@yahyahammad/venastine 0.3.8 → 0.3.10
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.env.example +19 -19
- package/CONFIG_ARCHITECTURE.md +2139 -2002
- package/LICENSE +201 -201
- package/README.md +1043 -1034
- package/agents/builtin/compactor.md +75 -75
- package/agents/builtin/explore.md +77 -77
- package/agents/builtin/pipeline-reviewer.md +100 -100
- package/agents/builtin/plan.md +106 -106
- package/agents/builtin/review.md +74 -74
- package/agents/manager.py +261 -261
- package/agents/subagent_tool.py +472 -472
- package/agents/tui_commands.py +238 -146
- package/bin/venastine.mjs +584 -528
- package/bin/venastine.py +73 -71
- package/cli_commands.py +848 -835
- package/config.yaml +64 -0
- package/config_edit.py +1301 -1301
- package/config_schema.py +987 -966
- package/config_update.py +749 -749
- package/core/agent_activity.py +256 -256
- package/core/ansi.py +452 -452
- package/core/approval.py +137 -137
- package/core/call_view.py +364 -364
- package/core/cancel.py +121 -121
- package/core/chatgpt_auth.py +747 -747
- package/core/claims_view.py +107 -107
- package/core/client.py +2756 -2715
- package/core/commands.py +464 -454
- package/core/compaction.py +1735 -1735
- package/core/config_loader.py +1982 -1982
- package/core/connect.py +686 -686
- package/core/events.py +123 -123
- package/core/file_mentions.py +280 -280
- package/core/interaction.py +411 -411
- package/core/loop.py +2566 -2566
- package/core/memory.py +482 -482
- package/core/pipeline_models.py +190 -190
- package/core/provider_errors.py +274 -234
- package/core/reasoning/authorization.py +130 -130
- package/core/reasoning/base.py +219 -219
- package/core/reasoning/confidence_scoring.py +484 -484
- package/core/reasoning/events.py +149 -149
- package/core/reasoning/orchestrator.py +1561 -1561
- package/core/reasoning/output_writer.py +345 -345
- package/core/reasoning/payload_validation.py +524 -524
- package/core/reasoning/pipeline_storage.py +427 -427
- package/core/reasoning/scholar.py +467 -458
- package/core/reasoning/source_corpus.py +433 -433
- package/core/reasoning/source_scoring.py +855 -855
- package/core/replay.py +315 -315
- package/core/shell_sessions.py +1553 -1553
- package/credentials.py +307 -307
- package/database.py +333 -333
- package/docs/PRIVACY.md +78 -78
- package/docs/SECURITY.md +114 -108
- package/docs/THIRD_PARTY_NOTICES.md +148 -148
- package/env_secrets.py +118 -118
- package/json_store.py +343 -343
- package/logging_setup.py +295 -295
- package/main.py +3433 -3433
- package/mcp_client/client.py +780 -733
- package/mcp_client/config.py +484 -484
- package/mcp_client/registration.py +343 -343
- package/memories/manager.py +147 -147
- package/memories/tui_commands.py +319 -239
- package/package.json +303 -303
- package/project_init/config_files.py +282 -282
- package/project_init/generator.py +728 -728
- package/project_init/manifest.py +418 -418
- package/project_init/tui_commands.py +147 -147
- package/prompts/critic_pass.md +24 -24
- package/prompts/revalidate.md +25 -25
- package/prompts/revise.md +17 -17
- package/prompts/source_grounding.md +66 -66
- package/prompts/system_prompts.py +316 -316
- package/providers.json.example +94 -94
- package/pyproject.toml +149 -149
- package/requirements.txt +314 -295
- package/safety/policy_enforcement.py +1082 -1082
- package/security/askpass.py +419 -419
- package/security/capability.py +207 -207
- package/security/confine.py +484 -484
- package/security/hosted.py +182 -182
- package/security/permissions.py +237 -237
- package/security/protected_paths.py +501 -501
- package/security/sandbox.py +4108 -4108
- package/skills/builtin/math/formal-specification.md +67 -67
- package/skills/builtin/math/numerical-methods.md +66 -66
- package/skills/builtin/research/experiment-design.md +62 -62
- package/skills/builtin/research/reproducibility.md +55 -55
- package/skills/builtin/research/source-evaluation.md +70 -70
- package/skills/builtin/research/statistical-inference.md +63 -63
- package/skills/builtin/research/technical-writing.md +59 -59
- package/skills/builtin/software/code-review.md +61 -61
- package/skills/builtin/software/debugging.md +61 -61
- package/skills/builtin/software/test-design.md +67 -67
- package/skills/tui_commands.py +110 -110
- package/storage.py +1338 -1338
- package/tools/base.py +465 -465
- package/tools/builtin/_math_common.py +667 -667
- package/tools/builtin/_net_common.py +1145 -267
- package/tools/builtin/_page_text.py +490 -0
- package/tools/builtin/_pdf_text.py +113 -0
- package/tools/builtin/arxiv.py +685 -681
- package/tools/builtin/fetch_url.py +639 -371
- package/tools/builtin/file_ops.py +720 -720
- package/tools/builtin/geometry.py +184 -184
- package/tools/builtin/load_skill.py +76 -76
- package/tools/builtin/pin.py +227 -227
- package/tools/builtin/probability_stats.py +224 -224
- package/tools/builtin/project_docs.py +464 -464
- package/tools/builtin/remember.py +175 -175
- package/tools/builtin/shell.py +980 -980
- package/tools/builtin/todo.py +149 -149
- package/tools/builtin/web_search.py +402 -328
- package/tools/context.py +159 -159
- package/tools/isolation.py +387 -387
- package/tools/registry.py +1252 -1242
- package/tui/app.py +8372 -8307
- package/tui/app.tcss +754 -739
- package/tui/commands.py +26 -26
- package/tui/diffs.py +286 -286
- package/tui/history.py +128 -128
- package/tui/hosted_driver.py +102 -102
- package/tui/markdown.py +1329 -1329
- package/tui/meters.py +364 -364
- package/tui/ravens.py +146 -146
- package/tui/screens.py +1255 -1197
- package/tui/themes.py +599 -599
- package/tui/widgets.py +3861 -3796
package/core/claims_view.py
CHANGED
|
@@ -1,107 +1,107 @@
|
|
|
1
|
-
"""
|
|
2
|
-
core/claims_view.py
|
|
3
|
-
|
|
4
|
-
What a research run's claims view shows (ROADMAP_v2 §26), for either shell.
|
|
5
|
-
|
|
6
|
-
Moved out of `tui/screens.py` by ROADMAP_v3 §89 (#8), when the CLI got a
|
|
7
|
-
`/claims` of its own: which claims come first, which fields are shown and
|
|
8
|
-
how a malformed stored field reads are decisions about the RUN, not about
|
|
9
|
-
a modal, and the CLI printing its own copy of them is the disagreement
|
|
10
|
-
`core/replay.py` exists to prevent. The TUI's ClaimsScreen and the CLI's
|
|
11
|
-
`/claims` both draw what this returns.
|
|
12
|
-
|
|
13
|
-
Plain dicts, not Claim objects: the sources are `vars(c)` for a finished
|
|
14
|
-
in-session run, `pipeline_storage.load_pipeline_run()` for a stored one,
|
|
15
|
-
and the partial picture assembled from PipelineEvents mid-run.
|
|
16
|
-
"""
|
|
17
|
-
|
|
18
|
-
from __future__ import annotations
|
|
19
|
-
|
|
20
|
-
# Tier order for the claims view: worst first. A reader opening this is
|
|
21
|
-
# looking for what NOT to trust, and burying the unverified claims under
|
|
22
|
-
# the confident ones makes them the hardest thing to find.
|
|
23
|
-
TIER_ORDER = ("UNVERIFIED", "UNVERIFIED_COVERAGE", "LOW", "MEDIUM", "HIGH")
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
def as_display_list(value) -> list:
|
|
27
|
-
"""The members of a model-supplied list field, as text.
|
|
28
|
-
|
|
29
|
-
EVERY FIELD THIS GUARDS IS MODEL DATA READ BACK FROM A STORED RUN, AND
|
|
30
|
-
A STORED RUN MAY HAVE FAILED (batch 149). The orchestrator's except
|
|
31
|
-
block persists a snapshot before it re-raises, so a run that died in
|
|
32
|
-
Pass 5 on `assumption_flags: 123` leaves that value in the record --
|
|
33
|
-
and `/claims <run id>` then rendered it. `'; '.join(map(str, 123))`
|
|
34
|
-
raises TypeError inside `compose()`, which is NOT covered by the
|
|
35
|
-
containment batch 145 put at the `/` dispatch site: `push_screen`
|
|
36
|
-
returns an `await_mount`, so compose runs in the message pump rather
|
|
37
|
-
than in the handler's call stack, and the exception reaches Textual's
|
|
38
|
-
`_handle_exception` and takes the whole app down. Measured against the
|
|
39
|
-
real ClaimsScreen in a minimal host app.
|
|
40
|
-
|
|
41
|
-
The payload boundary now rejects each of these shapes at the pass that
|
|
42
|
-
sent them (§66 A25), which closes the producer -- but not for a record
|
|
43
|
-
written before it did, and this view's whole job is reading those.
|
|
44
|
-
|
|
45
|
-
A value that is not a list is shown as the one value it is, rather
|
|
46
|
-
than dropped: the reader asked what the run recorded, and "123" is the
|
|
47
|
-
answer.
|
|
48
|
-
"""
|
|
49
|
-
if isinstance(value, (list, tuple)):
|
|
50
|
-
return [str(member) for member in value]
|
|
51
|
-
return [str(value)]
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
def claims_header(title: str, count: int, partial: bool) -> str:
|
|
55
|
-
"""The view's first line: the run, how many claims, and whether the
|
|
56
|
-
run is still going.
|
|
57
|
-
|
|
58
|
-
The partial note is said rather than left to inference: mid-run, only
|
|
59
|
-
id, text and tier have been broadcast as events, and thin rows with no
|
|
60
|
-
note would read as a pipeline that recorded nothing.
|
|
61
|
-
"""
|
|
62
|
-
header = f"{title} ({count})"
|
|
63
|
-
if partial:
|
|
64
|
-
header += " — run in progress, metadata incomplete"
|
|
65
|
-
return header
|
|
66
|
-
|
|
67
|
-
|
|
68
|
-
def render_claims(claims: list) -> str:
|
|
69
|
-
if not claims:
|
|
70
|
-
return "No claims recorded for this run."
|
|
71
|
-
|
|
72
|
-
def _rank(claim):
|
|
73
|
-
tier = claim.get("confidence_tier")
|
|
74
|
-
return TIER_ORDER.index(tier) if tier in TIER_ORDER else len(TIER_ORDER)
|
|
75
|
-
|
|
76
|
-
out = []
|
|
77
|
-
for claim in sorted(claims, key=_rank):
|
|
78
|
-
tier = claim.get("confidence_tier") or "—"
|
|
79
|
-
head = f"{tier:<20} {claim.get('id', '?')} {claim.get('type') or ''}"
|
|
80
|
-
if claim.get("retry_count"):
|
|
81
|
-
head += f" revised x{claim['retry_count']}"
|
|
82
|
-
out.append(head.rstrip())
|
|
83
|
-
# final_text is what the claim carries into the report; text is
|
|
84
|
-
# what Pass 2 extracted. Showing final_text when they differ means
|
|
85
|
-
# the view matches the report rather than the draft behind it.
|
|
86
|
-
out.append(f" {claim.get('final_text') or claim.get('text') or ''}")
|
|
87
|
-
if claim.get("entities"):
|
|
88
|
-
out.append(" entities: "
|
|
89
|
-
+ ", ".join(as_display_list(claim["entities"])))
|
|
90
|
-
if claim.get("grounding_status") or claim.get("grounding_sources"):
|
|
91
|
-
sources = len(as_display_list(claim.get("grounding_sources")
|
|
92
|
-
or ()))
|
|
93
|
-
out.append(f" grounding: {claim.get('grounding_status') or 'none'}"
|
|
94
|
-
f" ({sources} source{'s' if sources != 1 else ''})")
|
|
95
|
-
for field, label in (("fallacies", "fallacies"),
|
|
96
|
-
("contradictions", "contradictions"),
|
|
97
|
-
("assumption_flags", "assumptions")):
|
|
98
|
-
if claim.get(field):
|
|
99
|
-
out.append(f" {label}: "
|
|
100
|
-
+ "; ".join(as_display_list(claim[field])))
|
|
101
|
-
if claim.get("score_breakdown"):
|
|
102
|
-
out.append(" score: " + ", ".join(
|
|
103
|
-
f"{k}={v}" for k, v in claim["score_breakdown"].items()))
|
|
104
|
-
if claim.get("annotation"):
|
|
105
|
-
out.append(f" {claim['annotation']}")
|
|
106
|
-
out.append("")
|
|
107
|
-
return "\n".join(out).rstrip()
|
|
1
|
+
"""
|
|
2
|
+
core/claims_view.py
|
|
3
|
+
|
|
4
|
+
What a research run's claims view shows (ROADMAP_v2 §26), for either shell.
|
|
5
|
+
|
|
6
|
+
Moved out of `tui/screens.py` by ROADMAP_v3 §89 (#8), when the CLI got a
|
|
7
|
+
`/claims` of its own: which claims come first, which fields are shown and
|
|
8
|
+
how a malformed stored field reads are decisions about the RUN, not about
|
|
9
|
+
a modal, and the CLI printing its own copy of them is the disagreement
|
|
10
|
+
`core/replay.py` exists to prevent. The TUI's ClaimsScreen and the CLI's
|
|
11
|
+
`/claims` both draw what this returns.
|
|
12
|
+
|
|
13
|
+
Plain dicts, not Claim objects: the sources are `vars(c)` for a finished
|
|
14
|
+
in-session run, `pipeline_storage.load_pipeline_run()` for a stored one,
|
|
15
|
+
and the partial picture assembled from PipelineEvents mid-run.
|
|
16
|
+
"""
|
|
17
|
+
|
|
18
|
+
from __future__ import annotations
|
|
19
|
+
|
|
20
|
+
# Tier order for the claims view: worst first. A reader opening this is
|
|
21
|
+
# looking for what NOT to trust, and burying the unverified claims under
|
|
22
|
+
# the confident ones makes them the hardest thing to find.
|
|
23
|
+
TIER_ORDER = ("UNVERIFIED", "UNVERIFIED_COVERAGE", "LOW", "MEDIUM", "HIGH")
|
|
24
|
+
|
|
25
|
+
|
|
26
|
+
def as_display_list(value) -> list:
|
|
27
|
+
"""The members of a model-supplied list field, as text.
|
|
28
|
+
|
|
29
|
+
EVERY FIELD THIS GUARDS IS MODEL DATA READ BACK FROM A STORED RUN, AND
|
|
30
|
+
A STORED RUN MAY HAVE FAILED (batch 149). The orchestrator's except
|
|
31
|
+
block persists a snapshot before it re-raises, so a run that died in
|
|
32
|
+
Pass 5 on `assumption_flags: 123` leaves that value in the record --
|
|
33
|
+
and `/claims <run id>` then rendered it. `'; '.join(map(str, 123))`
|
|
34
|
+
raises TypeError inside `compose()`, which is NOT covered by the
|
|
35
|
+
containment batch 145 put at the `/` dispatch site: `push_screen`
|
|
36
|
+
returns an `await_mount`, so compose runs in the message pump rather
|
|
37
|
+
than in the handler's call stack, and the exception reaches Textual's
|
|
38
|
+
`_handle_exception` and takes the whole app down. Measured against the
|
|
39
|
+
real ClaimsScreen in a minimal host app.
|
|
40
|
+
|
|
41
|
+
The payload boundary now rejects each of these shapes at the pass that
|
|
42
|
+
sent them (§66 A25), which closes the producer -- but not for a record
|
|
43
|
+
written before it did, and this view's whole job is reading those.
|
|
44
|
+
|
|
45
|
+
A value that is not a list is shown as the one value it is, rather
|
|
46
|
+
than dropped: the reader asked what the run recorded, and "123" is the
|
|
47
|
+
answer.
|
|
48
|
+
"""
|
|
49
|
+
if isinstance(value, (list, tuple)):
|
|
50
|
+
return [str(member) for member in value]
|
|
51
|
+
return [str(value)]
|
|
52
|
+
|
|
53
|
+
|
|
54
|
+
def claims_header(title: str, count: int, partial: bool) -> str:
|
|
55
|
+
"""The view's first line: the run, how many claims, and whether the
|
|
56
|
+
run is still going.
|
|
57
|
+
|
|
58
|
+
The partial note is said rather than left to inference: mid-run, only
|
|
59
|
+
id, text and tier have been broadcast as events, and thin rows with no
|
|
60
|
+
note would read as a pipeline that recorded nothing.
|
|
61
|
+
"""
|
|
62
|
+
header = f"{title} ({count})"
|
|
63
|
+
if partial:
|
|
64
|
+
header += " — run in progress, metadata incomplete"
|
|
65
|
+
return header
|
|
66
|
+
|
|
67
|
+
|
|
68
|
+
def render_claims(claims: list) -> str:
|
|
69
|
+
if not claims:
|
|
70
|
+
return "No claims recorded for this run."
|
|
71
|
+
|
|
72
|
+
def _rank(claim):
|
|
73
|
+
tier = claim.get("confidence_tier")
|
|
74
|
+
return TIER_ORDER.index(tier) if tier in TIER_ORDER else len(TIER_ORDER)
|
|
75
|
+
|
|
76
|
+
out = []
|
|
77
|
+
for claim in sorted(claims, key=_rank):
|
|
78
|
+
tier = claim.get("confidence_tier") or "—"
|
|
79
|
+
head = f"{tier:<20} {claim.get('id', '?')} {claim.get('type') or ''}"
|
|
80
|
+
if claim.get("retry_count"):
|
|
81
|
+
head += f" revised x{claim['retry_count']}"
|
|
82
|
+
out.append(head.rstrip())
|
|
83
|
+
# final_text is what the claim carries into the report; text is
|
|
84
|
+
# what Pass 2 extracted. Showing final_text when they differ means
|
|
85
|
+
# the view matches the report rather than the draft behind it.
|
|
86
|
+
out.append(f" {claim.get('final_text') or claim.get('text') or ''}")
|
|
87
|
+
if claim.get("entities"):
|
|
88
|
+
out.append(" entities: "
|
|
89
|
+
+ ", ".join(as_display_list(claim["entities"])))
|
|
90
|
+
if claim.get("grounding_status") or claim.get("grounding_sources"):
|
|
91
|
+
sources = len(as_display_list(claim.get("grounding_sources")
|
|
92
|
+
or ()))
|
|
93
|
+
out.append(f" grounding: {claim.get('grounding_status') or 'none'}"
|
|
94
|
+
f" ({sources} source{'s' if sources != 1 else ''})")
|
|
95
|
+
for field, label in (("fallacies", "fallacies"),
|
|
96
|
+
("contradictions", "contradictions"),
|
|
97
|
+
("assumption_flags", "assumptions")):
|
|
98
|
+
if claim.get(field):
|
|
99
|
+
out.append(f" {label}: "
|
|
100
|
+
+ "; ".join(as_display_list(claim[field])))
|
|
101
|
+
if claim.get("score_breakdown"):
|
|
102
|
+
out.append(" score: " + ", ".join(
|
|
103
|
+
f"{k}={v}" for k, v in claim["score_breakdown"].items()))
|
|
104
|
+
if claim.get("annotation"):
|
|
105
|
+
out.append(f" {claim['annotation']}")
|
|
106
|
+
out.append("")
|
|
107
|
+
return "\n".join(out).rstrip()
|