@yahyahammad/venastine 0.2.0 → 0.3.8

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.
Files changed (129) hide show
  1. package/.env.example +19 -14
  2. package/CONFIG_ARCHITECTURE.md +2002 -0
  3. package/NOTICE +1 -1
  4. package/README.md +233 -94
  5. package/agents/builtin/build.md +73 -0
  6. package/agents/builtin/general.md +80 -0
  7. package/agents/builtin/plan.md +23 -5
  8. package/agents/builtin/test.md +65 -0
  9. package/agents/builtin/writer.md +67 -0
  10. package/agents/manager.py +261 -195
  11. package/agents/subagent_tool.py +277 -43
  12. package/agents/tui_commands.py +7 -12
  13. package/bin/venastine.mjs +528 -382
  14. package/bin/venastine.py +71 -0
  15. package/cli_commands.py +835 -0
  16. package/config.py +95 -1198
  17. package/config.yaml +1090 -0
  18. package/config_edit.py +1301 -0
  19. package/config_schema.py +966 -0
  20. package/config_update.py +749 -0
  21. package/core/agent_activity.py +158 -20
  22. package/core/ansi.py +452 -0
  23. package/core/approval.py +27 -5
  24. package/core/call_view.py +364 -0
  25. package/core/cancel.py +121 -0
  26. package/core/chatgpt_auth.py +747 -0
  27. package/core/claims_view.py +107 -0
  28. package/core/client.py +1238 -82
  29. package/core/commands.py +454 -0
  30. package/core/compaction.py +830 -66
  31. package/core/config_loader.py +516 -97
  32. package/core/connect.py +686 -0
  33. package/core/events.py +68 -6
  34. package/core/file_mentions.py +280 -0
  35. package/core/interaction.py +87 -4
  36. package/core/line_pattern.py +128 -0
  37. package/core/loop.py +1516 -440
  38. package/core/memory.py +120 -8
  39. package/core/pipeline_models.py +23 -9
  40. package/core/provider_errors.py +234 -0
  41. package/core/reasoning/confidence_scoring.py +70 -54
  42. package/core/reasoning/json_retry.py +19 -1
  43. package/core/reasoning/orchestrator.py +232 -53
  44. package/core/reasoning/output_writer.py +90 -31
  45. package/core/reasoning/payload_validation.py +133 -7
  46. package/core/reasoning/pipeline_storage.py +427 -361
  47. package/core/reasoning/review.py +659 -615
  48. package/core/reasoning/scholar.py +38 -14
  49. package/core/reasoning/source_corpus.py +110 -4
  50. package/core/replay.py +158 -26
  51. package/core/session.py +36 -12
  52. package/core/session_view.py +259 -0
  53. package/core/session_wake.py +259 -0
  54. package/core/shell_sessions.py +1553 -0
  55. package/core/workspace_trust.py +1 -1
  56. package/credentials.py +308 -180
  57. package/database.py +82 -1
  58. package/{PRIVACY.md → docs/PRIVACY.md} +19 -7
  59. package/docs/SECURITY.md +108 -0
  60. package/docs/THIRD_PARTY_NOTICES.md +148 -0
  61. package/env_secrets.py +118 -77
  62. package/json_store.py +236 -16
  63. package/logging_setup.py +121 -11
  64. package/main.py +1671 -199
  65. package/mcp_client/client.py +257 -23
  66. package/mcp_client/config.py +484 -379
  67. package/mcp_client/registration.py +163 -2
  68. package/memories/manager.py +15 -0
  69. package/memories/tui_commands.py +7 -12
  70. package/package.json +234 -4
  71. package/project_init/config_files.py +4 -0
  72. package/project_init/generator.py +201 -37
  73. package/project_init/manifest.py +418 -406
  74. package/project_init/tui_commands.py +7 -4
  75. package/prompts/revalidate.md +1 -1
  76. package/prompts/revise.md +17 -15
  77. package/prompts/system_prompts.py +64 -10
  78. package/providers.json.example +94 -92
  79. package/pyproject.toml +51 -2
  80. package/requirements.txt +252 -29
  81. package/safety/policy_enforcement.py +353 -17
  82. package/security/askpass.py +419 -0
  83. package/security/capability.py +37 -7
  84. package/security/confine.py +484 -0
  85. package/security/hosted.py +182 -0
  86. package/security/permissions.py +54 -7
  87. package/security/posture.py +124 -12
  88. package/security/protected_paths.py +198 -6
  89. package/security/sandbox.py +3262 -139
  90. package/security/secrets.py +855 -0
  91. package/skills/tui_commands.py +3 -4
  92. package/storage.py +368 -35
  93. package/tools/base.py +465 -305
  94. package/tools/builtin/_math_common.py +306 -16
  95. package/tools/builtin/_net_common.py +267 -61
  96. package/tools/builtin/arxiv.py +681 -223
  97. package/tools/builtin/ask_user.py +231 -213
  98. package/tools/builtin/discrete_math.py +35 -4
  99. package/tools/builtin/fetch_url.py +239 -21
  100. package/tools/builtin/file_ops.py +341 -33
  101. package/tools/builtin/geometry.py +184 -148
  102. package/tools/builtin/get_time.py +2 -2
  103. package/tools/builtin/linear_algebra.py +66 -23
  104. package/tools/builtin/load_skill.py +10 -0
  105. package/tools/builtin/logic.py +171 -12
  106. package/tools/builtin/pin.py +46 -12
  107. package/tools/builtin/probability_stats.py +224 -228
  108. package/tools/builtin/project_docs.py +80 -21
  109. package/tools/builtin/remember.py +21 -4
  110. package/tools/builtin/shell.py +586 -67
  111. package/tools/builtin/shell_sessions.py +613 -0
  112. package/tools/builtin/symbolic_math.py +80 -10
  113. package/tools/builtin/todo.py +5 -3
  114. package/tools/builtin/web_search.py +329 -151
  115. package/tools/context.py +23 -0
  116. package/tools/isolation.py +387 -314
  117. package/tools/registry.py +347 -30
  118. package/tui/app.py +4645 -438
  119. package/tui/app.tcss +185 -2
  120. package/tui/commands.py +10 -154
  121. package/tui/diffs.py +286 -265
  122. package/tui/hosted_driver.py +102 -0
  123. package/tui/markdown.py +60 -10
  124. package/tui/meters.py +103 -7
  125. package/tui/screens.py +346 -92
  126. package/tui/themes.py +59 -19
  127. package/tui/widgets.py +1575 -111
  128. package/SECURITY.md +0 -88
  129. package/THIRD_PARTY_NOTICES.md +0 -100
@@ -19,7 +19,9 @@ a translation -- but it is still done HERE, once, rather than teaching
19
19
  core/client.py about MCP types.
20
20
  """
21
21
 
22
+ import json
22
23
  import logging
24
+ import re
23
25
 
24
26
  from security import permissions
25
27
  from tools.base import BUDGET_IO, GRANT_ANYWHERE, ToolSpec
@@ -28,6 +30,50 @@ logger = logging.getLogger(__name__)
28
30
 
29
31
  NAME_PREFIX = "mcp__"
30
32
 
33
+ # WHAT A PROVIDER WILL ACCEPT AS A TOOL NAME. Anthropic, OpenAI and Google
34
+ # all publish the same rule -- `^[a-zA-Z0-9_-]{1,64}$` -- and BOTH halves
35
+ # of the composed name are outside this repository's control: the server
36
+ # half is a free-form key in mcp.json, the tool half comes from the server.
37
+ # Verified against the installed mcp==2.0.0, and again on 2.2.0 and 2.3.0:
38
+ # Tool.model_fields["name"] is a bare required `str` with empty metadata --
39
+ # no pattern, no max length.
40
+ #
41
+ # The prefix spends 7 of the 64 before either name starts.
42
+ #
43
+ # Checked HERE because this is the last place the fact is knowable:
44
+ # registration is the one function that sees the server name and the tool
45
+ # name together, and neither core/client.py (which translates schemas) nor
46
+ # tools/registry.py (which dispatches) has a reason to know what a
47
+ # provider will accept. All 31 of this harness's own tool names satisfy the
48
+ # rule, which is what makes it a bound on third-party names rather than a
49
+ # constraint the project would have to live down to.
50
+ MAX_TOOL_NAME_CHARS = 64
51
+ _NAME_CHAR = re.compile(r"[A-Za-z0-9_-]")
52
+
53
+ # THE TWO BOUNDS ON THIRD-PARTY TEXT IN AN ADVERTISED SCHEMA. Both are
54
+ # paid on EVERY model call for the rest of the session, because
55
+ # registry.schemas() is recomputed at the top of each one (K7) -- which is
56
+ # the difference between these and the bound on a tool RESULT (A79),
57
+ # paid once per call.
58
+ #
59
+ # Deliberately NOT config.MAX_CATALOG_TEXT_CHARS. 300 is right for a
60
+ # one-line catalog bullet and would mutilate a legitimate tool
61
+ # description; sharing the constant would couple two bounds that answer
62
+ # different questions. 2000 is 2.5x the longest description this harness
63
+ # ships (786, `shell_interactive`) and 25x smaller than the 50 KB a probe
64
+ # drove through here.
65
+ MAX_TOOL_DESCRIPTION_CHARS = 2000
66
+
67
+ # And the whole advertised schema, because `description` is not the only
68
+ # third-party string in it: input_schema arrives from the server untouched
69
+ # and a property description is model-facing text in the same request.
70
+ # Capping the strings INSIDE it was declined -- a truncated enum or
71
+ # pattern is a tool that fails at call time instead of load time -- so an
72
+ # oversized schema loses the tool instead, by name, which is the outcome
73
+ # shape the collision branch below already has. 16384 is ~3.6x the largest
74
+ # schema this harness ships (4544, `shell`).
75
+ MAX_TOOL_SCHEMA_CHARS = 16384
76
+
31
77
  # server name -> the exact tool names registered for it.
32
78
  #
33
79
  # THE PREFIX IS NOT A PARSER. `mcp__<server>__<tool>` cannot be taken
@@ -49,6 +95,14 @@ NAME_PREFIX = "mcp__"
49
95
  # Recording what was actually registered fixes both without rejecting any
50
96
  # mcp.json content, which decision G rules out: validation here is by
51
97
  # CONNECTING, not by schema.
98
+ #
99
+ # THAT SENTENCE IS ABOUT THESE TWO HAZARDS AND NOT ABOUT NAMES IN GENERAL,
100
+ # and batch 150 is where the difference started to matter. A name that
101
+ # COLLIDES is recorded rather than rejected, as above. A name no provider
102
+ # will accept is refused at registration (A30), which is not up-front
103
+ # schema validation of mcp.json -- nothing below reads the file -- and is
104
+ # decision G's own remedy: that one server's named failure rather than a
105
+ # startup error for everything else.
52
106
  _registered: dict = {}
53
107
 
54
108
 
@@ -56,6 +110,70 @@ def tool_name_for(server_name: str, tool_name: str) -> str:
56
110
  return f"{NAME_PREFIX}{server_name}__{tool_name}"
57
111
 
58
112
 
113
+ def _illegal_chars(text: str) -> list:
114
+ return sorted({c for c in text if not _NAME_CHAR.match(c)})
115
+
116
+
117
+ def _unadvertisable(name: str) -> str:
118
+ """Why no provider will accept this as a tool name, or "".
119
+
120
+ A REASON rather than a boolean, because it is going into a log line
121
+ somebody has to act on: the server is theirs to rename and the tool is
122
+ not, and which one they can fix depends on which half is wrong.
123
+ """
124
+ if not name:
125
+ return "it is empty"
126
+ if len(name) > MAX_TOOL_NAME_CHARS:
127
+ return (f"it is {len(name)} characters and the limit is "
128
+ f"{MAX_TOOL_NAME_CHARS}")
129
+ bad = _illegal_chars(name)
130
+ if bad:
131
+ return ("it contains " + ", ".join(repr(c) for c in bad)
132
+ + "; only letters, digits, '_' and '-' are accepted")
133
+ return ""
134
+
135
+
136
+ def _bounded(text: str) -> str:
137
+ """One line, capped -- A5's two rules on a string that enters every
138
+ model call, applied at the producer.
139
+
140
+ The same shape as core/config_loader._catalog_text, including the
141
+ VISIBLE ellipsis (a model reading a cut-off description should be able
142
+ to tell), with its own constant for the reason stated above. A5's
143
+ security half applies here too, one step removed: a tool schema is a
144
+ JSON array rather than a rendered list, so a newline cannot leave its
145
+ bullet the way #131's did -- but the newlines still arrive, and what
146
+ they can do then depends on a consumer. That dependence is what A5
147
+ exists to remove, and this project has already paid for it once at the
148
+ consumer end: core/reasoning/authorization.py takes
149
+ `description.strip().splitlines()[0]` for the grant picker, which is a
150
+ consumer defending itself against this producer.
151
+ """
152
+ collapsed = " ".join(str(text).split())
153
+ if len(collapsed) <= MAX_TOOL_DESCRIPTION_CHARS:
154
+ return collapsed
155
+ return collapsed[:MAX_TOOL_DESCRIPTION_CHARS - 1].rstrip() + "\u2026"
156
+
157
+
158
+ def _too_big(schema: dict) -> str:
159
+ """Why this schema cannot be advertised, or "".
160
+
161
+ Measured on the SERIALISED form, because that is what is paid: the
162
+ dict goes to the provider as JSON on every call. A schema that will
163
+ not serialise at all is refused by the same gate -- it could not be
164
+ sent either way, and finding that out here names the server instead
165
+ of raising inside the provider branch in core/client.py.
166
+ """
167
+ try:
168
+ size = len(json.dumps(schema))
169
+ except (TypeError, ValueError) as e:
170
+ return f"its schema is not JSON-serialisable ({e})"
171
+ if size > MAX_TOOL_SCHEMA_CHARS:
172
+ return (f"its schema is {size} characters and the limit is "
173
+ f"{MAX_TOOL_SCHEMA_CHARS}")
174
+ return ""
175
+
176
+
59
177
  def schema_for(server_name: str, tool) -> dict:
60
178
  """MCP Tool -> the Anthropic-native schema shape every TOOL_SCHEMA in
61
179
  tools/builtin/ uses.
@@ -77,7 +195,11 @@ def schema_for(server_name: str, tool) -> dict:
77
195
  description = f"{title}. {description}".strip()
78
196
  return {
79
197
  "name": tool_name_for(server_name, tool.name),
80
- "description": f"[MCP: {server_name}] {description}".strip(),
198
+ # BOUNDED AFTER the prefix and the title are composed in, so one
199
+ # call covers all three strings: `title` is server-supplied too
200
+ # and reaches the model through this field, which a cap applied
201
+ # before the concatenation would have missed.
202
+ "description": _bounded(f"[MCP: {server_name}] {description}"),
81
203
  "input_schema": input_schema,
82
204
  }
83
205
 
@@ -98,6 +220,20 @@ def register_all(client, registry) -> list:
98
220
  for server_name in sorted(client.clients):
99
221
  cfg = client.server_configs.get(server_name)
100
222
  auto_approve = bool(getattr(cfg, "auto_approve", False))
223
+ # ONE WARNING FOR THE SERVER, not one per tool: a character the
224
+ # rule rejects in the SERVER half poisons every name it composes,
225
+ # and `"my.server"` or `"api.example.com"` is an ordinary
226
+ # mcp.json key. The tool half is checked per tool below, where
227
+ # only that tool is lost.
228
+ bad_chars = _illegal_chars(server_name)
229
+ if bad_chars:
230
+ logger.warning(
231
+ "MCP server %r can compose no advertisable tool name: its "
232
+ "name contains %s, and a provider accepts only letters, "
233
+ "digits, '_' and '-' in a tool name. Rename the server in "
234
+ "mcp.json; none of its tools are registered.",
235
+ server_name, ", ".join(repr(c) for c in bad_chars))
236
+ continue
101
237
  try:
102
238
  tools = client.list_tools(server_name)
103
239
  except Exception as e:
@@ -107,6 +243,31 @@ def register_all(client, registry) -> list:
107
243
 
108
244
  for tool in tools:
109
245
  name = tool_name_for(server_name, tool.name)
246
+ # REFUSE, DO NOT MANGLE (A30). A silently renamed tool is one
247
+ # the model cannot name back, and the prefix is not a parser,
248
+ # so a generated name maps to no server. Refusing is also the
249
+ # only outcome that keeps this module's own invariant the
250
+ # right way up: "one unreachable server must never take down
251
+ # the others" has an inversion, and advertising a name no
252
+ # provider accepts is it -- one list comprehension builds the
253
+ # whole `tools` array, so one bad name makes EVERY request
254
+ # malformed, in plain chat and all ten research passes, and a
255
+ # 400 is not transient so nothing retries it.
256
+ #
257
+ # Within decision G rather than against it: G says an entry we
258
+ # cannot make sense of becomes that one server's NAMED
259
+ # failure instead of a startup error for everything else,
260
+ # which is exactly what this is. What G rules out is
261
+ # up-front SCHEMA validation of mcp.json, and nothing here
262
+ # reads the file.
263
+ schema = schema_for(server_name, tool)
264
+ refusal = _unadvertisable(name) or _too_big(schema)
265
+ if refusal:
266
+ logger.warning(
267
+ "MCP tool %r from server %r cannot be advertised: %s. "
268
+ "Skipping it; the server's other tools are unaffected.",
269
+ name, server_name, refusal)
270
+ continue
110
271
  if name in registry._tools:
111
272
  # Never overwrite. The colliding registration would replace
112
273
  # BOTH the handler and (via set_dynamic_approval_default
@@ -124,7 +285,7 @@ def register_all(client, registry) -> list:
124
285
  permissions.set_dynamic_approval_default(name, not auto_approve)
125
286
  registry.register(ToolSpec(
126
287
  name=name,
127
- schema=schema_for(server_name, tool),
288
+ schema=schema,
128
289
  handler=_make_handler(client, server_name, tool.name),
129
290
  # §25 R13. Stated here rather than left to
130
291
  # registry.grant_policy's mcp__* fallback: this is the
@@ -103,6 +103,15 @@ class MemoryManager:
103
103
  truncated list can say so, while a log line reaches nobody
104
104
  mid-conversation. Same "no silent caps" rule as §20's
105
105
  MAX_REVIEW_FINDINGS.
106
+
107
+ A CAP OF 0 MEANS OFF, IN THIS HALF TOO (batch 154). config.yaml
108
+ says so where the value is set, and `remember`'s available_check
109
+ already honours it by hiding the tool. This half did not: `rows[:0]`
110
+ is empty, `total > 0` fires the truncation branch, and every turn
111
+ got a "## Remembered" header announcing "the 0 most recent of 2"
112
+ with a WARNING behind it -- telling the model that remembered
113
+ content exists and is being withheld, while the tool that could
114
+ have earned it is not on the menu.
106
115
  """
107
116
  if not self.wants_memories(agent):
108
117
  return ""
@@ -111,6 +120,12 @@ class MemoryManager:
111
120
  return ""
112
121
 
113
122
  cap = config.MAX_INJECTED_MEMORIES
123
+ if cap <= 0:
124
+ # `<= 0` because the sentence is "a non-positive cap is off".
125
+ # Only 0 is reachable -- config_schema types this key
126
+ # NonNegativeInt -- so the other half of the comparison is the
127
+ # schema's job and not this line's.
128
+ return ""
114
129
  shown, total = rows[:cap], len(rows)
115
130
  header = "## Remembered"
116
131
  if total > cap:
@@ -20,8 +20,9 @@ something the user relies on.
20
20
 
21
21
  from uuid import UUID
22
22
 
23
- from tui.commands import SlashCommand, registry as commands
24
23
  from memories.manager import manager
24
+ from tui.commands import bind
25
+ from tui.commands import registry as commands
25
26
 
26
27
 
27
28
  def _cmd_memories(app, args: str) -> None:
@@ -129,9 +130,9 @@ def _cmd_ref(app, args: str) -> None:
129
130
  steer another without their say-so. A model-callable `search_threads`
130
131
  stays possible later, on this same retrieval and behind D26's gate.
131
132
  """
133
+ import storage
132
134
  from core.loop import clear_refs
133
135
  from tui.screens import ThreadPickerScreen
134
- import storage
135
136
 
136
137
  flag = args.strip().lower()
137
138
  if flag in ("--list", "list"):
@@ -232,13 +233,7 @@ def register_memory_commands() -> None:
232
233
  §21c's two live here with §21b's rather than in a module of their own:
233
234
  they are the same section's family, and `/ref` reads what `/summary`
234
235
  writes."""
235
- commands.register(SlashCommand(
236
- "memories", "list durable memories visible here", _cmd_memories))
237
- commands.register(SlashCommand(
238
- "forget", "delete one durable memory", _cmd_forget, "<id>"))
239
- commands.register(SlashCommand(
240
- "summary", "distil this conversation without folding it",
241
- _cmd_summary))
242
- commands.register(SlashCommand(
243
- "ref", "reference another conversation's summary", _cmd_ref,
244
- "[--list|--clear]"))
236
+ commands.register(bind("memories", _cmd_memories))
237
+ commands.register(bind("forget", _cmd_forget))
238
+ commands.register(bind("summary", _cmd_summary))
239
+ commands.register(bind("ref", _cmd_ref))
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@yahyahammad/venastine",
3
- "version": "0.2.0",
3
+ "version": "0.3.8",
4
4
  "description": "A from-scratch agentic harness: tool-using chat plus a ten-pass self-critiquing research pipeline.",
5
5
  "keywords": [
6
6
  "agent",
@@ -34,10 +34,239 @@
34
34
  "prepublishOnly": "node scripts/prepublish-check.mjs",
35
35
  "check-pack": "node scripts/prepublish-check.mjs"
36
36
  },
37
+ "neverDistribute": [
38
+ {
39
+ "label": "provider credentials",
40
+ "match": "exact",
41
+ "value": [
42
+ "providers.json"
43
+ ],
44
+ "channels": [
45
+ "git",
46
+ "archive",
47
+ "npm"
48
+ ],
49
+ "samples": [
50
+ "providers.json"
51
+ ]
52
+ },
53
+ {
54
+ "label": "tool secrets",
55
+ "match": "basename",
56
+ "value": [
57
+ ".env"
58
+ ],
59
+ "channels": [
60
+ "git",
61
+ "archive",
62
+ "npm"
63
+ ],
64
+ "samples": [
65
+ ".env",
66
+ "sub/.env"
67
+ ]
68
+ },
69
+ {
70
+ "label": "the harness-held secret store",
71
+ "match": "exact",
72
+ "value": [
73
+ "secrets.json"
74
+ ],
75
+ "channels": [
76
+ "git",
77
+ "archive",
78
+ "npm"
79
+ ],
80
+ "samples": [
81
+ "secrets.json"
82
+ ]
83
+ },
84
+ {
85
+ "label": "ChatGPT plan tokens",
86
+ "match": "exact",
87
+ "value": [
88
+ "chatgpt_auth.json",
89
+ "chatgpt_auth.json.lock"
90
+ ],
91
+ "channels": [
92
+ "git",
93
+ "archive",
94
+ "npm"
95
+ ],
96
+ "samples": [
97
+ "chatgpt_auth.json",
98
+ "chatgpt_auth.json.lock"
99
+ ],
100
+ "note": "ROADMAP_v3 §83 (CG4). The ChatGPT-plan sign-in's access and refresh tokens, written beside providers.json; the .lock is the empty file its refreshes serialize on."
101
+ },
102
+ {
103
+ "label": "local databases and their WAL sidecars",
104
+ "match": "suffix",
105
+ "value": [
106
+ ".db",
107
+ ".db-wal",
108
+ ".db-shm"
109
+ ],
110
+ "channels": [
111
+ "git",
112
+ "archive",
113
+ "npm"
114
+ ],
115
+ "samples": [
116
+ "app.db",
117
+ "app.db-wal",
118
+ "app.db-shm"
119
+ ]
120
+ },
121
+ {
122
+ "label": "logs",
123
+ "match": "dir",
124
+ "value": [
125
+ "logs"
126
+ ],
127
+ "channels": [
128
+ "git",
129
+ "archive",
130
+ "npm"
131
+ ],
132
+ "samples": [
133
+ "logs/app.log"
134
+ ]
135
+ },
136
+ {
137
+ "label": "research output",
138
+ "match": "dir",
139
+ "value": [
140
+ "output"
141
+ ],
142
+ "channels": [
143
+ "git",
144
+ "archive",
145
+ "npm"
146
+ ],
147
+ "samples": [
148
+ "output/run-1/report.md"
149
+ ]
150
+ },
151
+ {
152
+ "label": "compiled python",
153
+ "match": "suffix-or-segment",
154
+ "value": [
155
+ ".pyc",
156
+ "__pycache__"
157
+ ],
158
+ "channels": [
159
+ "git",
160
+ "npm"
161
+ ],
162
+ "samples": [
163
+ "x.pyc",
164
+ "__pycache__/README.txt"
165
+ ],
166
+ "note": "No archive row by decision (S18-08): `git archive` ships only TRACKED content, so export-ignore on a gitignored path is belt and braces. The row that mattered there was .venastine/."
167
+ },
168
+ {
169
+ "label": "notebook checkpoints",
170
+ "match": "segment",
171
+ "value": [
172
+ ".ipynb_checkpoints"
173
+ ],
174
+ "channels": [
175
+ "git",
176
+ "archive",
177
+ "npm"
178
+ ],
179
+ "samples": [
180
+ ".ipynb_checkpoints/system.md"
181
+ ]
182
+ },
183
+ {
184
+ "label": "security scan artifacts",
185
+ "match": "prefix",
186
+ "value": [
187
+ "CLAUDE-SECURITY"
188
+ ],
189
+ "channels": [
190
+ "git",
191
+ "archive",
192
+ "npm"
193
+ ],
194
+ "samples": [
195
+ "CLAUDE-SECURITY-2026-01-01/report.md"
196
+ ]
197
+ },
198
+ {
199
+ "label": "atomic-write debris",
200
+ "match": "suffix",
201
+ "value": [
202
+ ".tmp"
203
+ ],
204
+ "channels": [
205
+ "git",
206
+ "archive",
207
+ "npm"
208
+ ],
209
+ "samples": [
210
+ "providers.json.a1b2c3d4.tmp",
211
+ "secrets.json.a1b2c3d4.tmp"
212
+ ]
213
+ },
214
+ {
215
+ "label": "the unsafe-mode branch marker",
216
+ "match": "exact",
217
+ "value": [
218
+ "UNSAFE_BRANCH"
219
+ ],
220
+ "channels": [
221
+ "npm"
222
+ ],
223
+ "samples": [
224
+ "UNSAFE_BRANCH"
225
+ ],
226
+ "note": "npm only, deliberately. UN5 keeps this marker OUT of the tarball while the unsafe branch must be able to track it, so gitignoring it would break the thing it exists to mark."
227
+ },
228
+ {
229
+ "label": "project MCP server configuration",
230
+ "match": "exact",
231
+ "value": [
232
+ ".venastine/mcp.json"
233
+ ],
234
+ "channels": [
235
+ "git",
236
+ "archive",
237
+ "npm"
238
+ ],
239
+ "samples": [
240
+ ".venastine/mcp.json"
241
+ ],
242
+ "note": "S18-01. `env` and `headers` are live fields and README names `Authorization` as where a hosted server's credential rides. The directory is NOT excluded wholesale: .venastine/agents and .venastine/skills are project content meant to travel with the repo, which is what D17's trust prompt exists for."
243
+ },
244
+ {
245
+ "label": "project settings",
246
+ "match": "exact",
247
+ "value": [
248
+ ".venastine/settings.json"
249
+ ],
250
+ "channels": [
251
+ "git",
252
+ "archive",
253
+ "npm"
254
+ ],
255
+ "samples": [
256
+ ".venastine/settings.json"
257
+ ],
258
+ "note": "The sibling `/init --config` writes in the same call as mcp.json, so it travels with it."
259
+ }
260
+ ],
37
261
  "files": [
38
262
  "bin/",
39
263
  "main.py",
40
264
  "config.py",
265
+ "config.yaml",
266
+ "config_schema.py",
267
+ "config_edit.py",
268
+ "config_update.py",
269
+ "cli_commands.py",
41
270
  "storage.py",
42
271
  "database.py",
43
272
  "credentials.py",
@@ -60,9 +289,10 @@
60
289
  "README.md",
61
290
  "LICENSE",
62
291
  "NOTICE",
63
- "THIRD_PARTY_NOTICES.md",
64
- "SECURITY.md",
65
- "PRIVACY.md",
292
+ "docs/THIRD_PARTY_NOTICES.md",
293
+ "CONFIG_ARCHITECTURE.md",
294
+ "docs/SECURITY.md",
295
+ "docs/PRIVACY.md",
66
296
  "providers.json.example",
67
297
  ".env.example",
68
298
  "!**/__pycache__",
@@ -101,6 +101,10 @@ OMITTED = {
101
101
  "vestigial -- the roster's length decides the count now, and the "
102
102
  "loader warns when this is set, so a template carrying it would "
103
103
  "ship that warning to everyone who ran the command",
104
+ "confidence.disagreement_penalty_factor":
105
+ "retired by ROADMAP_v3 §90 for the same reason as ensemble_n: "
106
+ "disagreement routes a claim to be checked and subtracts nothing, "
107
+ "and the loader warns when this is set",
104
108
  "ensemble_mode":
105
109
  "resolved as `config.ENSEMBLE_MODE if value is None`, so writing "
106
110
  "`false` is not a restatement of the default but a decision to "