@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
@@ -11,6 +11,18 @@ first-order proof assistant -- it can't verify a multi-step natural-
11
11
  deduction proof or reason over quantifiers. What it CAN do is give the
12
12
  model a deterministic ground-truth check for a specific logical claim,
13
13
  which is still real uplift over the model reasoning about it unaided.
14
+
15
+ AND THAT IS WHY THE DOMAIN IS CHECKED (batch 143). `safe_parse` admits
16
+ the whole 230-name allowlist, so `expression` need not be Boolean, and no
17
+ operation used to look. SymPy's `truth_table` substitutes 0/1 and coerces
18
+ the arithmetic value through `bool()`, so `x - 1` was returned as a
19
+ confident NOT table and `x + y` as the truth table of OR -- which has a
20
+ plausible caller, because in GF(2) and in coding theory `x + y` IS XOR.
21
+ `bool()` of an arithmetic value is not a ground-truth check for a logical
22
+ claim; it is a coincidence that agrees with OR for `+` and with AND for
23
+ `*` and diverges elsewhere. A tool whose stated value is that it is
24
+ mechanical has to refuse the inputs it cannot be mechanical about, so
25
+ `_require_boolean` is the paragraph above enforced rather than asserted.
14
26
  """
15
27
 
16
28
  from __future__ import annotations
@@ -18,24 +30,69 @@ from __future__ import annotations
18
30
  from typing import Literal, Optional
19
31
 
20
32
  import sympy
21
- from sympy.logic.boolalg import simplify_logic, truth_table
22
- from sympy.logic.inference import satisfiable
23
33
  from pydantic import BaseModel, Field
34
+ from sympy.logic.boolalg import Boolean, simplify_logic, truth_table
35
+ from sympy.logic.inference import satisfiable
24
36
 
25
- from tools.builtin._math_common import safe_parse, MathParseError
37
+ from tools.builtin._math_common import (
38
+ MathParseError,
39
+ _split_warnings,
40
+ parse_params,
41
+ safe_parse,
42
+ )
26
43
 
27
44
  _OPERATIONS = Literal["simplify", "equivalent", "satisfiable", "tautology", "truth_table"]
28
45
 
46
+ # THE OUTPUT BOUND (batch 143), on the INPUT rather than on the rendered
47
+ # output, because this is the one math operation whose size is predictable
48
+ # before anything is computed: `truth_table` materialises 2**n rows into a
49
+ # list where n comes from a model-supplied string. H5's move -- refuse at
50
+ # the boundary rather than pay a full budget to learn the answer cannot be
51
+ # returned -- and it does NOT touch H4 ("no maximum on any math
52
+ # parameter"), because the symbol count is derived from the expression and
53
+ # is not a parameter anyone can cap.
54
+ #
55
+ # Measured on the real `logic.run`, with `config.MAX_READ_CHARS` (50,000)
56
+ # as the ceiling the other five tools get in `serialize`:
57
+ #
58
+ # columns rows chars returned secs
59
+ # 7 128 15,207 0.01
60
+ # 8 256 33,575 0.01 <- the last one under
61
+ # 9 512 73,511 0.03
62
+ # 10 1,024 159,783 0.06
63
+ # 14 16,384 3,440,679 1.36 <- returned to the model
64
+ # 15 32,768 7,323,687 3.03 before this bound
65
+ #
66
+ # So 8 is where a character cap would have landed anyway; stating it in
67
+ # columns is what lets the refusal name the number the caller controls.
68
+ # Nothing in the suite's own battery exceeds 3 columns.
69
+ _MAX_TRUTH_TABLE_SYMBOLS = 8
70
+
29
71
 
30
72
  class LogicParams(BaseModel):
31
73
  operation: _OPERATIONS = Field(..., description="Which operation to perform")
32
74
  expression: str = Field(
33
75
  ...,
34
- description="Boolean expression using variables and operators & (and), | (or), ~ (not), >> (implies)",
76
+ description=(
77
+ "Boolean expression using variables and operators & (and), "
78
+ "| (or), ~ (not), >> (implies), Xor, Equivalent -- or a "
79
+ "comparison such as Eq(x, 5) or x > 0. It must be a "
80
+ "PROPOSITION: an arithmetic expression like 'x + y' has no "
81
+ "truth value and is refused."
82
+ ),
35
83
  )
36
84
  expression_b: Optional[str] = Field(None, description="Second expression, for the 'equivalent' operation")
37
85
  variables: Optional[list[str]] = Field(
38
- None, description="Explicit variable names/order -- mainly useful for truth_table"
86
+ None,
87
+ description=(
88
+ "Variable names. A name listed here is kept WHOLE -- that is "
89
+ "the load-bearing half: any multi-character name NOT listed is "
90
+ "read as a product of single letters, so 'rain' alone becomes "
91
+ "a*i*n*r (a 16-row table over a, i, n, r) and 'rain & snow' is "
92
+ "a parse error. It also sets the column set and order for "
93
+ "truth_table, and a listed name that does not occur in the "
94
+ "expression becomes a constant column."
95
+ ),
39
96
  )
40
97
 
41
98
 
@@ -51,12 +108,93 @@ TOOL_SCHEMA = {
51
108
  }
52
109
 
53
110
 
111
+ def _require_boolean(expr, field: str, rewritten: set) -> None:
112
+ """Refuse an expression that is not a proposition. See the module docstring.
113
+
114
+ ONE `isinstance`, not a union of four types, and that was measured
115
+ rather than assumed: `Symbol`, `BooleanTrue`, `BooleanFalse` and every
116
+ `Relational` are already `Boolean` subclasses, so the single check
117
+ accepts `a & b`, `a >> b`, `~a`, `Xor(a, b)`, `ITE(a, b, c)`,
118
+ `Eq(x, 5)`, `x > 0`, `a`, `true`, `1 = 1` and refuses `x - 1`,
119
+ `x + y`, `x**2`, `pi`, `2`, `Matrix([[1]])`. One name is also CLOSED
120
+ over whatever Boolean type a future SymPy adds, which an enumeration
121
+ of four would not be.
122
+
123
+ `a & 1` and `a | 0` never reach here -- SymPy's own `&` raises on an
124
+ Integer operand, so an arithmetic constant inside a formula was
125
+ already refused loudly (batch 143).
126
+ """
127
+ if isinstance(expr, Boolean):
128
+ return
129
+ message = (
130
+ "%s parsed as %s, which has no truth value. This tool checks "
131
+ "PROPOSITIONAL claims: combine variables with & (and), | (or), "
132
+ "~ (not), >> (implies), Xor or Equivalent, or use a comparison "
133
+ "such as Eq(x, 5) or x > 0. Substituting 0 and 1 into an "
134
+ "arithmetic expression answers a different question -- 'x + y' "
135
+ "would come back as the truth table of OR."
136
+ % (field, type(expr).__name__))
137
+ # The commonest way to land here is not an arithmetic expression at
138
+ # all: `truth_table expression="rain"` parses as `a*i*n*r`, a Mul, so
139
+ # the domain refusal fires on a perfectly good propositional name. The
140
+ # cause the caller can act on is the SPLIT, so it is appended rather
141
+ # than left for them to infer from "Mul" (batch 143).
142
+ if rewritten:
143
+ message += " " + " ".join(_split_warnings(rewritten, "variables"))
144
+ raise MathParseError(message)
145
+
146
+
147
+ def _truth_table_columns(expr, declared: list[str]) -> list:
148
+ """The column set and order for a truth table.
149
+
150
+ The declared names FIRST, in the order given, then every free symbol
151
+ the caller did not declare, sorted. The second half is not optional and
152
+ is why this is a function: handed a list that does not cover every free
153
+ symbol, SymPy's `truth_table` leaves the rest UNSUBSTITUTED and returns
154
+ the symbol itself as the output -- measured, `truth_table(a & b, [a])`
155
+ gives `([1], b)`, and `bool(Symbol('b'))` is then drawn as `true`. So
156
+ "the declared names" alone would have replaced an ignored parameter
157
+ with a wrong table.
158
+
159
+ A declared name absent from the expression becomes a constant column,
160
+ which is what a caller asking for an explicit order wants (batch 143).
161
+ """
162
+ free = {str(s): s for s in expr.free_symbols}
163
+ columns, seen = [], set()
164
+ for name in declared:
165
+ if name not in seen:
166
+ seen.add(name)
167
+ columns.append(free[name] if name in free else sympy.Symbol(name))
168
+ for name in sorted(free):
169
+ if name not in seen:
170
+ seen.add(name)
171
+ columns.append(free[name])
172
+ return columns
173
+
174
+
54
175
  def run(params: dict) -> dict:
55
- parsed = LogicParams(**params)
176
+ try:
177
+ parsed = parse_params(LogicParams, params)
178
+ except MathParseError as e:
179
+ return {"error": str(e)}
180
+
181
+ # The fourth parsing tool reports a split name too. Batch 131 added
182
+ # this to the three tools that could not reach `symbols=` at all and
183
+ # left this one out, because `variables` already reached it -- but a
184
+ # caller who does not pass it still gets an answer about a product of
185
+ # single letters, and in the FUNCTIONAL spellings that answer is
186
+ # Boolean and therefore silent: `Xor(rain, b)`, `Not(rain)`,
187
+ # `Eq(rain, 5)` and `Implies(rain, b)` all parse, all pass the domain
188
+ # check, and `truth_table` then draws 16 rows over a, i, n, r as a
189
+ # success. Only the infix spellings (`rain & b`) fail loudly, and they
190
+ # fail naming SymPy's operand types (batch 143).
191
+ rewritten: set = set()
56
192
 
57
193
  try:
58
194
  var_names = parsed.variables or []
59
- expr = safe_parse(parsed.expression, symbols=var_names)
195
+ expr = safe_parse(parsed.expression, symbols=var_names,
196
+ rewritten=rewritten)
197
+ _require_boolean(expr, "expression", rewritten)
60
198
  op = parsed.operation
61
199
 
62
200
  if op == "simplify":
@@ -65,7 +203,9 @@ def run(params: dict) -> dict:
65
203
  elif op == "equivalent":
66
204
  if not parsed.expression_b:
67
205
  return {"error": "expression_b is required for 'equivalent'"}
68
- expr_b = safe_parse(parsed.expression_b, symbols=var_names)
206
+ expr_b = safe_parse(parsed.expression_b, symbols=var_names,
207
+ rewritten=rewritten)
208
+ _require_boolean(expr_b, "expression_b", rewritten)
69
209
  result = simplify_logic(expr) == simplify_logic(expr_b)
70
210
 
71
211
  elif op == "satisfiable":
@@ -75,11 +215,27 @@ def run(params: dict) -> dict:
75
215
  result = not satisfiable(sympy.Not(expr))
76
216
 
77
217
  elif op == "truth_table":
78
- free_syms = sorted(expr.free_symbols, key=str)
79
- rows = list(truth_table(expr, free_syms))
218
+ # `variables` decides the columns when it is given, which is
219
+ # the half of its description that used to be inert: this list
220
+ # was `sorted(expr.free_symbols, key=str)` regardless, so
221
+ # neither the requested SET nor the requested ORDER had any
222
+ # effect, while the field was silently load-bearing for the
223
+ # parse (batch 143).
224
+ columns = _truth_table_columns(expr, var_names)
225
+ if len(columns) > _MAX_TRUTH_TABLE_SYMBOLS:
226
+ return {"error":
227
+ "truth_table would need %d columns, which is 2**%d "
228
+ "rows -- too large to return. The ceiling is %d "
229
+ "columns (%d rows). Ask about fewer variables, or "
230
+ "use 'satisfiable' or 'tautology', which answer "
231
+ "whether a row exists without enumerating any."
232
+ % (len(columns), len(columns),
233
+ _MAX_TRUTH_TABLE_SYMBOLS,
234
+ 2 ** _MAX_TRUTH_TABLE_SYMBOLS)}
235
+ rows = list(truth_table(expr, columns))
80
236
  result = [
81
237
  {
82
- "inputs": {str(s): bool(v) for s, v in zip(free_syms, row[0])},
238
+ "inputs": {str(s): bool(v) for s, v in zip(columns, row[0])},
83
239
  "output": bool(row[1]),
84
240
  }
85
241
  for row in rows
@@ -93,4 +249,7 @@ def run(params: dict) -> dict:
93
249
  except Exception as e:
94
250
  return {"error": f"Computation failed: {e}"}
95
251
 
96
- return {"result": result, "operation": op}
252
+ out: dict = {"result": result, "operation": op}
253
+ if rewritten:
254
+ out["warnings"] = _split_warnings(rewritten, "variables")
255
+ return out