dsh-math-modeling-agent 0.2.7 → 0.2.8
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/README.md
CHANGED
|
@@ -108,7 +108,7 @@ D3 方向/D4 裁决），`run-state.mjs gate` 在每次状态转移前强制校
|
|
|
108
108
|
安装：
|
|
109
109
|
|
|
110
110
|
```bash
|
|
111
|
-
dsh plugin --profile web add github:yohanchen1/MathModelingAgent#v0.2.
|
|
111
|
+
dsh plugin --profile web add github:yohanchen1/MathModelingAgent#v0.2.8
|
|
112
112
|
dsh --profile web --dump-config # 检查组合层（应看到 dsh-math-modeling-agent-skills 行）
|
|
113
113
|
dsh web # 重启以加载插件
|
|
114
114
|
```
|
package/package.json
CHANGED
|
@@ -46,7 +46,8 @@ until the user says "可以了". Mathematical modeling is a deep activity;
|
|
|
46
46
|
4. **隐含信息与歧义清单**：numbered; each entry = the LLM's reading + its
|
|
47
47
|
basis + the risk if the reading is wrong. This is the heart of the
|
|
48
48
|
exchange: the user corrects entries here.
|
|
49
|
-
5.
|
|
49
|
+
5. **各子问题数学本质判定**：every sub-question of the problem — however
|
|
50
|
+
many there are (one or ten) — is classified (fitting/optimization/
|
|
50
51
|
geometric optics/simulation/…), with why.
|
|
51
52
|
6. **待澄清问题**：open questions inviting the user to add background,
|
|
52
53
|
correct a reading, or supply their own references — never a yes/no list.
|
|
@@ -99,6 +100,19 @@ record to real files on disk.
|
|
|
99
100
|
batch runs may declare `scope.contractExempt: true` — production modeling
|
|
100
101
|
must never set it.
|
|
101
102
|
|
|
103
|
+
## Phase pace — interactive beats between phases (v2.8)
|
|
104
|
+
|
|
105
|
+
The exposition is delivered PHASE BY PHASE, never as a firehose. The phase
|
|
106
|
+
sequence is: restatement → analysis → assumptions → modeling → solving →
|
|
107
|
+
verification → robustness → evaluation. Between every two phases there MUST
|
|
108
|
+
be a user interaction point: the completed phase is presented, one line asks
|
|
109
|
+
whether to proceed to the next phase, and the user may interrupt, ask for a
|
|
110
|
+
deeper explanation of any step of the just-finished phase, or contribute
|
|
111
|
+
knowledge. Sensitivity/robustness NEVER follows solving automatically — it
|
|
112
|
+
only starts after the verification phase, the user's go-ahead, and its own
|
|
113
|
+
D-interaction. Within a phase with ≥3 major steps, offer the user one
|
|
114
|
+
interruption opportunity after the first step.
|
|
115
|
+
|
|
102
116
|
## L2 display points (MUST show, never ask)
|
|
103
117
|
|
|
104
118
|
- After every attempt round: the 8-section runlog digest (see report-contract.md).
|
|
@@ -24,21 +24,67 @@ Do not copy full prior solutions or private reasoning.
|
|
|
24
24
|
|
|
25
25
|
State the frozen question, final scientific status, answer/recommendation, verified claims and evidence links, conditional/unresolved claims, assumptions and sensitivity, candidate comparison, data and parameter sources, validation, failure cases, limitations, reproducibility commands, and resume conditions when non-SOLVED.
|
|
26
26
|
|
|
27
|
-
## Chat display — the runlog digest (MUST output, v2)
|
|
27
|
+
## Chat display — the runlog digest (MUST output, v2.7)
|
|
28
28
|
|
|
29
29
|
The following is shown DIRECTLY IN THE CHAT, not only on disk.
|
|
30
30
|
|
|
31
|
+
### Readability standard — write for a human reader, never a bullet list
|
|
32
|
+
|
|
33
|
+
Every section is an EXPOSITION, not a summary: state WHY the step is taken,
|
|
34
|
+
HOW the reasoning runs (derivation, mechanism, first-principles chain), WHAT
|
|
35
|
+
came out (numbers with units), and HOW it is checked. A reader who knows
|
|
36
|
+
nothing about the problem must be able to follow each section without
|
|
37
|
+
guessing.
|
|
38
|
+
|
|
39
|
+
Style: step-by-step numbered exposition with plain direct language (this is
|
|
40
|
+
a working document, not a final paper — no paper-style formal phrasing
|
|
41
|
+
required), but with MORE detail than a paper would carry: every parameter
|
|
42
|
+
choice gets its reason, every formula is followed by a plain-language
|
|
43
|
+
explanation of each symbol, every step ends with a transition line
|
|
44
|
+
("这一步保证了…，为下一步…做好准备"), and every conclusion states what it
|
|
45
|
+
would take to be wrong. Minimum expectations per section:
|
|
46
|
+
|
|
47
|
+
- 问题重述：restate the question in your own words, explaining every
|
|
48
|
+
parameter's meaning and where it comes from in the problem text.
|
|
49
|
+
- 问题分析：walk the physical/geometric mechanism step by step (e.g. why the
|
|
50
|
+
nodes move radially, how the focal geometry pins the paraboloid), each
|
|
51
|
+
step justified by problem text, data, or derivation.
|
|
52
|
+
- 模型假设：for EVERY assumption give the reason it is justified
|
|
53
|
+
(original wording / data statistic / derivation / literature) and what
|
|
54
|
+
happens if it fails.
|
|
55
|
+
- 模型建立与求解：write the derivation in full (equations, variable
|
|
56
|
+
definitions, units, how the equation is solved, numerical method and why
|
|
57
|
+
it was chosen), not just the final formula.
|
|
58
|
+
- 验证：explain the verification LOGIC for each obligation (what would
|
|
59
|
+
falsify it, what tolerance is used and why), then the outcome.
|
|
60
|
+
- 鲁棒性：for each perturbation say what was varied, why that parameter,
|
|
61
|
+
and what would count as a conclusion flip.
|
|
62
|
+
- 评价与改进 / 参考文献：state the comparison criterion and map each
|
|
63
|
+
reference to the exact method it supports.
|
|
64
|
+
|
|
31
65
|
### Intake digest (once, at intake completion) — the six-section deep restatement
|
|
32
66
|
|
|
33
|
-
【题目要点摘录】problem's own numbers and constraints, quoted
|
|
34
|
-
【背景解读】engineering/physical mechanism
|
|
35
|
-
【数据语义】every attachment column by column (meaning/units/coordinate system/magnitudes)
|
|
36
|
-
【隐含信息与歧义清单】numbered; each = LLM reading + basis + risk (the user corrects entries here)
|
|
37
|
-
|
|
67
|
+
【题目要点摘录】the problem's own numbers and constraints, quoted, each explained
|
|
68
|
+
【背景解读】engineering/physical mechanism, walked through
|
|
69
|
+
【数据语义】every attachment column by column (meaning/units/coordinate system/magnitudes), with observed statistics
|
|
70
|
+
【隐含信息与歧义清单】numbered; each = LLM reading + derivation/basis + risk (the user corrects entries here)
|
|
71
|
+
【各子问题数学本质判定】every sub-question (any count) classified + why
|
|
38
72
|
【待澄清问题】open questions inviting user background/corrections/references
|
|
39
73
|
|
|
40
74
|
After the D0 exchange, show the delta applied to problem-brief.md.
|
|
41
75
|
|
|
76
|
+
### Quality self-check (MUST run before every D4 verdict)
|
|
77
|
+
|
|
78
|
+
Before presenting a D4 verdict, run the self-check on the current attempt's
|
|
79
|
+
report and show the result in the chat: for each of the eight digest sections,
|
|
80
|
+
confirm whether it answers all four questions — WHY (why this step), HOW
|
|
81
|
+
(how the reasoning runs), WHAT (result with units), CHECK (how it is
|
|
82
|
+
verified, or what would falsify it). Any section failing one of the four is
|
|
83
|
+
flagged as incomplete and the round is NOT presented for verdict until fixed.
|
|
84
|
+
The self-check is a human-facing declaration: the user decides whether the
|
|
85
|
+
exposition is actually sufficient. This is the quality gate; the machine
|
|
86
|
+
gates above only guarantee existence and structure.
|
|
87
|
+
|
|
42
88
|
### Per-round digest (after EVERY attempt round)
|
|
43
89
|
|
|
44
90
|
```text
|