triflux 4.2.6 → 4.2.7
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 +2 -1
- package/skills/tfx-workspace/evals/evals.json +0 -79
- package/skills/tfx-workspace/iteration-1/benchmark.json +0 -162
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/eval_metadata.json +0 -11
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/old_skill/grading.json +0 -9
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/old_skill/outputs/analysis.md +0 -154
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/with_skill/grading.json +0 -9
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/with_skill/outputs/analysis.md +0 -126
- package/skills/tfx-workspace/iteration-1/codex-gemini-remap/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/eval_metadata.json +0 -11
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/old_skill/grading.json +0 -9
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/old_skill/outputs/analysis.md +0 -119
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/with_skill/grading.json +0 -9
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/with_skill/outputs/analysis.md +0 -115
- package/skills/tfx-workspace/iteration-1/doctor-diagnosis/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/eval_metadata.json +0 -10
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/old_skill/grading.json +0 -8
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/old_skill/outputs/analysis.md +0 -86
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/with_skill/grading.json +0 -8
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/with_skill/outputs/analysis.md +0 -81
- package/skills/tfx-workspace/iteration-1/hub-start-sequence/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/multi-team-creation/eval_metadata.json +0 -12
- package/skills/tfx-workspace/iteration-1/multi-team-creation/old_skill/grading.json +0 -10
- package/skills/tfx-workspace/iteration-1/multi-team-creation/old_skill/outputs/analysis.md +0 -316
- package/skills/tfx-workspace/iteration-1/multi-team-creation/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/multi-team-creation/with_skill/grading.json +0 -10
- package/skills/tfx-workspace/iteration-1/multi-team-creation/with_skill/outputs/analysis.md +0 -352
- package/skills/tfx-workspace/iteration-1/multi-team-creation/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/review.html +0 -1325
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/eval_metadata.json +0 -12
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/old_skill/grading.json +0 -10
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/old_skill/outputs/analysis.md +0 -97
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/with_skill/grading.json +0 -10
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/with_skill/outputs/analysis.md +0 -94
- package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/eval_metadata.json +0 -12
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/old_skill/grading.json +0 -10
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/old_skill/outputs/analysis.md +0 -209
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/with_skill/grading.json +0 -10
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/with_skill/outputs/analysis.md +0 -193
- package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-2/benchmark.json +0 -62
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/eval_metadata.json +0 -13
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/old_skill/grading.json +0 -11
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/old_skill/outputs/analysis.md +0 -382
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/old_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/with_skill/grading.json +0 -11
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/with_skill/outputs/analysis.md +0 -333
- package/skills/tfx-workspace/iteration-2/multi-team-creation-refactored/with_skill/timing.json +0 -5
- package/skills/tfx-workspace/iteration-2/review.html +0 -1325
- package/skills/tfx-workspace/skill-snapshot/tfx-auto/SKILL.md +0 -217
- package/skills/tfx-workspace/skill-snapshot/tfx-auto-codex/SKILL.md +0 -77
- package/skills/tfx-workspace/skill-snapshot/tfx-codex/SKILL.md +0 -65
- package/skills/tfx-workspace/skill-snapshot/tfx-doctor/SKILL.md +0 -94
- package/skills/tfx-workspace/skill-snapshot/tfx-gemini/SKILL.md +0 -82
- package/skills/tfx-workspace/skill-snapshot/tfx-hub/SKILL.md +0 -133
- package/skills/tfx-workspace/skill-snapshot/tfx-multi/SKILL.md +0 -426
- package/skills/tfx-workspace/skill-snapshot/tfx-setup/SKILL.md +0 -101
|
@@ -1,12 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"eval_id": 1,
|
|
3
|
-
"eval_name": "routing-implement-shortcut",
|
|
4
|
-
"prompt": "/implement JWT 인증 미들웨어 추가해줘",
|
|
5
|
-
"assertions": [
|
|
6
|
-
"Routes to 'executor' agent (not architect, not analyst)",
|
|
7
|
-
"Uses 'implement' MCP profile",
|
|
8
|
-
"Generates correct tfx-route.sh command syntax",
|
|
9
|
-
"Does NOT trigger triage (single command shortcut)",
|
|
10
|
-
"Does NOT delegate to tfx-multi"
|
|
11
|
-
]
|
|
12
|
-
}
|
|
@@ -1,10 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"expectations": [
|
|
3
|
-
{"text": "Routes to 'executor' agent (not architect, not analyst)", "passed": true, "evidence": "Agent output: 'Agent chosen: executor — From the Codex 직행 shortcut table'"},
|
|
4
|
-
{"text": "Uses 'implement' MCP profile", "passed": true, "evidence": "Agent output: 'MCP profile: implement — Assigned directly by the shortcut table'"},
|
|
5
|
-
{"text": "Generates correct tfx-route.sh command syntax", "passed": true, "evidence": "Command: bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement"},
|
|
6
|
-
{"text": "Does NOT trigger triage (single command shortcut)", "passed": true, "evidence": "Agent output: 'Triage triggered: No. Command shortcut mode explicitly skips triage'"},
|
|
7
|
-
{"text": "Does NOT delegate to tfx-multi", "passed": true, "evidence": "Agent output: 'tfx-multi delegation: No'"}
|
|
8
|
-
],
|
|
9
|
-
"summary": {"passed": 5, "failed": 0, "total": 5, "pass_rate": 1.0}
|
|
10
|
-
}
|
package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/old_skill/outputs/analysis.md
DELETED
|
@@ -1,97 +0,0 @@
|
|
|
1
|
-
# Routing Analysis: `/implement JWT 인증 미들웨어 추가해줘`
|
|
2
|
-
|
|
3
|
-
## Input
|
|
4
|
-
|
|
5
|
-
```
|
|
6
|
-
/implement JWT 인증 미들웨어 추가해줘
|
|
7
|
-
```
|
|
8
|
-
|
|
9
|
-
---
|
|
10
|
-
|
|
11
|
-
## Mode Selected: Command Shortcut (커맨드 숏컷)
|
|
12
|
-
|
|
13
|
-
**Reason:** The user input begins with `/implement`, which is one of the explicitly listed triggers in SKILL.md. The skill's mode table states:
|
|
14
|
-
|
|
15
|
-
> `/implement JWT 추가` → 커맨드 숏컷 | 트리아지: 없음 (즉시 실행)
|
|
16
|
-
|
|
17
|
-
The command word `implement` directly maps to the "Codex 직행" shortcut table. No triage step is performed.
|
|
18
|
-
|
|
19
|
-
---
|
|
20
|
-
|
|
21
|
-
## Agent Selected: `executor`
|
|
22
|
-
|
|
23
|
-
**Reason:** From the "커맨드 숏컷 → Codex 직행" table:
|
|
24
|
-
|
|
25
|
-
| 커맨드 | 에이전트 | MCP |
|
|
26
|
-
|--------|---------|-----|
|
|
27
|
-
| `implement` | executor | implement |
|
|
28
|
-
|
|
29
|
-
The command keyword `implement` maps to the `executor` agent.
|
|
30
|
-
|
|
31
|
-
---
|
|
32
|
-
|
|
33
|
-
## Exact Bash Command Generated
|
|
34
|
-
|
|
35
|
-
```bash
|
|
36
|
-
bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement
|
|
37
|
-
```
|
|
38
|
-
|
|
39
|
-
**Explanation of each argument:**
|
|
40
|
-
- `executor` — agent resolved from the `implement` shortcut
|
|
41
|
-
- `'JWT 인증 미들웨어 추가해줘'` — the prompt is everything after the command keyword
|
|
42
|
-
- `implement` — MCP profile resolved from the shortcut table
|
|
43
|
-
|
|
44
|
-
The full invocation as specified by SKILL.md section "커맨드 숏컷":
|
|
45
|
-
|
|
46
|
-
> `Bash("bash ~/.claude/scripts/tfx-route.sh {에이전트} '{PROMPT}' {MCP}")`
|
|
47
|
-
|
|
48
|
-
So the complete call is:
|
|
49
|
-
|
|
50
|
-
```
|
|
51
|
-
Bash("bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement")
|
|
52
|
-
```
|
|
53
|
-
|
|
54
|
-
`run_in_background` is not specified for single shortcut execution (it is only explicitly required for background parallel tasks in SEQUENTIAL/DAG triage flows).
|
|
55
|
-
|
|
56
|
-
---
|
|
57
|
-
|
|
58
|
-
## MCP Profile Used: `implement`
|
|
59
|
-
|
|
60
|
-
**Reason:** The shortcut table assigns `implement` as the MCP profile for the `implement` command. This is confirmed by the "MCP 프로필 자동 결정" table:
|
|
61
|
-
|
|
62
|
-
| 에이전트 | MCP |
|
|
63
|
-
|----------|-----|
|
|
64
|
-
| executor, build-fixer, debugger, deep-executor | implement |
|
|
65
|
-
|
|
66
|
-
---
|
|
67
|
-
|
|
68
|
-
## Triage Triggered: NO
|
|
69
|
-
|
|
70
|
-
Command shortcut mode explicitly bypasses triage. From the mode table:
|
|
71
|
-
|
|
72
|
-
> 커맨드 숏컷 → 트리아지: **없음 (즉시 실행)**
|
|
73
|
-
|
|
74
|
-
The triage pipeline (Codex `--full-auto` classification + Opus inline decomposition) is only activated in **auto** mode (`/tfx-auto "..."`) or **manual** mode (`/tfx-auto N:agent_type "..."`).
|
|
75
|
-
|
|
76
|
-
---
|
|
77
|
-
|
|
78
|
-
## Delegation to tfx-multi: NO
|
|
79
|
-
|
|
80
|
-
`tfx-multi` delegation only occurs after triage produces **2 or more subtasks**. Since triage is skipped entirely in command shortcut mode, tfx-multi is never invoked. The task is executed directly as a single `tfx-route.sh` call.
|
|
81
|
-
|
|
82
|
-
> From SKILL.md: "서브태스크 수 1개 → tfx-auto 직접 실행 ... 2개+ → tfx-multi Phase 3"
|
|
83
|
-
> Shortcut mode never reaches the subtask-count decision point.
|
|
84
|
-
|
|
85
|
-
---
|
|
86
|
-
|
|
87
|
-
## Summary Table
|
|
88
|
-
|
|
89
|
-
| Dimension | Value |
|
|
90
|
-
|-----------|-------|
|
|
91
|
-
| Mode | Command Shortcut (커맨드 숏컷) |
|
|
92
|
-
| Triage | None (즉시 실행) |
|
|
93
|
-
| Agent | `executor` (Codex) |
|
|
94
|
-
| MCP Profile | `implement` |
|
|
95
|
-
| Bash command | `bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement` |
|
|
96
|
-
| tfx-multi delegation | No |
|
|
97
|
-
| run_in_background | Not applied (single shortcut, foreground) |
|
|
@@ -1,10 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"expectations": [
|
|
3
|
-
{"text": "Routes to 'executor' agent (not architect, not analyst)", "passed": true, "evidence": "Agent output: 'Agent: executor — Directly mapped from the implement shortcut row'"},
|
|
4
|
-
{"text": "Uses 'implement' MCP profile", "passed": true, "evidence": "Agent output: 'MCP Profile: implement — Mapped from both the shortcut table and the MCP 프로필 자동 결정 table'"},
|
|
5
|
-
{"text": "Generates correct tfx-route.sh command syntax", "passed": true, "evidence": "Command: bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement"},
|
|
6
|
-
{"text": "Does NOT trigger triage (single command shortcut)", "passed": true, "evidence": "Agent output: 'Triage: Not triggered — Command shortcut mode explicitly skips the Codex --full-auto classification'"},
|
|
7
|
-
{"text": "Does NOT delegate to tfx-multi", "passed": true, "evidence": "Agent output: 'tfx-multi: Not delegated — only activates after triage produces 2+ subtasks'"}
|
|
8
|
-
],
|
|
9
|
-
"summary": {"passed": 5, "failed": 0, "total": 5, "pass_rate": 1.0}
|
|
10
|
-
}
|
package/skills/tfx-workspace/iteration-1/routing-implement-shortcut/with_skill/outputs/analysis.md
DELETED
|
@@ -1,94 +0,0 @@
|
|
|
1
|
-
# Routing Analysis: `/implement JWT 인증 미들웨어 추가해줘`
|
|
2
|
-
|
|
3
|
-
## Input
|
|
4
|
-
|
|
5
|
-
```
|
|
6
|
-
/implement JWT 인증 미들웨어 추가해줘
|
|
7
|
-
```
|
|
8
|
-
|
|
9
|
-
---
|
|
10
|
-
|
|
11
|
-
## Mode Selected: Command Shortcut
|
|
12
|
-
|
|
13
|
-
The trigger word is `implement`, which is listed in the SKILL.md `triggers` list and matches a **커맨드 숏컷 (Command Shortcut)** entry in the "Codex 직행" table.
|
|
14
|
-
|
|
15
|
-
Per the skill definition (모드 표):
|
|
16
|
-
|
|
17
|
-
| 입력 형식 | 모드 | 트리아지 |
|
|
18
|
-
|-----------|------|----------|
|
|
19
|
-
| `/implement JWT 추가` | 커맨드 숏컷 | 없음 (즉시 실행) |
|
|
20
|
-
|
|
21
|
-
The input `/implement JWT 인증 미들웨어 추가해줘` matches this exact pattern.
|
|
22
|
-
|
|
23
|
-
---
|
|
24
|
-
|
|
25
|
-
## Agent Selected: `executor`
|
|
26
|
-
|
|
27
|
-
From the "Codex 직행" shortcut table:
|
|
28
|
-
|
|
29
|
-
| 커맨드 | 에이전트 | MCP |
|
|
30
|
-
|--------|---------|-----|
|
|
31
|
-
| `implement` | executor | implement |
|
|
32
|
-
|
|
33
|
-
`executor` is selected because the command shortcut `implement` maps directly to it. No classification or decomposition step occurs.
|
|
34
|
-
|
|
35
|
-
---
|
|
36
|
-
|
|
37
|
-
## Exact Bash Command That Would Be Generated
|
|
38
|
-
|
|
39
|
-
```
|
|
40
|
-
bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement
|
|
41
|
-
```
|
|
42
|
-
|
|
43
|
-
This follows the shortcut execution pattern defined in the skill:
|
|
44
|
-
|
|
45
|
-
> 패턴: `Bash("bash ~/.claude/scripts/tfx-route.sh {에이전트} '{PROMPT}' {MCP}")`
|
|
46
|
-
|
|
47
|
-
The PROMPT is the remainder of the user's input after the command keyword (`JWT 인증 미들웨어 추가해줘`).
|
|
48
|
-
|
|
49
|
-
---
|
|
50
|
-
|
|
51
|
-
## MCP Profile Used: `implement`
|
|
52
|
-
|
|
53
|
-
From the shortcut table, `implement` → `executor` → MCP profile `implement`.
|
|
54
|
-
|
|
55
|
-
This is also confirmed by the "MCP 프로필 자동 결정" table:
|
|
56
|
-
|
|
57
|
-
| 에이전트 | MCP |
|
|
58
|
-
|----------|-----|
|
|
59
|
-
| executor, build-fixer, debugger, deep-executor | implement |
|
|
60
|
-
|
|
61
|
-
---
|
|
62
|
-
|
|
63
|
-
## Triage Triggered: No
|
|
64
|
-
|
|
65
|
-
The skill definition explicitly states that command shortcuts bypass triage:
|
|
66
|
-
|
|
67
|
-
> 커맨드명 매칭 시 트리아지 없이 즉시 실행.
|
|
68
|
-
|
|
69
|
-
The mode table also confirms: 트리아지 = **없음 (즉시 실행)** for command shortcut inputs.
|
|
70
|
-
|
|
71
|
-
---
|
|
72
|
-
|
|
73
|
-
## Delegation to tfx-multi: No
|
|
74
|
-
|
|
75
|
-
`tfx-multi` delegation is only triggered **after triage** when subtask count >= 2:
|
|
76
|
-
|
|
77
|
-
> 트리아지 결과 서브태스크가 2개 이상이면 tfx-multi Native Teams 모드로 자동 전환한다.
|
|
78
|
-
|
|
79
|
-
Since triage is skipped entirely in command shortcut mode, there are no subtasks to count, and the execution goes directly to `tfx-auto` single-agent execution. `tfx-multi` is **not** invoked.
|
|
80
|
-
|
|
81
|
-
---
|
|
82
|
-
|
|
83
|
-
## Summary
|
|
84
|
-
|
|
85
|
-
| Field | Value |
|
|
86
|
-
|-------|-------|
|
|
87
|
-
| Mode | Command Shortcut |
|
|
88
|
-
| Triage | None (skipped) |
|
|
89
|
-
| Agent | `executor` |
|
|
90
|
-
| CLI | Codex |
|
|
91
|
-
| MCP Profile | `implement` |
|
|
92
|
-
| tfx-multi delegation | No |
|
|
93
|
-
| Bash command | `bash ~/.claude/scripts/tfx-route.sh executor 'JWT 인증 미들웨어 추가해줘' implement` |
|
|
94
|
-
| run_in_background | Not specified for single shortcut (no explicit flag in shortcut pattern; background flag applies to triage-spawned workers) |
|
|
@@ -1,12 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"eval_id": 2,
|
|
3
|
-
"eval_name": "routing-multi-task-triage",
|
|
4
|
-
"prompt": "/tfx-auto 프론트엔드 리팩터링하고 보안 리뷰도 해줘",
|
|
5
|
-
"assertions": [
|
|
6
|
-
"Identifies this as auto mode (not command shortcut)",
|
|
7
|
-
"Triggers Codex classification step",
|
|
8
|
-
"Decomposes into at least 2 subtasks",
|
|
9
|
-
"Notes delegation to tfx-multi for subtasks >= 2",
|
|
10
|
-
"Does NOT try to execute all subtasks directly"
|
|
11
|
-
]
|
|
12
|
-
}
|
|
@@ -1,10 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"expectations": [
|
|
3
|
-
{"text": "Identifies this as auto mode (not command shortcut)", "passed": true, "evidence": "Agent output: 'Mode selected: Auto (자동) — matches none of the command shortcut keywords'"},
|
|
4
|
-
{"text": "Triggers Codex classification step", "passed": true, "evidence": "Agent output: 'Codex --full-auto classification to identify task parts'"},
|
|
5
|
-
{"text": "Decomposes into at least 2 subtasks", "passed": true, "evidence": "Agent output: 2 subtasks — executor(Frontend refactoring) + security-reviewer(Security review)"},
|
|
6
|
-
{"text": "Notes delegation to tfx-multi for subtasks >= 2", "passed": true, "evidence": "Agent output: 'subtasks.length = 2 >= 2, hands off to tfx-multi Phase 3'"},
|
|
7
|
-
{"text": "Does NOT try to execute all subtasks directly", "passed": true, "evidence": "Agent correctly delegates to tfx-multi Phase 3"}
|
|
8
|
-
],
|
|
9
|
-
"summary": {"passed": 5, "failed": 0, "total": 5, "pass_rate": 1.0}
|
|
10
|
-
}
|
package/skills/tfx-workspace/iteration-1/routing-multi-task-triage/old_skill/outputs/analysis.md
DELETED
|
@@ -1,209 +0,0 @@
|
|
|
1
|
-
# tfx-auto 라우팅 분석
|
|
2
|
-
|
|
3
|
-
## 입력
|
|
4
|
-
```
|
|
5
|
-
/tfx-auto 프론트엔드 리팩터링하고 보안 리뷰도 해줘
|
|
6
|
-
```
|
|
7
|
-
|
|
8
|
-
---
|
|
9
|
-
|
|
10
|
-
## 1. 모드 선택
|
|
11
|
-
|
|
12
|
-
**선택된 모드: 자동(Auto)**
|
|
13
|
-
|
|
14
|
-
입력 형식이 `/tfx-auto "..."` 형태이며, 커맨드 숏컷 키워드(`implement`, `build`, `research` 등) 없이 자연어 복합 요청이 주어졌다. 또한 수동 모드의 `N:agent_type` 접두어도 없다.
|
|
15
|
-
|
|
16
|
-
SKILL.md 모드 판정표에 따라:
|
|
17
|
-
|
|
18
|
-
| 입력 형식 | 모드 | 트리아지 |
|
|
19
|
-
|-----------|------|----------|
|
|
20
|
-
| `/tfx-auto "리팩터링 + UI"` | 자동 | Codex 분류 → Opus 분해 |
|
|
21
|
-
|
|
22
|
-
따라서 이 요청은 **자동 모드**로 처리되며, 트리아지 파이프라인이 전부 실행된다.
|
|
23
|
-
|
|
24
|
-
---
|
|
25
|
-
|
|
26
|
-
## 2. 트리아지 실행 여부 및 방식
|
|
27
|
-
|
|
28
|
-
**트리아지가 트리거된다.**
|
|
29
|
-
|
|
30
|
-
자동 모드에서는 다음 2단계 트리아지가 수행된다:
|
|
31
|
-
|
|
32
|
-
### 2-1단계: Codex 분류
|
|
33
|
-
|
|
34
|
-
```bash
|
|
35
|
-
codex exec --full-auto --skip-git-repo-check
|
|
36
|
-
```
|
|
37
|
-
|
|
38
|
-
프롬프트를 Codex에 전달하여 태스크를 부분(parts)으로 분류한다. 반환 형식:
|
|
39
|
-
|
|
40
|
-
```json
|
|
41
|
-
{
|
|
42
|
-
"parts": [
|
|
43
|
-
{ "description": "프론트엔드 리팩터링", "agent": "codex" },
|
|
44
|
-
{ "description": "보안 리뷰", "agent": "codex" }
|
|
45
|
-
]
|
|
46
|
-
}
|
|
47
|
-
```
|
|
48
|
-
|
|
49
|
-
### 2-2단계: Opus 인라인 분해
|
|
50
|
-
|
|
51
|
-
Codex 분류 결과를 바탕으로 Opus가 서브태스크를 분해한다. 반환 형식:
|
|
52
|
-
|
|
53
|
-
```json
|
|
54
|
-
{
|
|
55
|
-
"graph_type": "INDEPENDENT",
|
|
56
|
-
"subtasks": [
|
|
57
|
-
{
|
|
58
|
-
"id": "t1",
|
|
59
|
-
"description": "프론트엔드 코드 리팩터링",
|
|
60
|
-
"scope": "frontend",
|
|
61
|
-
"agent": "executor",
|
|
62
|
-
"mcp_profile": "implement",
|
|
63
|
-
"depends_on": [],
|
|
64
|
-
"context_output": "refactor-summary",
|
|
65
|
-
"context_input": null
|
|
66
|
-
},
|
|
67
|
-
{
|
|
68
|
-
"id": "t2",
|
|
69
|
-
"description": "보안 취약점 리뷰",
|
|
70
|
-
"scope": "security",
|
|
71
|
-
"agent": "security-reviewer",
|
|
72
|
-
"mcp_profile": "review",
|
|
73
|
-
"depends_on": [],
|
|
74
|
-
"context_output": "security-report",
|
|
75
|
-
"context_input": null
|
|
76
|
-
}
|
|
77
|
-
]
|
|
78
|
-
}
|
|
79
|
-
```
|
|
80
|
-
|
|
81
|
-
"리팩터링"과 "보안 리뷰"는 서로 독립적이므로 `graph_type: "INDEPENDENT"`로 분해될 가능성이 높다. (보안 리뷰가 리팩터링 후 코드를 대상으로 해야 한다면 `SEQUENTIAL`이 될 수도 있으나, 요청 문맥상 기존 코드를 대상으로 한 독립 병렬 작업으로 분류하는 것이 자연스럽다.)
|
|
82
|
-
|
|
83
|
-
### 트리아지 실패 시
|
|
84
|
-
|
|
85
|
-
Codex 분류 실패 → Opus가 직접 분류 + 분해를 수행한다 (SKILL.md: "실패 시 Opus가 직접 분류+분해").
|
|
86
|
-
|
|
87
|
-
---
|
|
88
|
-
|
|
89
|
-
## 3. 태스크 분해
|
|
90
|
-
|
|
91
|
-
요청 "프론트엔드 리팩터링하고 보안 리뷰도 해줘"는 명확히 두 가지 독립 태스크를 포함한다:
|
|
92
|
-
|
|
93
|
-
| # | 서브태스크 | 분류 근거 | 예상 Agent | 예상 MCP |
|
|
94
|
-
|---|-----------|---------|-----------|---------|
|
|
95
|
-
| t1 | 프론트엔드 코드 리팩터링 | "리팩터링" 키워드 → 코드 수정 작업 | executor | implement |
|
|
96
|
-
| t2 | 보안 취약점 리뷰 | "보안 리뷰" 키워드 → 보안 분석 작업 | security-reviewer | review |
|
|
97
|
-
|
|
98
|
-
에이전트 매핑 근거 (SKILL.md "에이전트 매핑" 섹션):
|
|
99
|
-
- `executor` → Codex, MCP: implement
|
|
100
|
-
- `security-reviewer` → Codex (review), MCP: review
|
|
101
|
-
|
|
102
|
-
---
|
|
103
|
-
|
|
104
|
-
## 4. 서브태스크 수 >= 2 일 때의 동작 (tfx-multi 위임)
|
|
105
|
-
|
|
106
|
-
**서브태스크가 2개이므로 tfx-multi Native Teams 모드로 자동 전환된다.**
|
|
107
|
-
|
|
108
|
-
SKILL.md 규칙:
|
|
109
|
-
> 트리아지 결과 서브태스크가 2개 이상이면 tfx-multi Native Teams 모드로 자동 전환한다.
|
|
110
|
-
|
|
111
|
-
| 서브태스크 수 | 실행 경로 |
|
|
112
|
-
|---|---|
|
|
113
|
-
| 1개 | tfx-auto 직접 실행 |
|
|
114
|
-
| **2개+ (현재 케이스)** | **tfx-multi Phase 3** (TeamCreate → TaskCreate → Agent 래퍼) |
|
|
115
|
-
|
|
116
|
-
전환 시 주요 규칙:
|
|
117
|
-
- 트리아지 결과(서브태스크 배열)를 그대로 tfx-multi Phase 3에 전달한다.
|
|
118
|
-
- tfx-multi의 Phase 2(트리아지)는 건너뛴다 (이미 수행했으므로).
|
|
119
|
-
- Phase 3a(TeamCreate)부터 시작한다.
|
|
120
|
-
|
|
121
|
-
```
|
|
122
|
-
subtasks.length >= 2
|
|
123
|
-
→ tfx-multi Phase 3 실행 (트리아지 결과 재사용)
|
|
124
|
-
→ TeamCreate → TaskCreate × 2 → Agent 래퍼 spawn (Phase 3a~3c)
|
|
125
|
-
→ Phase 4 결과 수집 → Phase 5 정리
|
|
126
|
-
```
|
|
127
|
-
|
|
128
|
-
---
|
|
129
|
-
|
|
130
|
-
## 5. 실행 시퀀스 (전체 흐름)
|
|
131
|
-
|
|
132
|
-
```
|
|
133
|
-
[Step 1] 모드 판정
|
|
134
|
-
입력: /tfx-auto 프론트엔드 리팩터링하고 보안 리뷰도 해줘
|
|
135
|
-
→ 커맨드 숏컷 없음, N:agent 없음
|
|
136
|
-
→ 자동 모드 확정
|
|
137
|
-
|
|
138
|
-
[Step 2] 트리아지 1단계 — Codex 분류
|
|
139
|
-
codex exec --full-auto --skip-git-repo-check
|
|
140
|
-
프롬프트: "프론트엔드 리팩터링하고 보안 리뷰도 해줘"
|
|
141
|
-
→ 반환: { parts: [ {description, agent}, ... ] }
|
|
142
|
-
(실패 시: Step 3으로 넘어가되 Opus가 직접 분류)
|
|
143
|
-
|
|
144
|
-
[Step 3] 트리아지 2단계 — Opus 인라인 분해
|
|
145
|
-
Codex 분류 결과를 입력으로 Opus 호출
|
|
146
|
-
→ 반환: { graph_type: "INDEPENDENT", subtasks: [ t1(리팩터링), t2(보안리뷰) ] }
|
|
147
|
-
|
|
148
|
-
[Step 4] 서브태스크 수 판정
|
|
149
|
-
subtasks.length = 2 → >= 2 조건 충족
|
|
150
|
-
→ tfx-multi Phase 3으로 전환
|
|
151
|
-
|
|
152
|
-
[Step 5] tfx-multi Phase 3a — TeamCreate
|
|
153
|
-
팀 생성: 2개의 워커 슬롯 확보
|
|
154
|
-
|
|
155
|
-
[Step 6] tfx-multi Phase 3b — TaskCreate × 2
|
|
156
|
-
Task 1: 프론트엔드 리팩터링 (executor / implement)
|
|
157
|
-
Task 2: 보안 리뷰 (security-reviewer / review)
|
|
158
|
-
|
|
159
|
-
[Step 7] tfx-multi Phase 3c — Agent 래퍼 spawn (병렬)
|
|
160
|
-
graph_type = INDEPENDENT이므로 두 태스크를 동시에 실행:
|
|
161
|
-
|
|
162
|
-
워커 1 (t1 - 리팩터링):
|
|
163
|
-
Bash("bash ~/.claude/scripts/tfx-route.sh executor '프론트엔드 코드 리팩터링' implement",
|
|
164
|
-
run_in_background=true)
|
|
165
|
-
|
|
166
|
-
워커 2 (t2 - 보안 리뷰):
|
|
167
|
-
Bash("bash ~/.claude/scripts/tfx-route.sh security-reviewer '보안 취약점 리뷰' review",
|
|
168
|
-
run_in_background=true)
|
|
169
|
-
|
|
170
|
-
(graph_type이 SEQUENTIAL이었다면 t1 완료 후 context_output을
|
|
171
|
-
.omc/context/{sid}/combined-{tid}.md에 저장하고 t2에 5번째 인자로 전달)
|
|
172
|
-
|
|
173
|
-
[Step 8] tfx-multi Phase 4 — 결과 수집
|
|
174
|
-
각 워커의 exit_code 판정:
|
|
175
|
-
exit 0 + success → === OUTPUT === 섹션 추출
|
|
176
|
-
exit 124 + timeout → === PARTIAL OUTPUT === 사용
|
|
177
|
-
exit ≠0 + failed → stderr 수집 → Claude fallback 실행
|
|
178
|
-
1차 fallback: Agent(subagent_type="oh-my-claudecode:executor", model="sonnet")
|
|
179
|
-
2차 연속 실패: 실패 보고 + 성공 결과만 종합
|
|
180
|
-
|
|
181
|
-
[Step 9] tfx-multi Phase 5 — 정리 및 보고
|
|
182
|
-
보고 형식:
|
|
183
|
-
## tfx-auto 완료
|
|
184
|
-
**모드**: auto | **그래프**: INDEPENDENT | **레벨**: 0
|
|
185
|
-
| # | 서브태스크 | Agent | CLI | MCP | 레벨 | 상태 | 시간 |
|
|
186
|
-
| 1 | 프론트엔드 리팩터링 | executor | Codex | implement | 0 | ✓ | Xs |
|
|
187
|
-
| 2 | 보안 리뷰 | security-reviewer | Codex | review | 0 | ✓ | Xs |
|
|
188
|
-
### 워커 1: 프론트엔드 리팩터링
|
|
189
|
-
(출력 요약)
|
|
190
|
-
### 워커 2: 보안 리뷰
|
|
191
|
-
(출력 요약)
|
|
192
|
-
### Token Savings Report
|
|
193
|
-
node ~/.claude/scripts/token-snapshot.mjs report {session-id}
|
|
194
|
-
```
|
|
195
|
-
|
|
196
|
-
---
|
|
197
|
-
|
|
198
|
-
## 6. 핵심 판단 포인트 요약
|
|
199
|
-
|
|
200
|
-
| 항목 | 결정값 | 근거 |
|
|
201
|
-
|------|--------|------|
|
|
202
|
-
| 모드 | 자동(Auto) | 커맨드 숏컷 없음, N:agent 없음 |
|
|
203
|
-
| 트리아지 | 실행 (Codex → Opus) | 자동 모드 필수 |
|
|
204
|
-
| 서브태스크 수 | 2개 | "리팩터링" + "보안 리뷰" 명확 분리 |
|
|
205
|
-
| 그래프 타입 | INDEPENDENT (예상) | 두 태스크 간 의존 없음 |
|
|
206
|
-
| 실행 경로 | tfx-multi Phase 3 | subtasks.length >= 2 |
|
|
207
|
-
| 워커 1 에이전트 | executor / Codex | 코드 수정 → implement MCP |
|
|
208
|
-
| 워커 2 에이전트 | security-reviewer / Codex | 보안 분석 → review MCP |
|
|
209
|
-
| 병렬 실행 | 예 | INDEPENDENT → run_in_background=true |
|
|
@@ -1,10 +0,0 @@
|
|
|
1
|
-
{
|
|
2
|
-
"expectations": [
|
|
3
|
-
{"text": "Identifies this as auto mode (not command shortcut)", "passed": true, "evidence": "Agent output: 'Mode selected: AUTO — does not match any command shortcut keyword'"},
|
|
4
|
-
{"text": "Triggers Codex classification step", "passed": true, "evidence": "Agent output: 'Step 1: Codex runs --full-auto --skip-git-repo-check and classifies the input'"},
|
|
5
|
-
{"text": "Decomposes into at least 2 subtasks", "passed": true, "evidence": "Agent output: 2 subtasks — executor(프론트엔드 리팩터링) + security-reviewer(보안 리뷰)"},
|
|
6
|
-
{"text": "Notes delegation to tfx-multi for subtasks >= 2", "passed": true, "evidence": "Agent output: 'subtasks.length = 2 >= 2, tfx-auto automatically hands off to tfx-multi Phase 3'"},
|
|
7
|
-
{"text": "Does NOT try to execute all subtasks directly", "passed": true, "evidence": "Agent correctly delegates to tfx-multi instead of direct execution"}
|
|
8
|
-
],
|
|
9
|
-
"summary": {"passed": 5, "failed": 0, "total": 5, "pass_rate": 1.0}
|
|
10
|
-
}
|