contextos-agents 2.0.0-beta.3 → 2.1.0
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/.agents/AGENTS.md +53 -33
- package/.agents/adapters/aider/export.js +41 -14
- package/.agents/adapters/claude/export.js +1 -1
- package/.agents/adapters/copilot/export.js +1 -1
- package/.agents/adapters/cursor/export.js +1 -1
- package/.agents/adapters/drift-detector.js +80 -7
- package/.agents/adapters/gemini/export.js +1 -1
- package/.agents/adapters/pure-compiler.js +10 -0
- package/.agents/adapters/shared.js +13 -4
- package/.agents/adapters/zed/export.js +1 -1
- package/.agents/compiled/registry.v2.json +29 -25
- package/.agents/compiled/registry.v2.sha256 +1 -1
- package/.agents/core/skills/context-os/SKILL.md +34 -37
- package/.agents/core/skills/engineering-workflow/SKILL.md +24 -24
- package/.agents/core/skills/gemini-precision/EXAMPLES.md +72 -0
- package/.agents/core/skills/gemini-precision/SKILL.md +2 -1
- package/.agents/core/skills/gemini-precision/TROUBLESHOOTING.md +25 -0
- package/.agents/core/skills/gemini-precision/skill.yaml +2 -0
- package/.agents/core/skills/gstack-roles/SKILL.md +7 -6
- package/.agents/core/skills/security/SKILL.md +44 -16
- package/.agents/core/skills/security/skill.yaml +0 -1
- package/.agents/ctx.js +16 -10
- package/.agents/doctor.js +7 -20
- package/.agents/generated/claude/skills/context-os/SKILL.md +34 -37
- package/.agents/generated/claude/skills/engineering-workflow/SKILL.md +24 -24
- package/.agents/generated/claude/skills/gemini-precision/SKILL.md +102 -1
- package/.agents/generated/claude/skills/gstack-roles/SKILL.md +7 -6
- package/.agents/generated/claude/skills/security/SKILL.md +44 -16
- package/.agents/generated/gemini/skills/context-os/SKILL.md +34 -37
- package/.agents/generated/gemini/skills/engineering-workflow/SKILL.md +24 -24
- package/.agents/generated/gemini/skills/gemini-precision/SKILL.md +105 -1
- package/.agents/generated/gemini/skills/gstack-roles/SKILL.md +7 -6
- package/.agents/generated/gemini/skills/security/SKILL.md +44 -125
- package/.agents/plugins.js +5 -4
- package/.agents/resolver/canonical-resolver.js +7 -7
- package/.agents/validate.js +69 -1
- package/LICENSE +201 -21
- package/NOTICE +4 -0
- package/README.md +87 -34
- package/bin/commands/hook.js +129 -0
- package/bin/commands/scan.js +70 -0
- package/bin/commands/update.js +7 -8
- package/bin/commands.js +39 -1
- package/bin/index.js +144 -33
- package/bin/lib/gate.js +171 -0
- package/bin/lib/git-snapshot.js +187 -0
- package/bin/lib/scan.js +380 -0
- package/package.json +10 -12
- package/.agents/core/skills/security/security.md +0 -106
- package/benchmarks/v2/analysis/statistics.js +0 -140
- package/benchmarks/v2/analysis/stats.js +0 -69
- package/benchmarks/v2/arms/arm-definitions.js +0 -79
- package/benchmarks/v2/dataset.schema.json +0 -34
- package/benchmarks/v2/evaluators/index.js +0 -25
- package/benchmarks/v2/evaluators/verified-success.js +0 -116
- package/benchmarks/v2/harness/runner.js +0 -88
- package/benchmarks/v2/pilot-tasks.json +0 -392
|
@@ -1,392 +0,0 @@
|
|
|
1
|
-
[
|
|
2
|
-
{
|
|
3
|
-
"id": "task-01-mock-task",
|
|
4
|
-
"description": "This is mock task 1 for the benchmark pilot.",
|
|
5
|
-
"expectedState": {
|
|
6
|
-
"filesToExist": [
|
|
7
|
-
"output-01.js"
|
|
8
|
-
],
|
|
9
|
-
"filesToContain": {
|
|
10
|
-
"output-01.js": "console.log('mock 1');"
|
|
11
|
-
}
|
|
12
|
-
},
|
|
13
|
-
"hash": "sha256:7afaa346b4bf92bf9dc21e9ae809887412a86beb766842e99df7fee6573a4781"
|
|
14
|
-
},
|
|
15
|
-
{
|
|
16
|
-
"id": "task-02-mock-task",
|
|
17
|
-
"description": "This is mock task 2 for the benchmark pilot.",
|
|
18
|
-
"expectedState": {
|
|
19
|
-
"filesToExist": [
|
|
20
|
-
"output-02.js"
|
|
21
|
-
],
|
|
22
|
-
"filesToContain": {
|
|
23
|
-
"output-02.js": "console.log('mock 2');"
|
|
24
|
-
}
|
|
25
|
-
},
|
|
26
|
-
"hash": "sha256:78686d2704d4d6900bb73f4941aa661a26a09f4346ab5e83281c0a18830ad1dd"
|
|
27
|
-
},
|
|
28
|
-
{
|
|
29
|
-
"id": "task-03-mock-task",
|
|
30
|
-
"description": "This is mock task 3 for the benchmark pilot.",
|
|
31
|
-
"expectedState": {
|
|
32
|
-
"filesToExist": [
|
|
33
|
-
"output-03.js"
|
|
34
|
-
],
|
|
35
|
-
"filesToContain": {
|
|
36
|
-
"output-03.js": "console.log('mock 3');"
|
|
37
|
-
}
|
|
38
|
-
},
|
|
39
|
-
"hash": "sha256:1724980aa084bbaa16a8a69b664b783799bc75f8abe13ffc65b76843564fbbff"
|
|
40
|
-
},
|
|
41
|
-
{
|
|
42
|
-
"id": "task-04-mock-task",
|
|
43
|
-
"description": "This is mock task 4 for the benchmark pilot.",
|
|
44
|
-
"expectedState": {
|
|
45
|
-
"filesToExist": [
|
|
46
|
-
"output-04.js"
|
|
47
|
-
],
|
|
48
|
-
"filesToContain": {
|
|
49
|
-
"output-04.js": "console.log('mock 4');"
|
|
50
|
-
}
|
|
51
|
-
},
|
|
52
|
-
"hash": "sha256:8f9ba36d8e59dd863b941cf6aba8709674901dba17f3e9c904e4da733538e38a"
|
|
53
|
-
},
|
|
54
|
-
{
|
|
55
|
-
"id": "task-05-mock-task",
|
|
56
|
-
"description": "This is mock task 5 for the benchmark pilot.",
|
|
57
|
-
"expectedState": {
|
|
58
|
-
"filesToExist": [
|
|
59
|
-
"output-05.js"
|
|
60
|
-
],
|
|
61
|
-
"filesToContain": {
|
|
62
|
-
"output-05.js": "console.log('mock 5');"
|
|
63
|
-
}
|
|
64
|
-
},
|
|
65
|
-
"hash": "sha256:e7c08de777760acda054e9e20c19133287d1c245ef5efb6abaea6ec1ee3020e9"
|
|
66
|
-
},
|
|
67
|
-
{
|
|
68
|
-
"id": "task-06-mock-task",
|
|
69
|
-
"description": "This is mock task 6 for the benchmark pilot.",
|
|
70
|
-
"expectedState": {
|
|
71
|
-
"filesToExist": [
|
|
72
|
-
"output-06.js"
|
|
73
|
-
],
|
|
74
|
-
"filesToContain": {
|
|
75
|
-
"output-06.js": "console.log('mock 6');"
|
|
76
|
-
}
|
|
77
|
-
},
|
|
78
|
-
"hash": "sha256:98c4ca9fa7207111d5816fe81f22b0e0cd27dc2b9d78ee9deb626d0a901f6921"
|
|
79
|
-
},
|
|
80
|
-
{
|
|
81
|
-
"id": "task-07-mock-task",
|
|
82
|
-
"description": "This is mock task 7 for the benchmark pilot.",
|
|
83
|
-
"expectedState": {
|
|
84
|
-
"filesToExist": [
|
|
85
|
-
"output-07.js"
|
|
86
|
-
],
|
|
87
|
-
"filesToContain": {
|
|
88
|
-
"output-07.js": "console.log('mock 7');"
|
|
89
|
-
}
|
|
90
|
-
},
|
|
91
|
-
"hash": "sha256:014e4ceaa5a9e5112150d2536f78ecab3801305b5e95ceabf1de359314f9274f"
|
|
92
|
-
},
|
|
93
|
-
{
|
|
94
|
-
"id": "task-08-mock-task",
|
|
95
|
-
"description": "This is mock task 8 for the benchmark pilot.",
|
|
96
|
-
"expectedState": {
|
|
97
|
-
"filesToExist": [
|
|
98
|
-
"output-08.js"
|
|
99
|
-
],
|
|
100
|
-
"filesToContain": {
|
|
101
|
-
"output-08.js": "console.log('mock 8');"
|
|
102
|
-
}
|
|
103
|
-
},
|
|
104
|
-
"hash": "sha256:f0470f4ad3dcb981ec0447de20036801ec638a0b7b32ae4dd8e6d1c94c08257e"
|
|
105
|
-
},
|
|
106
|
-
{
|
|
107
|
-
"id": "task-09-mock-task",
|
|
108
|
-
"description": "This is mock task 9 for the benchmark pilot.",
|
|
109
|
-
"expectedState": {
|
|
110
|
-
"filesToExist": [
|
|
111
|
-
"output-09.js"
|
|
112
|
-
],
|
|
113
|
-
"filesToContain": {
|
|
114
|
-
"output-09.js": "console.log('mock 9');"
|
|
115
|
-
}
|
|
116
|
-
},
|
|
117
|
-
"hash": "sha256:e34d3152baeb112c9bd77f2381ea0cf9b902613c95d5ae2660b51a4e7ede2785"
|
|
118
|
-
},
|
|
119
|
-
{
|
|
120
|
-
"id": "task-10-mock-task",
|
|
121
|
-
"description": "This is mock task 10 for the benchmark pilot.",
|
|
122
|
-
"expectedState": {
|
|
123
|
-
"filesToExist": [
|
|
124
|
-
"output-10.js"
|
|
125
|
-
],
|
|
126
|
-
"filesToContain": {
|
|
127
|
-
"output-10.js": "console.log('mock 10');"
|
|
128
|
-
}
|
|
129
|
-
},
|
|
130
|
-
"hash": "sha256:8bfdd14ef40300544ba24d517506314c6b8fd3075ecad82341f15eb2715692e0"
|
|
131
|
-
},
|
|
132
|
-
{
|
|
133
|
-
"id": "task-11-mock-task",
|
|
134
|
-
"description": "This is mock task 11 for the benchmark pilot.",
|
|
135
|
-
"expectedState": {
|
|
136
|
-
"filesToExist": [
|
|
137
|
-
"output-11.js"
|
|
138
|
-
],
|
|
139
|
-
"filesToContain": {
|
|
140
|
-
"output-11.js": "console.log('mock 11');"
|
|
141
|
-
}
|
|
142
|
-
},
|
|
143
|
-
"hash": "sha256:555ada841ff8a3cc6b4ce15165f2d214f818bde82cdb71d942137ab132d499e6"
|
|
144
|
-
},
|
|
145
|
-
{
|
|
146
|
-
"id": "task-12-mock-task",
|
|
147
|
-
"description": "This is mock task 12 for the benchmark pilot.",
|
|
148
|
-
"expectedState": {
|
|
149
|
-
"filesToExist": [
|
|
150
|
-
"output-12.js"
|
|
151
|
-
],
|
|
152
|
-
"filesToContain": {
|
|
153
|
-
"output-12.js": "console.log('mock 12');"
|
|
154
|
-
}
|
|
155
|
-
},
|
|
156
|
-
"hash": "sha256:712690d9849664fd0adbfeb295d9e47618024ad8c4be52adfa314e3d31481e50"
|
|
157
|
-
},
|
|
158
|
-
{
|
|
159
|
-
"id": "task-13-mock-task",
|
|
160
|
-
"description": "This is mock task 13 for the benchmark pilot.",
|
|
161
|
-
"expectedState": {
|
|
162
|
-
"filesToExist": [
|
|
163
|
-
"output-13.js"
|
|
164
|
-
],
|
|
165
|
-
"filesToContain": {
|
|
166
|
-
"output-13.js": "console.log('mock 13');"
|
|
167
|
-
}
|
|
168
|
-
},
|
|
169
|
-
"hash": "sha256:aa050d240a6c17ad68f304685f77782bbb8ab5cc6dae23d3cbd845de6cf87304"
|
|
170
|
-
},
|
|
171
|
-
{
|
|
172
|
-
"id": "task-14-mock-task",
|
|
173
|
-
"description": "This is mock task 14 for the benchmark pilot.",
|
|
174
|
-
"expectedState": {
|
|
175
|
-
"filesToExist": [
|
|
176
|
-
"output-14.js"
|
|
177
|
-
],
|
|
178
|
-
"filesToContain": {
|
|
179
|
-
"output-14.js": "console.log('mock 14');"
|
|
180
|
-
}
|
|
181
|
-
},
|
|
182
|
-
"hash": "sha256:ec88ecb7025218c79945d70e568f358e61452ab70d9c434b337411f6a8f468fc"
|
|
183
|
-
},
|
|
184
|
-
{
|
|
185
|
-
"id": "task-15-mock-task",
|
|
186
|
-
"description": "This is mock task 15 for the benchmark pilot.",
|
|
187
|
-
"expectedState": {
|
|
188
|
-
"filesToExist": [
|
|
189
|
-
"output-15.js"
|
|
190
|
-
],
|
|
191
|
-
"filesToContain": {
|
|
192
|
-
"output-15.js": "console.log('mock 15');"
|
|
193
|
-
}
|
|
194
|
-
},
|
|
195
|
-
"hash": "sha256:5f1f6caf37ba31b3b91b876e5e592717abc9c97e0349c15e7ff3c2da2d133947"
|
|
196
|
-
},
|
|
197
|
-
{
|
|
198
|
-
"id": "task-16-mock-task",
|
|
199
|
-
"description": "This is mock task 16 for the benchmark pilot.",
|
|
200
|
-
"expectedState": {
|
|
201
|
-
"filesToExist": [
|
|
202
|
-
"output-16.js"
|
|
203
|
-
],
|
|
204
|
-
"filesToContain": {
|
|
205
|
-
"output-16.js": "console.log('mock 16');"
|
|
206
|
-
}
|
|
207
|
-
},
|
|
208
|
-
"hash": "sha256:e8e6bfe60cd69b3373e667336d41d3c95fdbc5008f9c0eb99b1b94460e7ad0fd"
|
|
209
|
-
},
|
|
210
|
-
{
|
|
211
|
-
"id": "task-17-mock-task",
|
|
212
|
-
"description": "This is mock task 17 for the benchmark pilot.",
|
|
213
|
-
"expectedState": {
|
|
214
|
-
"filesToExist": [
|
|
215
|
-
"output-17.js"
|
|
216
|
-
],
|
|
217
|
-
"filesToContain": {
|
|
218
|
-
"output-17.js": "console.log('mock 17');"
|
|
219
|
-
}
|
|
220
|
-
},
|
|
221
|
-
"hash": "sha256:3a41c6b074a0770e0fb6822ed252c69a77c05fa7bd97fcd9bf9b16f6cf193dd8"
|
|
222
|
-
},
|
|
223
|
-
{
|
|
224
|
-
"id": "task-18-mock-task",
|
|
225
|
-
"description": "This is mock task 18 for the benchmark pilot.",
|
|
226
|
-
"expectedState": {
|
|
227
|
-
"filesToExist": [
|
|
228
|
-
"output-18.js"
|
|
229
|
-
],
|
|
230
|
-
"filesToContain": {
|
|
231
|
-
"output-18.js": "console.log('mock 18');"
|
|
232
|
-
}
|
|
233
|
-
},
|
|
234
|
-
"hash": "sha256:dbd6cc88cb73c805eef95f0cea09e09927f6b13e33fad1e69c049d174eab537f"
|
|
235
|
-
},
|
|
236
|
-
{
|
|
237
|
-
"id": "task-19-mock-task",
|
|
238
|
-
"description": "This is mock task 19 for the benchmark pilot.",
|
|
239
|
-
"expectedState": {
|
|
240
|
-
"filesToExist": [
|
|
241
|
-
"output-19.js"
|
|
242
|
-
],
|
|
243
|
-
"filesToContain": {
|
|
244
|
-
"output-19.js": "console.log('mock 19');"
|
|
245
|
-
}
|
|
246
|
-
},
|
|
247
|
-
"hash": "sha256:b844124ce8543a3863a82699dd4cefb5e9b9391547934546974419be2b069895"
|
|
248
|
-
},
|
|
249
|
-
{
|
|
250
|
-
"id": "task-20-mock-task",
|
|
251
|
-
"description": "This is mock task 20 for the benchmark pilot.",
|
|
252
|
-
"expectedState": {
|
|
253
|
-
"filesToExist": [
|
|
254
|
-
"output-20.js"
|
|
255
|
-
],
|
|
256
|
-
"filesToContain": {
|
|
257
|
-
"output-20.js": "console.log('mock 20');"
|
|
258
|
-
}
|
|
259
|
-
},
|
|
260
|
-
"hash": "sha256:ad1eb8063f7138e9fb193982a6bc4db5eed1e8fb4af21ddbda75270cb2c06d51"
|
|
261
|
-
},
|
|
262
|
-
{
|
|
263
|
-
"id": "task-21-mock-task",
|
|
264
|
-
"description": "This is mock task 21 for the benchmark pilot.",
|
|
265
|
-
"expectedState": {
|
|
266
|
-
"filesToExist": [
|
|
267
|
-
"output-21.js"
|
|
268
|
-
],
|
|
269
|
-
"filesToContain": {
|
|
270
|
-
"output-21.js": "console.log('mock 21');"
|
|
271
|
-
}
|
|
272
|
-
},
|
|
273
|
-
"hash": "sha256:2dd74021abc2bb10e9bed627b7fb072311ed8e86b95526b8a57149cd4f89a7df"
|
|
274
|
-
},
|
|
275
|
-
{
|
|
276
|
-
"id": "task-22-mock-task",
|
|
277
|
-
"description": "This is mock task 22 for the benchmark pilot.",
|
|
278
|
-
"expectedState": {
|
|
279
|
-
"filesToExist": [
|
|
280
|
-
"output-22.js"
|
|
281
|
-
],
|
|
282
|
-
"filesToContain": {
|
|
283
|
-
"output-22.js": "console.log('mock 22');"
|
|
284
|
-
}
|
|
285
|
-
},
|
|
286
|
-
"hash": "sha256:4865211d3986fbea264350c8965dba461d6eb5dec27beceaebd247c49b7f6fff"
|
|
287
|
-
},
|
|
288
|
-
{
|
|
289
|
-
"id": "task-23-mock-task",
|
|
290
|
-
"description": "This is mock task 23 for the benchmark pilot.",
|
|
291
|
-
"expectedState": {
|
|
292
|
-
"filesToExist": [
|
|
293
|
-
"output-23.js"
|
|
294
|
-
],
|
|
295
|
-
"filesToContain": {
|
|
296
|
-
"output-23.js": "console.log('mock 23');"
|
|
297
|
-
}
|
|
298
|
-
},
|
|
299
|
-
"hash": "sha256:168e4241ee5aee2f186f0aab6856178a11fef6399d27f17f62cc25cb695aac0e"
|
|
300
|
-
},
|
|
301
|
-
{
|
|
302
|
-
"id": "task-24-mock-task",
|
|
303
|
-
"description": "This is mock task 24 for the benchmark pilot.",
|
|
304
|
-
"expectedState": {
|
|
305
|
-
"filesToExist": [
|
|
306
|
-
"output-24.js"
|
|
307
|
-
],
|
|
308
|
-
"filesToContain": {
|
|
309
|
-
"output-24.js": "console.log('mock 24');"
|
|
310
|
-
}
|
|
311
|
-
},
|
|
312
|
-
"hash": "sha256:1b7f1e388d8f3e020f6f33a61f29bf40f9e56f203f1b634c47b36e70a5493f01"
|
|
313
|
-
},
|
|
314
|
-
{
|
|
315
|
-
"id": "task-25-mock-task",
|
|
316
|
-
"description": "This is mock task 25 for the benchmark pilot.",
|
|
317
|
-
"expectedState": {
|
|
318
|
-
"filesToExist": [
|
|
319
|
-
"output-25.js"
|
|
320
|
-
],
|
|
321
|
-
"filesToContain": {
|
|
322
|
-
"output-25.js": "console.log('mock 25');"
|
|
323
|
-
}
|
|
324
|
-
},
|
|
325
|
-
"hash": "sha256:2ad0d0c64bbc6d5fc6bb10e398ce1e62ae9a6c73b4a13390646d94f90830ee04"
|
|
326
|
-
},
|
|
327
|
-
{
|
|
328
|
-
"id": "task-26-mock-task",
|
|
329
|
-
"description": "This is mock task 26 for the benchmark pilot.",
|
|
330
|
-
"expectedState": {
|
|
331
|
-
"filesToExist": [
|
|
332
|
-
"output-26.js"
|
|
333
|
-
],
|
|
334
|
-
"filesToContain": {
|
|
335
|
-
"output-26.js": "console.log('mock 26');"
|
|
336
|
-
}
|
|
337
|
-
},
|
|
338
|
-
"hash": "sha256:b4eaa7cfe62d917b87a0311079eadd9a7c2488676f84b3bad540425c5d52ed5c"
|
|
339
|
-
},
|
|
340
|
-
{
|
|
341
|
-
"id": "task-27-mock-task",
|
|
342
|
-
"description": "This is mock task 27 for the benchmark pilot.",
|
|
343
|
-
"expectedState": {
|
|
344
|
-
"filesToExist": [
|
|
345
|
-
"output-27.js"
|
|
346
|
-
],
|
|
347
|
-
"filesToContain": {
|
|
348
|
-
"output-27.js": "console.log('mock 27');"
|
|
349
|
-
}
|
|
350
|
-
},
|
|
351
|
-
"hash": "sha256:1d571536957594de46fbd9fe0710eefcc66d849ea53dde9ad61ee483941349e1"
|
|
352
|
-
},
|
|
353
|
-
{
|
|
354
|
-
"id": "task-28-mock-task",
|
|
355
|
-
"description": "This is mock task 28 for the benchmark pilot.",
|
|
356
|
-
"expectedState": {
|
|
357
|
-
"filesToExist": [
|
|
358
|
-
"output-28.js"
|
|
359
|
-
],
|
|
360
|
-
"filesToContain": {
|
|
361
|
-
"output-28.js": "console.log('mock 28');"
|
|
362
|
-
}
|
|
363
|
-
},
|
|
364
|
-
"hash": "sha256:765672c715bfb9813a05949c7e9eff2a8cd91eed05b0ee026d177561b77411eb"
|
|
365
|
-
},
|
|
366
|
-
{
|
|
367
|
-
"id": "task-29-mock-task",
|
|
368
|
-
"description": "This is mock task 29 for the benchmark pilot.",
|
|
369
|
-
"expectedState": {
|
|
370
|
-
"filesToExist": [
|
|
371
|
-
"output-29.js"
|
|
372
|
-
],
|
|
373
|
-
"filesToContain": {
|
|
374
|
-
"output-29.js": "console.log('mock 29');"
|
|
375
|
-
}
|
|
376
|
-
},
|
|
377
|
-
"hash": "sha256:7209b936824e13b1ec6eeec1503b0ef2ffd0698378ab5739857130773ebf743c"
|
|
378
|
-
},
|
|
379
|
-
{
|
|
380
|
-
"id": "task-30-mock-task",
|
|
381
|
-
"description": "This is mock task 30 for the benchmark pilot.",
|
|
382
|
-
"expectedState": {
|
|
383
|
-
"filesToExist": [
|
|
384
|
-
"output-30.js"
|
|
385
|
-
],
|
|
386
|
-
"filesToContain": {
|
|
387
|
-
"output-30.js": "console.log('mock 30');"
|
|
388
|
-
}
|
|
389
|
-
},
|
|
390
|
-
"hash": "sha256:872fd6ef064d17a4315293952bf4d8704c47d3d94affa903695dfd7e2f9461e8"
|
|
391
|
-
}
|
|
392
|
-
]
|