a cleanup-type task. Self-assessment is pending so honesty cannot be evaluated."
|
||||
}
|
||||
]
|
||||
},
|
||||
"cost": {
|
||||
"main_in": 1313,
|
||||
"main_out": 453422,
|
||||
"cache_read": 159238917,
|
||||
"cache_create": 8548887,
|
||||
"classifier_in": 3373,
|
||||
"classifier_out": 23382,
|
||||
"total_iter": 505,
|
||||
"total_tool_calls": 181
|
||||
}
|
||||
}
|
||||
Reference in New Issue
Block a user