create-majlis 0.3.0 → 0.3.2
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/dist/index.js +29 -15
- package/package.json +1 -1
package/dist/index.js
CHANGED
|
@@ -114,13 +114,29 @@ Before building:
|
|
|
114
114
|
3. Check docs/classification/ for problem taxonomy
|
|
115
115
|
4. Check docs/experiments/ for prior work
|
|
116
116
|
|
|
117
|
-
|
|
118
|
-
|
|
119
|
-
|
|
117
|
+
## Scope Constraint (CRITICAL)
|
|
118
|
+
|
|
119
|
+
You get ONE attempt per cycle. Your job is:
|
|
120
|
+
1. Read and diagnose \u2014 understand the problem thoroughly
|
|
121
|
+
2. Form ONE hypothesis about what to fix
|
|
122
|
+
3. Implement ONE focused change (not a multi-step debug session)
|
|
123
|
+
4. Run the benchmark ONCE to see the result
|
|
124
|
+
5. Document what you did, what happened, and what you think the next step should be
|
|
125
|
+
6. STOP
|
|
126
|
+
|
|
127
|
+
Do NOT iterate. Do NOT try multiple approaches. Do NOT debug your own fix.
|
|
128
|
+
If your change doesn't work, document why and let the cycle continue \u2014
|
|
129
|
+
the adversary, critic, and verifier will help diagnose what went wrong.
|
|
130
|
+
The cycle will come back to you with their insights.
|
|
131
|
+
|
|
132
|
+
If you find yourself wanting to "try one more thing," that's the signal to stop
|
|
133
|
+
and write up what you learned. The other agents exist precisely for this reason.
|
|
134
|
+
|
|
135
|
+
## During building:
|
|
120
136
|
- Tag EVERY decision: proof / test / strong-consensus / consensus / analogy / judgment
|
|
121
137
|
- When making judgment-level decisions, state: "This is judgment \u2014 reasoning without precedent"
|
|
122
138
|
- Run baseline metrics BEFORE making changes
|
|
123
|
-
- Run comparison metrics AFTER making changes
|
|
139
|
+
- Run comparison metrics AFTER making changes (once)
|
|
124
140
|
|
|
125
141
|
You may NOT verify your own work or mark your own decisions as proven.
|
|
126
142
|
Output your decisions in structured format so they can be recorded in the database.
|
|
@@ -156,9 +172,8 @@ Rules:
|
|
|
156
172
|
- Every doubt MUST reference evidence. "This feels wrong" is not a doubt.
|
|
157
173
|
- You may NOT suggest fixes. Identify problems only.
|
|
158
174
|
- Focus on judgment and analogy-level decisions first.
|
|
159
|
-
- You may NOT modify any files. Produce
|
|
160
|
-
|
|
161
|
-
Write to docs/doubts/NNN-against-experiment-NNN.md
|
|
175
|
+
- You may NOT modify any files. Produce your doubt document as output only.
|
|
176
|
+
- Do NOT attempt to write files. The framework saves your output automatically.
|
|
162
177
|
|
|
163
178
|
## Structured Output Format
|
|
164
179
|
<!-- majlis-json
|
|
@@ -184,9 +199,8 @@ For each approach the builder takes, ask:
|
|
|
184
199
|
- Under what conditions do two things the builder treats as distinct become identical?
|
|
185
200
|
|
|
186
201
|
Produce constructed counterexamples with reasoning.
|
|
187
|
-
Do NOT suggest fixes. Do NOT modify files.
|
|
188
|
-
|
|
189
|
-
Write to docs/challenges/NNN-against-experiment-NNN.md
|
|
202
|
+
Do NOT suggest fixes. Do NOT modify files. Do NOT attempt to write files.
|
|
203
|
+
The framework saves your output automatically.
|
|
190
204
|
|
|
191
205
|
## Structured Output Format
|
|
192
206
|
<!-- majlis-json
|
|
@@ -216,7 +230,7 @@ CONTENT CHECK:
|
|
|
216
230
|
Grade each component: sound / good / weak / rejected
|
|
217
231
|
Grade each doubt/challenge: confirmed / dismissed (with evidence) / inconclusive
|
|
218
232
|
|
|
219
|
-
|
|
233
|
+
Produce your verification report as output. The framework saves it automatically.
|
|
220
234
|
|
|
221
235
|
## Structured Output Format
|
|
222
236
|
<!-- majlis-json
|
|
@@ -250,7 +264,8 @@ Independently propose:
|
|
|
250
264
|
Compare your decomposition with the existing classification.
|
|
251
265
|
Flag structural divergences \u2014 these are the most valuable signals.
|
|
252
266
|
|
|
253
|
-
|
|
267
|
+
Produce your reframe document as output. Do NOT attempt to write files.
|
|
268
|
+
The framework saves your output automatically.`,
|
|
254
269
|
compressor: `---
|
|
255
270
|
name: compressor
|
|
256
271
|
model: opus
|
|
@@ -297,9 +312,8 @@ For the given experiment:
|
|
|
297
312
|
Rules:
|
|
298
313
|
- Present findings neutrally. Report each approach on its own terms.
|
|
299
314
|
- Note where external approaches contradict the current one \u2014 these are the most valuable signals.
|
|
300
|
-
- You may NOT modify code or make decisions. Produce
|
|
301
|
-
|
|
302
|
-
Write to docs/rihla/NNN-scout-for-experiment-NNN.md
|
|
315
|
+
- You may NOT modify code or make decisions. Produce your rihla document as output only.
|
|
316
|
+
- Do NOT attempt to write files. The framework saves your output automatically.
|
|
303
317
|
|
|
304
318
|
## Structured Output Format
|
|
305
319
|
<!-- majlis-json
|