@yahyahammad/venastine 0.2.0 → 0.3.9

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.
Files changed (152) hide show
  1. package/.env.example +9 -4
  2. package/CONFIG_ARCHITECTURE.md +2030 -0
  3. package/LICENSE +201 -201
  4. package/NOTICE +1 -1
  5. package/README.md +1034 -895
  6. package/agents/builtin/build.md +73 -0
  7. package/agents/builtin/compactor.md +75 -75
  8. package/agents/builtin/explore.md +77 -77
  9. package/agents/builtin/general.md +80 -0
  10. package/agents/builtin/pipeline-reviewer.md +100 -100
  11. package/agents/builtin/plan.md +106 -88
  12. package/agents/builtin/review.md +74 -74
  13. package/agents/builtin/test.md +65 -0
  14. package/agents/builtin/writer.md +67 -0
  15. package/agents/manager.py +68 -2
  16. package/agents/subagent_tool.py +472 -238
  17. package/agents/tui_commands.py +238 -151
  18. package/bin/venastine.mjs +150 -4
  19. package/bin/venastine.py +71 -0
  20. package/cli_commands.py +838 -0
  21. package/config.py +95 -1198
  22. package/config.yaml +1103 -0
  23. package/config_edit.py +1301 -0
  24. package/config_schema.py +971 -0
  25. package/config_update.py +749 -0
  26. package/core/agent_activity.py +256 -118
  27. package/core/ansi.py +452 -0
  28. package/core/approval.py +137 -115
  29. package/core/call_view.py +364 -0
  30. package/core/cancel.py +121 -0
  31. package/core/chatgpt_auth.py +747 -0
  32. package/core/claims_view.py +107 -0
  33. package/core/client.py +2715 -1559
  34. package/core/commands.py +460 -0
  35. package/core/compaction.py +1735 -971
  36. package/core/config_loader.py +1982 -1563
  37. package/core/connect.py +686 -0
  38. package/core/events.py +123 -61
  39. package/core/file_mentions.py +280 -0
  40. package/core/interaction.py +411 -328
  41. package/core/line_pattern.py +128 -0
  42. package/core/loop.py +2566 -1490
  43. package/core/memory.py +482 -370
  44. package/core/pipeline_models.py +190 -176
  45. package/core/provider_errors.py +234 -0
  46. package/core/reasoning/authorization.py +130 -130
  47. package/core/reasoning/base.py +219 -219
  48. package/core/reasoning/confidence_scoring.py +484 -468
  49. package/core/reasoning/events.py +149 -149
  50. package/core/reasoning/json_retry.py +19 -1
  51. package/core/reasoning/orchestrator.py +1561 -1382
  52. package/core/reasoning/output_writer.py +345 -286
  53. package/core/reasoning/payload_validation.py +524 -398
  54. package/core/reasoning/pipeline_storage.py +88 -22
  55. package/core/reasoning/review.py +659 -615
  56. package/core/reasoning/scholar.py +458 -434
  57. package/core/reasoning/source_corpus.py +433 -327
  58. package/core/reasoning/source_scoring.py +855 -855
  59. package/core/replay.py +315 -183
  60. package/core/session.py +36 -12
  61. package/core/session_view.py +259 -0
  62. package/core/session_wake.py +259 -0
  63. package/core/shell_sessions.py +1553 -0
  64. package/core/workspace_trust.py +1 -1
  65. package/credentials.py +148 -20
  66. package/database.py +333 -252
  67. package/{PRIVACY.md → docs/PRIVACY.md} +78 -66
  68. package/docs/SECURITY.md +108 -0
  69. package/docs/THIRD_PARTY_NOTICES.md +148 -0
  70. package/env_secrets.py +50 -9
  71. package/json_store.py +343 -123
  72. package/logging_setup.py +295 -185
  73. package/main.py +3433 -1961
  74. package/mcp_client/client.py +733 -499
  75. package/mcp_client/config.py +126 -21
  76. package/mcp_client/registration.py +343 -182
  77. package/memories/manager.py +147 -132
  78. package/memories/tui_commands.py +319 -244
  79. package/package.json +303 -73
  80. package/project_init/config_files.py +282 -278
  81. package/project_init/generator.py +728 -564
  82. package/project_init/manifest.py +18 -6
  83. package/project_init/tui_commands.py +147 -144
  84. package/prompts/critic_pass.md +24 -24
  85. package/prompts/revalidate.md +25 -25
  86. package/prompts/revise.md +2 -0
  87. package/prompts/source_grounding.md +66 -66
  88. package/prompts/system_prompts.py +316 -262
  89. package/providers.json.example +4 -2
  90. package/pyproject.toml +149 -100
  91. package/requirements.txt +295 -72
  92. package/safety/policy_enforcement.py +1082 -746
  93. package/security/askpass.py +419 -0
  94. package/security/capability.py +207 -177
  95. package/security/confine.py +484 -0
  96. package/security/hosted.py +182 -0
  97. package/security/permissions.py +237 -190
  98. package/security/posture.py +124 -12
  99. package/security/protected_paths.py +501 -309
  100. package/security/sandbox.py +4108 -985
  101. package/security/secrets.py +855 -0
  102. package/skills/builtin/math/formal-specification.md +67 -67
  103. package/skills/builtin/math/numerical-methods.md +66 -66
  104. package/skills/builtin/research/experiment-design.md +62 -62
  105. package/skills/builtin/research/reproducibility.md +55 -55
  106. package/skills/builtin/research/source-evaluation.md +70 -70
  107. package/skills/builtin/research/statistical-inference.md +63 -63
  108. package/skills/builtin/research/technical-writing.md +59 -59
  109. package/skills/builtin/software/code-review.md +61 -61
  110. package/skills/builtin/software/debugging.md +61 -61
  111. package/skills/builtin/software/test-design.md +67 -67
  112. package/skills/tui_commands.py +110 -111
  113. package/storage.py +1338 -1005
  114. package/tools/base.py +161 -1
  115. package/tools/builtin/_math_common.py +667 -377
  116. package/tools/builtin/_net_common.py +225 -19
  117. package/tools/builtin/arxiv.py +497 -39
  118. package/tools/builtin/ask_user.py +231 -213
  119. package/tools/builtin/discrete_math.py +35 -4
  120. package/tools/builtin/fetch_url.py +371 -153
  121. package/tools/builtin/file_ops.py +720 -412
  122. package/tools/builtin/geometry.py +48 -12
  123. package/tools/builtin/get_time.py +2 -2
  124. package/tools/builtin/linear_algebra.py +66 -23
  125. package/tools/builtin/load_skill.py +76 -66
  126. package/tools/builtin/logic.py +171 -12
  127. package/tools/builtin/pin.py +227 -193
  128. package/tools/builtin/probability_stats.py +12 -16
  129. package/tools/builtin/project_docs.py +464 -405
  130. package/tools/builtin/remember.py +175 -158
  131. package/tools/builtin/shell.py +980 -461
  132. package/tools/builtin/shell_sessions.py +613 -0
  133. package/tools/builtin/symbolic_math.py +80 -10
  134. package/tools/builtin/todo.py +149 -147
  135. package/tools/builtin/web_search.py +191 -13
  136. package/tools/context.py +159 -136
  137. package/tools/isolation.py +107 -34
  138. package/tools/registry.py +1242 -925
  139. package/tui/app.py +8318 -4100
  140. package/tui/app.tcss +747 -556
  141. package/tui/commands.py +26 -170
  142. package/tui/diffs.py +34 -13
  143. package/tui/history.py +128 -128
  144. package/tui/hosted_driver.py +102 -0
  145. package/tui/markdown.py +1329 -1279
  146. package/tui/meters.py +364 -268
  147. package/tui/ravens.py +146 -146
  148. package/tui/screens.py +1255 -943
  149. package/tui/themes.py +599 -559
  150. package/tui/widgets.py +3813 -2332
  151. package/SECURITY.md +0 -88
  152. package/THIRD_PARTY_NOTICES.md +0 -100
package/config.yaml ADDED
@@ -0,0 +1,1103 @@
1
+ # Venastine harness configuration.
2
+ #
3
+ # Every tunable value the harness has. `config.py` reads this file once at
4
+ # import through `config_schema.py` and publishes the values as module
5
+ # attributes, so the names here are the lower-cased form of the constants the
6
+ # code uses: `max_tokens` is `config.MAX_TOKENS`.
7
+ #
8
+ # WHY THE HISTORY IS NOT IN HERE. Each of these numbers was argued, several
9
+ # were reversed once, and the arguments run to hundreds of lines -- they live
10
+ # in CONFIG_ARCHITECTURE.md, section by section, under the name used here.
11
+ # Read it before changing a value. The comments below say what a value DOES;
12
+ # that document says why it is what it is.
13
+ #
14
+ # THREE THINGS TO KNOW BEFORE EDITING.
15
+ #
16
+ # 1. This file ships with the harness and has exactly ONE location, beside
17
+ # config.py. There is no user-level or project-level copy and no
18
+ # environment variable that redirects it. That absence is deliberate and
19
+ # is a security property rather than an oversight: settings.json refuses
20
+ # shell_approval_mode, ensemble_models, critic_model and embedder_model by
21
+ # name precisely because a project's settings.json beats the user's and
22
+ # arrives with a directory you cloned.
23
+ #
24
+ # 2. Keys marked AUTHORITY decide what this harness may do to its host, or
25
+ # name a provider it sends your research content to. Nothing UNATTENDED
26
+ # can change them -- not a tool call, not a settings file, not an
27
+ # environment variable. This file is the only route and it needs a
28
+ # human: `/config` writes one, but only after saying what the key
29
+ # permits and asking (batch 84).
30
+ #
31
+ # 3. Values are read ONCE, at startup. Editing this file while the harness is
32
+ # running changes nothing until the next launch.
33
+ #
34
+ # A missing key, an unknown key, a duplicated key or a value of the wrong
35
+ # type is a startup error naming the key. There are no defaults behind this
36
+ # file.
37
+
38
+ # ======================================================================
39
+ # Model and loop
40
+ # ======================================================================
41
+
42
+ # The model every path uses when nothing more specific speaks.
43
+ # --model > settings.json default_model > this. AGENT_MODEL overrides.
44
+ model_name: claude-sonnet-5
45
+
46
+ # Per-call OUTPUT ceiling, sent as max_tokens / max_output_tokens /
47
+ # max_completion_tokens. On current Anthropic models it bounds thinking
48
+ # and response text together, so effort spends from the same budget.
49
+ max_tokens: 16000
50
+
51
+ # The default step ceiling: every chat turn, every research pass, and any
52
+ # agent whose frontmatter names no max_steps or names an invalid one.
53
+ max_iterations: 50
54
+
55
+ # How many more times ONE model call is attempted when the provider fails in
56
+ # a way that can pass: a dropped connection, a 408/409/429/5xx, or an error
57
+ # sent inside a stream that had already opened. Total attempts per call is
58
+ # this plus one. The openai and anthropic SDKs still retry a request that
59
+ # fails before its stream opens, inside each attempt. 0 turns this off.
60
+ model_call_max_retries: 2
61
+
62
+ # Seconds before the first of those retries. Each later retry waits twice as
63
+ # long, with up to 25% jitter either way; a server's retry-after can make a
64
+ # wait longer, never shorter, capped at 60s.
65
+ model_call_retry_base_delay_s: 3.0
66
+
67
+ # ======================================================================
68
+ # Subagents
69
+ # ======================================================================
70
+
71
+ # How deep spawn_subagent may nest, checked against
72
+ # ToolContext.subagent_depth.
73
+ subagent_max_depth: 2
74
+
75
+ # How many calls of one response may run at once, for tools whose ToolSpec
76
+ # declares `parallel`. A ceiling, not a target: more calls run in waves of
77
+ # this size. 1 makes the parallel path behave exactly like the sequential
78
+ # one, which is the escape hatch if concurrency is ever suspect.
79
+ subagent_max_parallel: 3
80
+
81
+ # ======================================================================
82
+ # The active agent
83
+ # ======================================================================
84
+
85
+ # The agents shift+tab steps through in the TUI's prompt, in this order.
86
+ # The plain harness ("default") is always the first stop and is not
87
+ # listed. A name no agent tier provides is skipped and named once. An
88
+ # empty list gives shift+tab back to moving focus. /agent <name> reaches
89
+ # every agent, listed here or not.
90
+ agent_cycle:
91
+ - plan
92
+ - build
93
+
94
+ # ======================================================================
95
+ # Deep research pipeline
96
+ # ======================================================================
97
+
98
+ # Revise/re-validate loop iterations per claim before the pipeline falls
99
+ # back.
100
+ max_pipeline_retries: 2
101
+
102
+ # Corrective follow-up attempts when a pass returns malformed JSON.
103
+ # Total attempts per pass is this plus one.
104
+ max_json_retries: 2
105
+
106
+ # ======================================================================
107
+ # Ensemble mode
108
+ # ======================================================================
109
+
110
+ # Run Pass 1 once per `ensemble_models` entry. A settings.json
111
+ # `ensemble_mode` outranks this; the roster below has no settings key.
112
+ ensemble_mode: false
113
+
114
+ # AUTHORITY -- no settings.json key exists at any tier, because a roster chooses
115
+ # which providers N research passes call. N is its length and is never
116
+ # configured separately. Fewer than two DISTINCT (provider_name, model)
117
+ # pairs is refused: one model cannot disagree with itself.
118
+ # null, with ensemble_mode false, is the shipped configuration. Example:
119
+ # ensemble_models:
120
+ # - {provider_name: ANTHROPIC, model: claude-opus-5}
121
+ # - {provider_name: OPENAI, model: gpt-5.1}
122
+ ensemble_models: null
123
+
124
+ # ======================================================================
125
+ # Sampling-parameter support
126
+ # ======================================================================
127
+
128
+ # Models that reject temperature / top_p / top_k outright, or ignore them.
129
+ # core/client.py omits those parameters entirely for anything listed, with a
130
+ # WARNING. Steering on these happens through prompting and
131
+ # model_effort_levels instead.
132
+ #
133
+ # The two gemini names IGNORE the parameters (ROADMAP_v3 §87): no error and
134
+ # no effect. Google says every later Gemini will do the same, then refuse
135
+ # them with a 400, so a new Gemini model belongs here when it ships. OpenAI's
136
+ # reasoning models are not listed: core/client.py drops the parameters on
137
+ # OPENAI whenever a reasoning effort other than "none" is sent, which is
138
+ # OpenAI's own rule and needs no model names.
139
+ #
140
+ # On ANTHROPIC every model drops them since batch 234 (API version
141
+ # 2026-10-21 refuses them on all of them), so the claude names below matter
142
+ # for Claude reached through an OpenAI-compatible gateway, by its bare name.
143
+ models_rejecting_sampling_params:
144
+ - claude-fable-5
145
+ - claude-fable-5-1
146
+ - claude-haiku-5-5
147
+ - claude-mythos-5
148
+ - claude-mythos-5-1
149
+ - claude-opus-4-7
150
+ - claude-opus-4-8
151
+ - claude-opus-5
152
+ - claude-opus-5-5
153
+ - claude-sonnet-5
154
+ - claude-sonnet-5-5
155
+ - gemini-3.5-flash-lite
156
+ - gemini-3.6-flash
157
+
158
+ # ======================================================================
159
+ # Critic and embedder routing
160
+ # ======================================================================
161
+
162
+ # AUTHORITY -- names a provider every claim is sent to. `/critic` writes a
163
+ # user-tier store that OUTRANKS this, and settings.json rejects the key by
164
+ # name at every tier. null means no special routing.
165
+ critic_model: null
166
+
167
+ # AUTHORITY -- names a provider that receives claim text and the text of every
168
+ # page a run fetched. `/embedder` outranks this. null means no special
169
+ # routing.
170
+ embedder_model: null
171
+
172
+ # Per-embedder similarity floor and ceiling, matched by model-name PREFIX
173
+ # (the longest matching prefix wins).
174
+ similarity_calibration:
175
+ text-embedding-3:
176
+ floor: 0.1
177
+ ceiling: 0.65
178
+ text-embedding-ada:
179
+ floor: 0.7
180
+ ceiling: 0.92
181
+ e5-:
182
+ floor: 0.7
183
+ ceiling: 0.92
184
+ multilingual-e5:
185
+ floor: 0.7
186
+ ceiling: 0.92
187
+ bge-:
188
+ floor: 0.55
189
+ ceiling: 0.85
190
+ gte-:
191
+ floor: 0.55
192
+ ceiling: 0.88
193
+ nomic-embed:
194
+ floor: 0.4
195
+ ceiling: 0.8
196
+ embed-english:
197
+ floor: 0.25
198
+ ceiling: 0.8
199
+ voyage:
200
+ floor: 0.35
201
+ ceiling: 0.85
202
+ gemini-embedding:
203
+ floor: 0.35
204
+ ceiling: 0.85
205
+
206
+ # For an embedder that matches no prefix above.
207
+ default_similarity_calibration:
208
+ floor: 0.25
209
+ ceiling: 0.85
210
+
211
+ # The query/passage prefixes asymmetric embedders need, matched by
212
+ # model-name prefix. An embedder absent from here is sent bare text.
213
+ embedder_prefixes:
214
+ e5-:
215
+ query: 'query: '
216
+ passage: 'passage: '
217
+ multilingual-e5:
218
+ query: 'query: '
219
+ passage: 'passage: '
220
+ bge-:
221
+ query: 'Represent this sentence for searching relevant passages: '
222
+ passage: ''
223
+ gte-:
224
+ query: ''
225
+ passage: ''
226
+ nomic-embed:
227
+ query: 'search_query: '
228
+ passage: 'search_document: '
229
+ gemini-embedding:
230
+ query_task_type: RETRIEVAL_QUERY
231
+ passage_task_type: RETRIEVAL_DOCUMENT
232
+ text-embedding-004:
233
+ query_task_type: RETRIEVAL_QUERY
234
+ passage_task_type: RETRIEVAL_DOCUMENT
235
+
236
+ # ======================================================================
237
+ # Source authority: domain classes
238
+ # ======================================================================
239
+
240
+ # Class name -> authority score. The second table, below, maps hosts onto
241
+ # these names, so a score is changed in one place.
242
+ domain_authority_classes:
243
+ restricted_registry: 0.9 # .gov .mil .int .edu, gov.uk, ac.uk...
244
+ intergovernmental: 0.9 # WHO, UN, World Bank, OECD, EU
245
+ standards_body: 0.88 # IETF, W3C, ISO, NIST, IEC
246
+ peer_reviewed_publisher: 0.85
247
+ established_news: 0.65
248
+ preprint_server: 0.6 # not peer reviewed; §45 SQ5 refines this
249
+ reference_tertiary: 0.55 # encyclopaedias -- a pointer to a source
250
+ generic_org: 0.55 # an unrestricted registry, small bump
251
+ generic_commercial: 0.45
252
+ blog_platform: 0.3
253
+ forum_qa: 0.25
254
+ social_media: 0.15
255
+ content_farm: 0.1
256
+
257
+ # The score for a host that matches no suffix below; its class reads
258
+ # "unknown".
259
+ default_domain_authority: 0.45
260
+
261
+ # Host suffix -> a class name from the table above. The `<kind>.<cc>`
262
+ # second-level forms are derived in source_scoring.py, not listed here.
263
+ domain_authority_suffixes:
264
+
265
+ # Restricted registries. The bare TLDs; the `<kind>.<cc>`
266
+ # second-level forms are derived in source_scoring.py rather than
267
+ # listed.
268
+ gov: restricted_registry
269
+ mil: restricted_registry
270
+ int: restricted_registry
271
+ edu: restricted_registry
272
+
273
+ # Intergovernmental bodies, whose hosts are ordinary .org/.int
274
+ # names.
275
+ who.int: intergovernmental
276
+ un.org: intergovernmental
277
+ europa.eu: intergovernmental
278
+ oecd.org: intergovernmental
279
+ imf.org: intergovernmental
280
+ worldbank.org: intergovernmental
281
+ iaea.org: intergovernmental
282
+
283
+ # Standards bodies.
284
+ ietf.org: standards_body
285
+ rfc-editor.org: standards_body
286
+ w3.org: standards_body
287
+ iso.org: standards_body
288
+ iec.ch: standards_body
289
+ ieee.org: standards_body
290
+ unicode.org: standards_body
291
+ ecma-international.org: standards_body
292
+
293
+ # Peer-reviewed publishers and indexes.
294
+ nature.com: peer_reviewed_publisher
295
+ science.org: peer_reviewed_publisher
296
+ sciencemag.org: peer_reviewed_publisher
297
+ nejm.org: peer_reviewed_publisher
298
+ thelancet.com: peer_reviewed_publisher
299
+ bmj.com: peer_reviewed_publisher
300
+ cell.com: peer_reviewed_publisher
301
+ pnas.org: peer_reviewed_publisher
302
+ sciencedirect.com: peer_reviewed_publisher
303
+ springer.com: peer_reviewed_publisher
304
+ link.springer.com: peer_reviewed_publisher
305
+ onlinelibrary.wiley.com: peer_reviewed_publisher
306
+ wiley.com: peer_reviewed_publisher
307
+ tandfonline.com: peer_reviewed_publisher
308
+ sagepub.com: peer_reviewed_publisher
309
+ acm.org: peer_reviewed_publisher
310
+ aps.org: peer_reviewed_publisher
311
+ acs.org: peer_reviewed_publisher
312
+ plos.org: peer_reviewed_publisher
313
+ frontiersin.org: peer_reviewed_publisher
314
+ mdpi.com: peer_reviewed_publisher
315
+ jstor.org: peer_reviewed_publisher
316
+ pubmed.ncbi.nlm.nih.gov: peer_reviewed_publisher
317
+
318
+ # Preprint servers and repositories -- not peer reviewed.
319
+ arxiv.org: preprint_server
320
+ biorxiv.org: preprint_server
321
+ medrxiv.org: preprint_server
322
+ chemrxiv.org: preprint_server
323
+ ssrn.com: preprint_server
324
+ papers.ssrn.com: preprint_server
325
+ osf.io: preprint_server
326
+ researchsquare.com: preprint_server
327
+ semanticscholar.org: preprint_server
328
+
329
+ # News organisations with a correction policy and a masthead.
330
+ # An ALLOWLIST, because "news site" is not a property of a
331
+ # domain suffix and the alternative is scoring every .com that
332
+ # publishes articles.
333
+ reuters.com: established_news
334
+ apnews.com: established_news
335
+ bbc.com: established_news
336
+ bbc.co.uk: established_news
337
+ ft.com: established_news
338
+ economist.com: established_news
339
+ nytimes.com: established_news
340
+ washingtonpost.com: established_news
341
+ wsj.com: established_news
342
+ theguardian.com: established_news
343
+ npr.org: established_news
344
+ pbs.org: established_news
345
+ bloomberg.com: established_news
346
+ nikkei.com: established_news
347
+ afp.com: established_news
348
+ dw.com: established_news
349
+
350
+ # Tertiary references -- they summarise sources rather than
351
+ # being one.
352
+ wikipedia.org: reference_tertiary
353
+ wikimedia.org: reference_tertiary
354
+ wiktionary.org: reference_tertiary
355
+ britannica.com: reference_tertiary
356
+ org: generic_org
357
+
358
+ # Blog and newsletter platforms: the host says nothing about
359
+ # the author.
360
+ medium.com: blog_platform
361
+ substack.com: blog_platform
362
+ wordpress.com: blog_platform
363
+ blogspot.com: blog_platform
364
+ blogger.com: blog_platform
365
+ tumblr.com: blog_platform
366
+ wixsite.com: blog_platform
367
+ ghost.io: blog_platform
368
+ reddit.com: forum_qa
369
+ quora.com: forum_qa
370
+
371
+ # Forums and Q&A. Often correct, never attributable.
372
+ stackexchange.com: forum_qa
373
+ stackoverflow.com: forum_qa
374
+ news.ycombinator.com: forum_qa
375
+ 4chan.org: content_farm
376
+
377
+ # Social media. Speculation and unattributed claims travel here
378
+ # fastest, which is the whole reason a claim needs grounding
379
+ # elsewhere.
380
+ x.com: social_media
381
+ twitter.com: social_media
382
+ facebook.com: social_media
383
+ instagram.com: social_media
384
+ tiktok.com: social_media
385
+ threads.net: social_media
386
+ linkedin.com: social_media
387
+ mastodon.social: social_media
388
+ bsky.app: social_media
389
+ youtube.com: social_media
390
+ vk.com: social_media
391
+ weibo.com: social_media
392
+ t.me: social_media
393
+
394
+ # Aggregators that republish without attribution, and image
395
+ # boards with no text to ground anything in. The measured
396
+ # case: a Pinterest pin scored similarity 1.0 on a Nobel
397
+ # Prize claim.
398
+ pinterest.com: content_farm
399
+ answers.com: content_farm
400
+ ehow.com: content_farm
401
+ scribd.com: content_farm
402
+ coursehero.com: content_farm
403
+ chegg.com: content_farm
404
+
405
+ # How far a stated reason may move a source's authority score, in either
406
+ # direction.
407
+ authority_adjustment_cap: 0.15
408
+
409
+ # ======================================================================
410
+ # The sources/ artifact
411
+ # ======================================================================
412
+
413
+ # Whether the sources/ artifact keeps the fetched page text alongside the
414
+ # scores, or the scores alone.
415
+ persist_source_text: true
416
+
417
+ # ======================================================================
418
+ # Scholarly authority (OpenAlex)
419
+ # ======================================================================
420
+
421
+ # Whether the OpenAlex lookup runs at all. Off ships no outbound
422
+ # scholarly requests.
423
+ scholar_lookup: false
424
+
425
+ # The OpenAlex base URL.
426
+ scholar_api_url: https://api.openalex.org
427
+
428
+ # Per-request timeout, in seconds.
429
+ scholar_timeout_s: 8.0
430
+
431
+ # Retries per request; total attempts is this plus one.
432
+ scholar_max_retries: 2
433
+
434
+ # How long a lookup is cached in-process, in seconds.
435
+ scholar_cache_ttl_s: 900
436
+
437
+ # Sent as OpenAlex's `mailto` parameter when non-empty, which is their
438
+ # polite-pool convention. Blank sends no address and is the shipped value:
439
+ # an address here leaves the machine on every lookup.
440
+ scholar_mailto: ''
441
+
442
+ # The three weights below split one score and a settings.json
443
+ # `source_scoring` block can override each. They are expected to sum to 1.
444
+ scholar_venue_weight: 0.4
445
+ scholar_citation_weight: 0.4
446
+ scholar_author_weight: 0.2
447
+
448
+ # OpenAlex location type -> venue credit. `unknown` is the fallback for an
449
+ # unrecognised type, and a work with no location at all takes
450
+ # `repository`.
451
+ scholar_venue_credit:
452
+ journal: 1.0
453
+ conference: 1.0
454
+ book series: 0.85
455
+ repository: 0.55 # arXiv, bioRxiv, institutional repositories
456
+ unknown: 0.6
457
+
458
+ # The h-index at which author credit saturates.
459
+ scholar_h_saturation: 60
460
+
461
+ # Below this many works in a field-year cohort, the citation percentile is
462
+ # not trusted.
463
+ scholar_min_cohort_size: 200
464
+
465
+ # A work younger than this has not had time to accrue citations, so the
466
+ # citation term is not held against it.
467
+ scholar_min_citation_age_days: 60
468
+
469
+ # The authority a retracted work is given outright, whatever else it
470
+ # scores.
471
+ scholar_retracted_authority: 0.05
472
+
473
+ # ======================================================================
474
+ # Reasoning effort
475
+ # ======================================================================
476
+
477
+ # Reasoning effort when nothing more specific speaks: --effort >
478
+ # settings.json `effort` > this, with settings.json `tui.effort` winning
479
+ # inside the TUI. null sends no effort at all.
480
+ default_effort: high
481
+
482
+ # Known effort levels per model, for providers whose API exposes no
483
+ # capability query. An EMPTY list means the model accepts none, which is
484
+ # different from being absent -- absent falls through to the list below.
485
+ model_effort_levels:
486
+ gpt-4o: []
487
+ gpt-4o-mini: []
488
+ gpt-4.1: []
489
+ gpt-4.1-mini: []
490
+ gpt-4.1-nano: []
491
+ gpt-3.5-turbo: []
492
+
493
+ # Used for a model with no entry above.
494
+ default_effort_levels:
495
+ - low
496
+ - medium
497
+ - high
498
+ - xhigh
499
+ - max
500
+ - ultra
501
+
502
+ # Effort level -> Google thinking_budget in tokens. -1 hands the choice to
503
+ # the model.
504
+ google_thinking_budgets:
505
+ low: 2048
506
+ medium: 8192
507
+ high: 16384
508
+ xhigh: 24576
509
+ ultra: 32768
510
+ max: -1
511
+
512
+ # ======================================================================
513
+ # Database
514
+ # ======================================================================
515
+
516
+ # The SQLite file behind the conversation database. APP_DB_PATH overrides;
517
+ # the npm launcher points it at ~/.config/venastine so state is global
518
+ # while artifacts stay local.
519
+ db_path: app.db
520
+
521
+ # ======================================================================
522
+ # File-ops workspace
523
+ # ======================================================================
524
+
525
+ # The file tools' read-write root, and a PERMISSION BOUNDARY: file_ops
526
+ # auto-approves writes inside it. AGENT_WORKSPACE overrides, and naming it
527
+ # also makes the resolved workspace the project path -- so pointing it at a
528
+ # home directory would auto-approve writes there for every project. Writing
529
+ # the path HERE moves the workspace alone; the key below moves the project
530
+ # with it.
531
+ workspace_dir: ./workspace
532
+
533
+ # Whether that workspace is also the PROJECT: what workspace trust asks about,
534
+ # which .venastine/ supplies settings and agents, where /init scaffolds, and
535
+ # what project-scoped memories belong to. Naming AGENT_WORKSPACE has always
536
+ # done this; the key is the same statement made in the file, for the reader
537
+ # who has no variable to set. Either route suffices, and neither is needed to
538
+ # work inside a project you launched from -- that is already the project.
539
+ workspace_is_project: false
540
+
541
+ # Hard reject before opening, in both directions (25 MB).
542
+ max_file_size_bytes: 25000000
543
+
544
+ # Per-read line ceiling. A larger `count` is clamped, with a note in the
545
+ # result rather than silently.
546
+ max_read_lines: 500
547
+
548
+ # Per-read character ceiling, the truncation point for sandboxed stdout and
549
+ # stderr, and the bound on an MCP tool's whole result.
550
+ max_read_chars: 50000
551
+
552
+ # ROADMAP_v3 §81 (IM7). The largest image `view_image` attaches, in bytes of
553
+ # the file (3.75 MiB). Base64 inflates it by 4/3, so the encoded payload stays
554
+ # under 5 MiB -- the smallest per-image limit of the wires that carry it --
555
+ # and the image is resent with every request for as long as it stays in view.
556
+ max_image_bytes: 3932160
557
+
558
+ # ======================================================================
559
+ # /init
560
+ # ======================================================================
561
+
562
+ # The agent /init runs.
563
+ initializer_agent: initializer
564
+
565
+ # Most characters read_project_doc returns per file.
566
+ init_read_chars: 20000
567
+
568
+ # Step ceiling for the /init run when its agent names none.
569
+ init_max_steps: 12
570
+
571
+ # Largest existing AGENTS.md /init will revise, in characters. Above it
572
+ # /init stops before any model call: the reply replaces the whole file.
573
+ init_max_hub_chars: 1000000
574
+
575
+ # Most lines of the proposed diff /init prints before asking.
576
+ init_diff_shown_lines: 2000
577
+
578
+ # ======================================================================
579
+ # MCP teardown
580
+ # ======================================================================
581
+
582
+ # Seconds allowed for MCP session teardown at shutdown before the harness
583
+ # stops waiting.
584
+ teardown_budget_s: 10.0
585
+
586
+ # ======================================================================
587
+ # MCP results
588
+ # ======================================================================
589
+
590
+ # Largest single server-sent event, in bytes, a Streamable HTTP ("http")
591
+ # MCP server may stream back. A bigger one fails that tool call with an
592
+ # error naming the limit; the session survives. An answer sent as plain
593
+ # JSON is not an event and is not bounded by this. Legacy "sse" servers
594
+ # keep the MCP SDK's own 1 MiB, which it gives no way to change. 16 MiB.
595
+ # What reaches the model is bounded separately, by max_read_chars.
596
+ mcp_max_event_bytes: 16777216
597
+
598
+ # ======================================================================
599
+ # Shell and sandbox
600
+ # ======================================================================
601
+
602
+ # The shell the sandbox runs commands through. EMPTY means auto-detect,
603
+ # which is the shipped value. AGENT_SHELL overrides.
604
+ shell_binary: ''
605
+
606
+ # AUTHORITY -- part of the frozen security posture. Enables the weak
607
+ # host-subprocess backend, which is unbounded code on the host. Bound once
608
+ # at first read and frozen, so no tool call, settings file, environment
609
+ # variable or slash command can move it.
610
+ allow_insecure_sandbox_fallback: false
611
+
612
+ # AUTHORITY -- runs that fallback with no per-run prompt. The pair above and
613
+ # here reads as worse than either alone. Honoured under `tiered` and
614
+ # `contained`; under `always` every command still asks, this one included.
615
+ auto_approve_sandbox_fallback: false
616
+
617
+ # AUTHORITY -- part of the frozen security posture. Lets the agent ask
618
+ # for a command to run in WSL (ROADMAP_v3 §49, slice 3). It is NOT a
619
+ # weaker container: a command there reaches this harness's own
620
+ # config.yaml and ~/.config, gets no memory, CPU or process limit, and
621
+ # can start Windows programs with your full authority -- Windows
622
+ # interop is a kernel feature there and survives an emptied
623
+ # environment. Only a read-only command whose every argument is inside
624
+ # the workspace runs there without asking; everything else meets a
625
+ # prompt under `tiered` and `contained` alike. Shown on the startup
626
+ # banner and the TUI badge whenever it is on.
627
+ allow_wsl_backend: false
628
+
629
+ # Which WSL distribution the backend runs in. EMPTY means the first one
630
+ # `wsl.exe -l -q` lists, which is WSL's own default. A name that is not
631
+ # installed is refused with the installed names, rather than passed to
632
+ # `wsl.exe` -- which answers an unknown distro by exiting 4294967295
633
+ # and writing its complaint to stdout in UTF-16, so the failure would
634
+ # read as the command's rather than the configuration's.
635
+ wsl_distro: ''
636
+
637
+ # AUTHORITY -- part of the frozen security posture. Lets the agent ask
638
+ # for a command to run on the remote host configured below (ROADMAP_v3
639
+ # §49, slice 4). A remote host is not a sandbox and not a weaker one: it
640
+ # is a machine this harness cannot look at. It gets no memory, CPU or
641
+ # process limit -- those belong to the remote -- and the credentials it
642
+ # uses are ones you issued, not ones this harness controls.
643
+ # NOTHING runs there without asking you first, not even a read-only
644
+ # command: "inside the workspace" is a question about a filesystem this
645
+ # process cannot resolve, so it is not asked. Shown on the startup
646
+ # banner and the TUI badge whenever it is on.
647
+ allow_ssh_backend: false
648
+
649
+ # The one remote host the SSH backend reaches. The agent asks for
650
+ # `backend: "ssh"` and gets THIS host; it never names a host itself,
651
+ # for the reason `wsl_distro` is not the agent's either -- which machine
652
+ # this harness may reach is your question, and which command needs it is
653
+ # the agent's.
654
+ ssh_host: ''
655
+ ssh_user: ''
656
+ ssh_port: 22
657
+
658
+ # The private key to authenticate with. EMPTY offers whatever a running
659
+ # ssh-agent holds. This harness NEVER asks for, holds or stores a
660
+ # passphrase: if the key is encrypted, load it into an agent with
661
+ # `ssh-add` first. Measured, and the reason the backend passes
662
+ # BatchMode=yes: without it, an encrypted key with no agent makes ssh
663
+ # wait forever for a passphrase on a terminal this process does not own.
664
+ ssh_identity_file: ''
665
+
666
+ # Where a command runs on the remote. REQUIRED when the backend is on,
667
+ # and deliberately not defaulted to the login home: a command that runs
668
+ # somewhere nobody named is the failure this backend is most able to
669
+ # cause, and nothing here can see where it landed.
670
+ ssh_remote_workspace: ''
671
+
672
+ # The remote's host key, PINNED. Run `ssh-keyscan -p <port> <host>` and
673
+ # paste its output here (the `#` banner line is ignored). A host whose
674
+ # key does not match this is refused -- not prompted about, because a
675
+ # prompt on first use is a question a headless run has nobody to ask,
676
+ # and because a key that CHANGES is exactly the case worth refusing.
677
+ ssh_host_key: ''
678
+
679
+ # Which `ssh` program to run. EMPTY prefers the operating system's own
680
+ # (%SYSTEMROOT%\System32\OpenSSH\ssh.exe on Windows) and falls back to
681
+ # PATH. Not PATH first: on the machine this was built, PATH resolves to
682
+ # a git-for-Windows build while the OS ships its own, and the two report
683
+ # different text for the same failure.
684
+ ssh_binary: ''
685
+
686
+ # AUTHORITY -- THE gate for shell approval, and part of the frozen posture.
687
+ # always -- every command asks
688
+ # tiered -- read-only commands from inert_commands, every argument
689
+ # inside the workspace, run unprompted; every other command
690
+ # asks, including anything that can run code
691
+ # contained -- anything that runs in the container without network runs
692
+ # unprompted, code included; reads outside the workspace and
693
+ # network commands still ask
694
+ # never -- nothing asks
695
+ # An unknown value raises at startup rather than falling back, because one
696
+ # direction of a fallback asks about everything and the other about
697
+ # nothing, and a typo cannot pick. Rejected by name from settings.json at
698
+ # every tier: a cloned repo could otherwise set `never`.
699
+ # Shell approval is governed solely here: `tool_approvals` deliberately has
700
+ # no `shell` key, so there is no second switch that can disagree with this.
701
+ shell_approval_mode: tiered
702
+
703
+ # The image sandboxed commands run in, under Docker or Podman (whichever
704
+ # runs the sandbox -- ROADMAP_v3 §49, SS22). Fully qualified so Podman
705
+ # resolves it without a short-name alias (SS23). AGENT_SANDBOX_IMAGE
706
+ # overrides.
707
+ sandbox_docker_image: docker.io/library/python:3.13-slim
708
+
709
+ # Wall-clock ceiling for one sandboxed command.
710
+ sandbox_timeout_seconds: 120
711
+
712
+ # Memory ceiling for the sandbox, in MB.
713
+ sandbox_memory_mb: 2048
714
+
715
+ # CPU-time ceiling for the sandbox, in seconds.
716
+ sandbox_cpu_seconds: 30
717
+
718
+ # Process ceiling for the sandbox.
719
+ sandbox_max_pids: 200
720
+
721
+ # Background and monitor shell sessions (ROADMAP_v3 §49, SS12-SS13).
722
+ #
723
+ # The longest a session may run. An agent asking for more is started with
724
+ # this and told so -- never refused.
725
+ shell_session_timeout_cap_s: 3600
726
+
727
+ # How long an INTERACTIVE shell may sit with nothing typed into it
728
+ # before it is closed (ROADMAP_v3 §49, SS25). Unlike a background
729
+ # command, an interactive session does not block your prompt, so this is
730
+ # the only thing that stops a forgotten shell holding a container open
731
+ # until its timeout cap. Measured from the last input, not the last
732
+ # output: a program printing to itself is still one nobody is using.
733
+ shell_session_idle_timeout_s: 600
734
+
735
+ # How many sessions of any kind may be live at once, counted across the
736
+ # main agent and every subagent. Each is its own container under the
737
+ # memory ceiling above. An interactive shell counts here -- it holds a
738
+ # container like any other -- even though it does not block your prompt.
739
+ shell_session_max_live: 4
740
+
741
+ # How many times in a row finished sessions may wake the agent with no
742
+ # message from you. Past it, waking stops, the prompt unblocks, and the
743
+ # results wait for your next message.
744
+ shell_session_max_consecutive_wakes: 10
745
+
746
+ # How much of a session's output is kept, in memory only and never on
747
+ # disk: its first characters and its most recent ones. The middle is
748
+ # dropped, and the agent is told where.
749
+ shell_session_output_head_chars: 10000
750
+ shell_session_output_tail_chars: 190000
751
+
752
+ # ======================================================================
753
+ # Output redaction and tool isolation
754
+ # ======================================================================
755
+
756
+ # AUTHORITY -- the master switch for pattern-based redaction of everything
757
+ # leaving a tool, and part of the frozen posture. There is deliberately no
758
+ # settings.json key. VENASTINE_REDACT_OFF can only ever turn redaction
759
+ # OFF for a run, never on, and is bound at the same moment as this.
760
+ redact_tool_outputs: true
761
+
762
+ # Wall clock for one BUDGET_COMPUTE tool call, and the CPU ceiling handed
763
+ # to the isolated child process that runs it.
764
+ tool_compute_timeout_s: 20
765
+
766
+ # First words that make a sandboxed command need the network. Anything not
767
+ # listed runs with networking off.
768
+ network_allowed_commands:
769
+ - pip
770
+ - pip3
771
+ - curl
772
+ - wget
773
+ - git
774
+ - npm
775
+ - yarn
776
+ - pnpm
777
+ - apt
778
+ - apt-get
779
+ - yum
780
+ - dnf
781
+ - brew
782
+ - cargo
783
+
784
+ # Read-only inspection commands eligible for the auto-approved inert tier
785
+ # under `shell_approval_mode: tiered`. Sound as a bare token list only
786
+ # because that tier executes argv with no shell, so the tokens ARE the
787
+ # arguments.
788
+ #
789
+ # WHAT BELONGS HERE, after batch 117 found two things that did not. A
790
+ # command qualifies only if it cannot write, and "cannot write" has to hold
791
+ # for its ARGUMENTS as well as its name:
792
+ #
793
+ # * A writing FLAG is guarded by `_DANGEROUS_FLAGS` in security/sandbox.py
794
+ # -- `tree -o FILE` and `date -s TIME` were inert with no entry there
795
+ # until batch 117, so the tier reported `writes=False` about a write.
796
+ # * A writing POSITIONAL cannot be guarded at all. `uniq IN OUT` writes to
797
+ # its second argument with no flag to catch, so `uniq` was REMOVED from
798
+ # this list rather than given an entry that could not work. Teaching the
799
+ # classifier which positions are outputs would make it a shell parser,
800
+ # which security/sandbox.py's own G2 argument forbids.
801
+ #
802
+ # The bound that still held throughout: `_escapes_workspace` reads every
803
+ # token after the first as a path, so a write aimed outside the workspace
804
+ # was refused regardless. What was wrong was the profile's claim, not the
805
+ # containment.
806
+ inert_commands:
807
+ - ls
808
+ - cat
809
+ - pwd
810
+ - echo
811
+ - grep
812
+ - wc
813
+ - head
814
+ - tail
815
+ - date
816
+ - whoami
817
+ - uname
818
+ - which
819
+ - where
820
+ - type
821
+ - file
822
+ - stat
823
+ - du
824
+ - df
825
+ - tree
826
+ - diff
827
+ - md5sum
828
+ - sha256sum
829
+
830
+ # ======================================================================
831
+ # Research grants and review
832
+ # ======================================================================
833
+
834
+ # Ceiling on granted tool calls in one run, across every grant.
835
+ max_granted_tool_calls: 150
836
+
837
+ # How long an attended approval waits for an answer before giving up on
838
+ # it.
839
+ attended_approval_timeout_s: 600
840
+
841
+ # Whether a finished research run is reviewed. A settings.json
842
+ # `research.subagent_review` outranks this; --no-review escapes it for one
843
+ # run.
844
+ subagent_review: false
845
+
846
+ # Findings kept from one review; the rest are dropped with a warning.
847
+ max_review_findings: 25
848
+
849
+ # Rounds a single finding may be refined before it is recorded as
850
+ # rejected.
851
+ max_review_refinements: 3
852
+
853
+ # ======================================================================
854
+ # Compaction
855
+ # ======================================================================
856
+
857
+ # How much of the model's context window a thread may fill before
858
+ # compaction folds it. What the fraction really names is the HEADROOM --
859
+ # the remainder, which the step after the trigger fires has to fit inside,
860
+ # because nothing in the harness recovers from a prompt that overflows the
861
+ # window. A bare share gets that wrong at both ends of the roster: 0.10 of
862
+ # 1M is 100k and 0.10 of 64k is 6.4k, while ONE capped tool read is 12,500
863
+ # proxy tokens. Hence the two clamps below.
864
+ compaction_trigger_fraction: 0.90
865
+
866
+ # The headroom never shrinks below this, however small the window. Sized
867
+ # from max_read_chars (50,000 chars = 12,500 proxy tokens, the largest a
868
+ # single bounded read can return) plus the assistant turn that asked for
869
+ # it. Binds under a ~160k window and nowhere above one, so a 262k or 1M
870
+ # thread gets exactly the fraction above and a 64k thread keeps a usable
871
+ # margin instead of 6.4k.
872
+ compaction_min_headroom_tokens: 16000
873
+
874
+ # ...and never grows past this share of the window, which is what stops
875
+ # the floor above from eating a small window whole -- 16k of a 32k window
876
+ # is half of it. This is the clamp that keeps the trigger positive and
877
+ # above keep_recent_tokens at the bottom of the range.
878
+ compaction_max_headroom_fraction: 0.25
879
+
880
+ # The trigger used when NO MODEL IS IN SCOPE. A settings.json
881
+ # `compaction.trigger_tokens` overrides it.
882
+ compaction_trigger_tokens: 40000
883
+
884
+ # The early warning fires this many tokens BEFORE the trigger. Must stay
885
+ # strictly below the trigger, or the warning arrives after the thing it
886
+ # warns about; the loader refuses a pair that cannot work.
887
+ compaction_warning_margin_tokens: 8000
888
+
889
+ # Recent conversation held out of a fold verbatim.
890
+ compaction_keep_recent_tokens: 4000
891
+
892
+ # Recent turns held out of a fold verbatim, whichever bound binds first.
893
+ compaction_keep_recent_turns: 3
894
+
895
+ # The share of the trigger that pinned content may occupy, so a pin cannot
896
+ # force permanent compaction.
897
+ pin_max_trigger_fraction: 0.5
898
+
899
+ # Default fold strength: one of the keys of `compaction_target_ratios`.
900
+ compaction_strength: 3
901
+
902
+ # Attempts at one fold before it is abandoned.
903
+ compaction_max_retries: 2
904
+
905
+ # Strength -> the fraction of the original character count a fold aims
906
+ # for. INTEGER keys, and they are the closed set a settings.json
907
+ # `compaction.strength` is checked against.
908
+ compaction_target_ratios:
909
+ 1: 0.4
910
+ 2: 0.25
911
+ 3: 0.15
912
+ 4: 0.1
913
+ 5: 0.05
914
+
915
+ # How far over target a fold may land before it is retried.
916
+ compaction_ratio_tolerance: 0.5
917
+
918
+ # Default strategy.
919
+ # chain -- fold the previous summary together with the new span
920
+ # rederive -- re-summarise the whole thread from the archive
921
+ compaction_strategy: chain
922
+
923
+ # The closed vocabulary `compaction_strategy` and any settings.json
924
+ # `compaction.strategy` are checked against.
925
+ compaction_strategies:
926
+ - rederive
927
+ - chain
928
+
929
+ # Memories injected into a prompt at once. 0 disables the `remember` tool
930
+ # rather than injecting nothing.
931
+ max_injected_memories: 50
932
+
933
+ # ======================================================================
934
+ # Catalog text
935
+ # ======================================================================
936
+
937
+ # Cap on a skill's or agent's `name` and `description` -- the two strings
938
+ # that reach every run's system prompt from a .md file with no tool call
939
+ # and no further consent. The newline collapse that comes with it matters
940
+ # more than the number.
941
+ max_catalog_text_chars: 300
942
+
943
+ # ======================================================================
944
+ # Summaries, references and the checklist
945
+ # ======================================================================
946
+
947
+ # The agent that does the summarising, for both folds and /summary.
948
+ compactor_agent: compactor
949
+
950
+ # How long a whole-thread summary may be, in characters. ABSOLUTE rather
951
+ # than a ratio: this summary is injected into every turn of a REFERENCING
952
+ # thread, where nothing it replaces bounds it.
953
+ summary_target_chars: 2000
954
+
955
+ # Referenced threads attachable to one thread at once. REFUSES rather than
956
+ # dropping the oldest, because a reference is something a person named.
957
+ max_injected_refs: 3
958
+
959
+ # The checklist vocabulary, and the enum in todo_write's own schema. A
960
+ # plain list rather than an enum so a list stored under an older
961
+ # vocabulary reads back as data instead of raising.
962
+ todo_statuses:
963
+ - pending
964
+ - in_progress
965
+ - completed
966
+
967
+ # Checklist ceiling. REFUSES rather than truncating: a list silently cut
968
+ # would leave the model believing it had recorded work it has forgotten.
969
+ max_todo_items: 50
970
+
971
+ # Research passes compact only this far from the model's actual context
972
+ # window -- a safety net against a hard provider error mid-pipeline, not a
973
+ # working-set policy.
974
+ compaction_pipeline_backstop_tokens: 20000
975
+
976
+ # ======================================================================
977
+ # Context windows
978
+ # ======================================================================
979
+
980
+ # Model -> input window, in tokens. A FALLBACK now, not the only source:
981
+ # core/client.py queries the provider where the SDK exposes it, and
982
+ # core/model_windows.py holds the user's own `/window` answer, which
983
+ # outranks both. A miss here warns once per model.
984
+ model_context_windows:
985
+ claude-opus-5: 1000000
986
+ claude-sonnet-5: 1000000
987
+ claude-fable-5: 1000000
988
+ claude-opus-4-8: 1000000
989
+ claude-opus-4-7: 1000000
990
+ claude-opus-4-6: 1000000
991
+ claude-sonnet-4-6: 1000000
992
+ claude-haiku-4-5: 200000
993
+ gpt-5.1: 400000
994
+ gemini-2.5-pro: 1000000
995
+ qwen3.8-max-preview: 983616
996
+
997
+ # For a model absent from the table above and unqueryable. Logged at
998
+ # WARNING once per model, because a wrong window now decides when every
999
+ # thread folds.
1000
+ default_context_window: 256000
1001
+
1002
+ # ======================================================================
1003
+ # Connecting a model
1004
+ # ======================================================================
1005
+
1006
+ # The Hugging Face OAuth app /connect huggingface signs in through, by its
1007
+ # client id -- a PUBLIC value, not a secret. This one is the project's own
1008
+ # app, registered without a secret, with the device code flow and the scopes
1009
+ # `openid profile inference-api` (ROADMAP_v3 §79, HD5). Another app's id
1010
+ # signs in through that app instead; empty hides the sign-in and leaves a
1011
+ # pasted token working.
1012
+ hf_oauth_client_id: "3a823316-767d-4abe-8e55-6c0354ce2c27"
1013
+
1014
+ # ======================================================================
1015
+ # Tool permissions and approvals
1016
+ # ======================================================================
1017
+
1018
+ # AUTHORITY -- whether each registered tool may be called AT ALL.
1019
+ # is_tool_allowed() reads this before any context or approval logic and a
1020
+ # tool context can never widen it, so a false here cannot be re-enabled at
1021
+ # runtime by an agent, a skill or a settings file -- only by editing this
1022
+ # file. read / view_image / write / edit / shell ship false for that reason,
1023
+ # and so do the five background-session tools after shell: a session is a
1024
+ # shell command that keeps running, and enabling one without the others is
1025
+ # allowed but rarely useful. view_image follows read because it reads a file
1026
+ # off disk the same way; it also needs a provider marked `accepts_images` in
1027
+ # providers.json before it is offered at all (ROADMAP_v3 §81, IM6).
1028
+ # Every registered tool must appear in BOTH tables or the harness refuses
1029
+ # to start (D24).
1030
+ tool_permissions:
1031
+ web_search: true
1032
+ fetch_url: true
1033
+ get_time: true
1034
+ arxiv_search: true
1035
+ symbolic_math: true
1036
+ linear_algebra: true
1037
+ probability_stats: true
1038
+ discrete_math: true
1039
+ logic: true
1040
+ geometry: true
1041
+ read: false
1042
+ view_image: false
1043
+ write: false
1044
+ edit: false
1045
+ shell: false
1046
+ shell_background: false
1047
+ shell_monitor: false
1048
+ shell_interactive: false
1049
+ shell_input: false
1050
+ shell_sessions: false
1051
+ shell_output: false
1052
+ shell_kill: false
1053
+ load_skill: true
1054
+ spawn_subagent: true
1055
+ pin: true
1056
+ unpin: true
1057
+ remember: true
1058
+ read_project_doc: true
1059
+ write_project_doc: true
1060
+ ask_user: true
1061
+ todo_write: true
1062
+
1063
+ # AUTHORITY -- whether each tool needs a human yes before it runs. Approval ORs
1064
+ # across every layer, so a true here is a one-way ratchet that can only
1065
+ # add prompts. `shell`, `shell_background`, `shell_monitor` and
1066
+ # `shell_interactive` are deliberately absent: their approval is governed
1067
+ # solely by `shell_approval_mode` above -- there is no second switch, so
1068
+ # there is nothing here that can disagree with it. Listing, reading and
1069
+ # stopping a session ask nothing: they act only on this conversation's own
1070
+ # sessions. `shell_input` is here and ships false because approving the
1071
+ # shell approved what you type into it (SS29) -- set it true to be asked
1072
+ # about every line instead.
1073
+ # A tool gated here is not merely denied where nothing can ask -- it stops
1074
+ # being advertised at all, which is why ask_user, pin and todo_write are
1075
+ # ungated.
1076
+ tool_approvals:
1077
+ web_search: false
1078
+ fetch_url: false
1079
+ get_time: false
1080
+ arxiv_search: false
1081
+ symbolic_math: false
1082
+ linear_algebra: false
1083
+ probability_stats: false
1084
+ discrete_math: false
1085
+ logic: false
1086
+ geometry: false
1087
+ read: false
1088
+ view_image: false
1089
+ write: false
1090
+ edit: false
1091
+ shell_input: false
1092
+ shell_sessions: false
1093
+ shell_output: false
1094
+ shell_kill: false
1095
+ load_skill: false
1096
+ spawn_subagent: true
1097
+ pin: false
1098
+ unpin: false
1099
+ remember: true
1100
+ read_project_doc: false
1101
+ write_project_doc: true
1102
+ ask_user: false
1103
+ todo_write: false