okstra 0.235.0 → 0.237.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/lib/host-registry-client.d.mts +9 -2
- package/dist/lib/host-registry-client.mjs +29 -3
- package/dist/lib/host-registry-client.mjs.map +1 -1
- package/dist/lib/runtime-payload.mjs +1 -0
- package/dist/lib/runtime-payload.mjs.map +1 -1
- package/docs/architecture/storage-model.md +1 -0
- package/docs/architecture.md +2 -2
- package/docs/cli.md +6 -0
- package/package.json +1 -1
- package/runtime/BUILD.json +2 -2
- package/runtime/bin/lib/okstra/usage.sh +8 -8
- package/runtime/bin/okstra-qwen-exec.sh +5 -0
- package/runtime/prompts/lead/convergence.md +4 -1
- package/runtime/prompts/lead/okstra-lead-contract.md +4 -4
- package/runtime/prompts/profiles/_clarification-recommendation.md +1 -1
- package/runtime/prompts/profiles/_common-contract.md +10 -1
- package/runtime/prompts/wizard/prompts.ko.json +16 -4
- package/runtime/python/okstra_ctl/adapters/providers/antigravity/adapter.py +10 -2
- package/runtime/python/okstra_ctl/adapters/providers/claude/adapter.py +16 -1
- package/runtime/python/okstra_ctl/adapters/providers/codex/adapter.py +20 -5
- package/runtime/python/okstra_ctl/adapters/providers/kimi/adapter.py +17 -9
- package/runtime/python/okstra_ctl/adapters/providers/qwen/adapter.py +88 -0
- package/runtime/python/okstra_ctl/adapters/providers/qwen/manifest.json +1 -0
- package/runtime/python/okstra_ctl/adapters/providers/qwen/settings.json +12 -0
- package/runtime/python/okstra_ctl/adapters/providers/zai/adapter.py +8 -2
- package/runtime/python/okstra_ctl/approval_decisions.py +22 -2
- package/runtime/python/okstra_ctl/business_flow/contracts.py +11 -0
- package/runtime/python/okstra_ctl/business_flow/engine.py +90 -32
- package/runtime/python/okstra_ctl/business_flow/hooks.py +4 -2
- package/runtime/python/okstra_ctl/business_flow/source.py +39 -7
- package/runtime/python/okstra_ctl/business_flow/store.py +94 -28
- package/runtime/python/okstra_ctl/convergence.py +95 -5
- package/runtime/python/okstra_ctl/convergence_critic_verify_prompt.py +11 -10
- package/runtime/python/okstra_ctl/convergence_engine.py +30 -15
- package/runtime/python/okstra_ctl/convergence_reverify_prompt.py +9 -2
- package/runtime/python/okstra_ctl/coverage_census.py +75 -12
- package/runtime/python/okstra_ctl/entrypoints/hosts.py +1 -1
- package/runtime/python/okstra_ctl/lead_progress.py +1 -1
- package/runtime/python/okstra_ctl/manager_cli.py +42 -5
- package/runtime/python/okstra_ctl/manager_launch.py +60 -1
- package/runtime/python/okstra_ctl/manager_store.py +45 -0
- package/runtime/python/okstra_ctl/model_discovery.py +1 -1
- package/runtime/python/okstra_ctl/models.py +2 -2
- package/runtime/python/okstra_ctl/phases/discussion/profile.md +2 -2
- package/runtime/python/okstra_ctl/phases/error_analysis/profile.md +1 -1
- package/runtime/python/okstra_ctl/phases/implementation_option_selection/profile.md +1 -1
- package/runtime/python/okstra_ctl/phases/implementation_planning/authoring.py +2 -1
- package/runtime/python/okstra_ctl/phases/implementation_planning/instructions/plan-body-verification.md +2 -2
- package/runtime/python/okstra_ctl/phases/implementation_planning/profile.md +3 -3
- package/runtime/python/okstra_ctl/phases/requirements_discovery/profile.md +5 -5
- package/runtime/python/okstra_ctl/plan_item_evidence.py +25 -1
- package/runtime/python/okstra_ctl/registry/provider_registry.py +1 -1
- package/runtime/python/okstra_ctl/render_final_report.py +1 -1
- package/runtime/python/okstra_ctl/run.py +1 -1
- package/runtime/python/okstra_ctl/run_context.py +18 -3
- package/runtime/python/okstra_ctl/schema_excerpt.py +14 -9
- package/runtime/python/okstra_ctl/verdict_blocks.py +7 -2
- package/runtime/python/okstra_ctl/wizard/cli.py +14 -2
- package/runtime/python/okstra_ctl/wizard/engine.py +28 -1
- package/runtime/python/okstra_ctl/wizard/ids.py +8 -0
- package/runtime/python/okstra_ctl/wizard/picker_navigation.py +64 -2
- package/runtime/python/okstra_ctl/wizard/roles.py +194 -45
- package/runtime/python/okstra_ctl/wizard/sources.py +10 -1
- package/runtime/python/okstra_ctl/wizard/state.py +71 -21
- package/runtime/python/okstra_ctl/wizard/statefile.py +8 -3
- package/runtime/python/okstra_ctl/wizard/steps_roles.py +6 -3
- package/runtime/python/okstra_ctl/wizard/worker_settings.py +129 -25
- package/runtime/python/okstra_ctl/worker_audit_ledger.py +50 -8
- package/runtime/python/okstra_ctl/worker_prompt_body.py +26 -2
- package/runtime/python/okstra_ctl/worker_prompt_headers.py +12 -9
- package/runtime/python/okstra_token_usage/claude.py +5 -1
- package/runtime/python/okstra_token_usage/collect.py +3 -3
- package/runtime/python/okstra_token_usage/pricing.py +1 -0
- package/runtime/skills/_fragments/wizard-relay-protocol.md +1 -1
- package/runtime/skills/okstra-discussion/SKILL.md +1 -1
- package/runtime/skills/okstra-manager/SKILL.md +8 -2
- package/runtime/skills/okstra-run/SKILL.md +2 -2
- package/runtime/skills/okstra-setup/SKILL.md +34 -5
- package/runtime/skills/okstra-setup/references/project-config.md +19 -0
- package/runtime/templates/worker-prompt-preamble.md +12 -2
|
@@ -2,6 +2,7 @@ from __future__ import annotations
|
|
|
2
2
|
|
|
3
3
|
import re
|
|
4
4
|
from collections import Counter
|
|
5
|
+
from collections.abc import Sequence
|
|
5
6
|
from pathlib import Path
|
|
6
7
|
from typing import Any, cast
|
|
7
8
|
|
|
@@ -29,6 +30,8 @@ from .source import candidate_projects, capture_project, knowledge_projects
|
|
|
29
30
|
|
|
30
31
|
CONFIRMED_LEVELS = frozenset({"static", "execution-verified"})
|
|
31
32
|
PACKET_CLAIM_LIMIT = 40
|
|
33
|
+
# Words every brief title or task id carries, which name no business.
|
|
34
|
+
_GENERIC_FOCUS_WORDS = frozenset({"task", "brief", "story", "feature", "with", "from", "that", "this"})
|
|
32
35
|
|
|
33
36
|
|
|
34
37
|
def task_link_path(project_root: Path, task_key: str) -> Path:
|
|
@@ -141,7 +144,7 @@ def _claim(
|
|
|
141
144
|
|
|
142
145
|
|
|
143
146
|
def _mark_conflicts(claims: list[StoredClaim]) -> None:
|
|
144
|
-
groups: dict[str, list[StoredClaim]] = {}
|
|
147
|
+
groups: dict[tuple[str, str], list[StoredClaim]] = {}
|
|
145
148
|
for claim in claims:
|
|
146
149
|
if claim["level"] not in CONFIRMED_LEVELS or claim["status"] == "superseded":
|
|
147
150
|
continue
|
|
@@ -149,10 +152,7 @@ def _mark_conflicts(claims: list[StoredClaim]) -> None:
|
|
|
149
152
|
not row["digest"] for row in claim["sourceStates"].values()
|
|
150
153
|
):
|
|
151
154
|
continue
|
|
152
|
-
|
|
153
|
-
groups.setdefault(
|
|
154
|
-
fingerprint({"key": claim["key"], "states": state}), []
|
|
155
|
-
).append(claim)
|
|
155
|
+
groups.setdefault((claim["key"], _source_version(claim)), []).append(claim)
|
|
156
156
|
for rows in groups.values():
|
|
157
157
|
if len({row["value"] for row in rows}) > 1:
|
|
158
158
|
for row in rows:
|
|
@@ -203,20 +203,16 @@ def validate_resolutions(
|
|
|
203
203
|
"conflict resolution references unknown or repeated claims"
|
|
204
204
|
)
|
|
205
205
|
rows = _resolution_rows(data["claims"], resolution)
|
|
206
|
-
|
|
207
|
-
|
|
206
|
+
expected_ids = next(
|
|
207
|
+
(
|
|
208
|
+
{row["id"] for row in candidate}
|
|
209
|
+
for candidate in conflict_groups(data["claims"])
|
|
210
|
+
if candidate[0]["key"] == rows[0]["key"]
|
|
211
|
+
and _source_version(candidate[0]) == _source_version(rows[0])
|
|
212
|
+
),
|
|
213
|
+
set(),
|
|
208
214
|
)
|
|
209
|
-
|
|
210
|
-
row["id"]
|
|
211
|
-
for row in data["claims"]
|
|
212
|
-
if row["key"] == rows[0]["key"]
|
|
213
|
-
and row["status"] == "conflict"
|
|
214
|
-
and fingerprint(
|
|
215
|
-
{key: source["digest"] for key, source in row["sourceStates"].items()}
|
|
216
|
-
)
|
|
217
|
-
== state
|
|
218
|
-
}
|
|
219
|
-
if ids != group:
|
|
215
|
+
if ids != expected_ids:
|
|
220
216
|
raise BusinessFlowError(
|
|
221
217
|
"conflict resolution must cover the complete active conflict"
|
|
222
218
|
)
|
|
@@ -232,17 +228,24 @@ def _resolution_rows(
|
|
|
232
228
|
if not ids.issubset(by_id) or resolution["acceptedClaimId"] not in ids:
|
|
233
229
|
raise BusinessFlowError("conflict resolution references unknown claims")
|
|
234
230
|
rows = [by_id[key] for key in ids]
|
|
235
|
-
|
|
236
|
-
fingerprint(
|
|
237
|
-
{key: state["digest"] for key, state in row["sourceStates"].items()}
|
|
238
|
-
)
|
|
239
|
-
for row in rows
|
|
240
|
-
}
|
|
241
|
-
if len({row["key"] for row in rows}) != 1 or len(states) != 1:
|
|
231
|
+
if len({row["key"] for row in rows}) != 1 or len({_source_version(row) for row in rows}) != 1:
|
|
242
232
|
raise BusinessFlowError("conflict resolution mixes subjects or source versions")
|
|
243
233
|
return rows
|
|
244
234
|
|
|
245
235
|
|
|
236
|
+
def conflict_groups(claims: Sequence[StoredClaim]) -> list[list[StoredClaim]]:
|
|
237
|
+
"""Claims in conflict on one key at one source version: what one resolution covers."""
|
|
238
|
+
groups: dict[tuple[str, str], dict[str, StoredClaim]] = {}
|
|
239
|
+
for claim in claims:
|
|
240
|
+
if claim["status"] == "conflict":
|
|
241
|
+
groups.setdefault((claim["key"], _source_version(claim)), {})[claim["id"]] = claim
|
|
242
|
+
return [list(group.values()) for group in groups.values()]
|
|
243
|
+
|
|
244
|
+
|
|
245
|
+
def _source_version(claim: StoredClaim) -> str:
|
|
246
|
+
return fingerprint({key: state["digest"] for key, state in claim["sourceStates"].items()})
|
|
247
|
+
|
|
248
|
+
|
|
246
249
|
def read_shared_knowledge(project_root: Path, project_group: str) -> dict[str, Any]:
|
|
247
250
|
claims, resolutions, errors = [], [], []
|
|
248
251
|
candidates = knowledge_projects(
|
|
@@ -250,6 +253,9 @@ def read_shared_knowledge(project_root: Path, project_group: str) -> dict[str, A
|
|
|
250
253
|
)
|
|
251
254
|
for candidate in candidates:
|
|
252
255
|
root = Path(candidate["root"])
|
|
256
|
+
# A project removed after it ran stays in run history; it has no store to read.
|
|
257
|
+
if not root.is_dir():
|
|
258
|
+
continue
|
|
253
259
|
try:
|
|
254
260
|
data = read_knowledge(root)
|
|
255
261
|
claims.extend({**claim, "ownerRoot": str(root)} for claim in data["claims"])
|
|
@@ -387,17 +393,69 @@ def source_freshness(
|
|
|
387
393
|
return "current" if source_states else "unverified"
|
|
388
394
|
|
|
389
395
|
|
|
396
|
+
def _words(text: str) -> set[str]:
|
|
397
|
+
return set(re.findall(r"[a-z0-9]+", text.casefold()))
|
|
398
|
+
|
|
399
|
+
|
|
400
|
+
def focus_terms(text: str) -> tuple[str, ...]:
|
|
401
|
+
"""The words of a task's identity (task group, brief title) that can name a business."""
|
|
402
|
+
return tuple(sorted(
|
|
403
|
+
word for word in _words(text)
|
|
404
|
+
if len(word) >= 4 and not word.isdigit() and word not in _GENERIC_FOCUS_WORDS
|
|
405
|
+
))
|
|
406
|
+
|
|
407
|
+
|
|
408
|
+
def task_focus(task_key: str, brief: str) -> tuple[str, ...]:
|
|
409
|
+
"""The task group and the brief's title: what this task is about, without the brief's template."""
|
|
410
|
+
group = task_key.split(":")[1:2]
|
|
411
|
+
title = [line for line in brief.splitlines() if line.startswith("# ")][:1]
|
|
412
|
+
return focus_terms(" ".join([*group, *title]))
|
|
413
|
+
|
|
414
|
+
|
|
415
|
+
def _shares_stem(word: str, term: str) -> bool:
|
|
416
|
+
"""Same word, its plural, or one a prefix of the other when both are long."""
|
|
417
|
+
if word == term or (min(len(word), len(term)) >= 4 and _singular(word) == _singular(term)):
|
|
418
|
+
return True
|
|
419
|
+
return min(len(word), len(term)) >= 6 and (word.startswith(term) or term.startswith(word))
|
|
420
|
+
|
|
421
|
+
|
|
422
|
+
def _singular(word: str) -> str:
|
|
423
|
+
return word[:-2] if word.endswith("es") and len(word) > 4 else word.removesuffix("s")
|
|
424
|
+
|
|
425
|
+
|
|
426
|
+
def _mentions(row: dict[str, Any], terms: Sequence[str]) -> bool:
|
|
427
|
+
words = _words(" ".join([row["business"], row["key"], row["value"], *row.get("aliases", [])]))
|
|
428
|
+
return any(_shares_stem(word, term) for word in words for term in terms)
|
|
429
|
+
|
|
430
|
+
|
|
431
|
+
def split_by_focus(
|
|
432
|
+
claims: list[dict[str, Any]], focus: Sequence[str]
|
|
433
|
+
) -> tuple[list[dict[str, Any]], list[dict[str, Any]]]:
|
|
434
|
+
"""Claims that mention one of the task's focus terms, and the rest."""
|
|
435
|
+
focused = [row for row in claims if _mentions(row, focus)]
|
|
436
|
+
return focused, [row for row in claims if not _mentions(row, focus)]
|
|
437
|
+
|
|
438
|
+
|
|
390
439
|
def knowledge_packet(
|
|
391
|
-
project_root: Path, question: str, source_root: Path | None = None
|
|
440
|
+
project_root: Path, question: str, source_root: Path | None = None,
|
|
441
|
+
focus: Sequence[str] = (),
|
|
392
442
|
) -> str:
|
|
443
|
+
"""The packet's knowledge section.
|
|
444
|
+
|
|
445
|
+
`question` is the whole brief, so its words select nearly every stored claim.
|
|
446
|
+
`focus` holds the task's identity terms;
|
|
447
|
+
when given, only claims that mention one of them are shown and the rest are
|
|
448
|
+
left to lookup.
|
|
449
|
+
"""
|
|
393
450
|
try:
|
|
394
451
|
result = query_knowledge(project_root, question, source_root)
|
|
395
452
|
except (OSError, BusinessFlowError, JsonBoundaryError) as exc:
|
|
396
453
|
return f"## Shared Business Knowledge\n\nKnowledge unavailable: {exc}\n"
|
|
397
|
-
# 질문이 task-brief 전체라 토큰 매칭이 거의 모든 claim 을 고르므로, 선택 결과는
|
|
398
|
-
# 관련도 순위가 아니다. 이 프로젝트가 기록한 claim 을 앞세우고 나머지는 조회로 넘긴다.
|
|
399
454
|
own = str(project_root.resolve())
|
|
400
455
|
claims = sorted(result["claims"], key=lambda row: row["ownerRoot"] != own)
|
|
456
|
+
unrelated: list[dict[str, Any]] = []
|
|
457
|
+
if focus:
|
|
458
|
+
claims, unrelated = split_by_focus(claims, focus)
|
|
401
459
|
shown, omitted = claims[:PACKET_CLAIM_LIMIT], claims[PACKET_CLAIM_LIMIT:]
|
|
402
460
|
lines = [
|
|
403
461
|
"## Shared Business Knowledge",
|
|
@@ -427,6 +485,14 @@ def knowledge_packet(
|
|
|
427
485
|
f"- {len(omitted)} more selected claims are omitted; look them up by business: "
|
|
428
486
|
+ ", ".join(f"`{name}` ({count})" for name, count in counts.most_common())
|
|
429
487
|
)
|
|
488
|
+
if unrelated:
|
|
489
|
+
counts = Counter(row["business"] for row in unrelated)
|
|
490
|
+
lines.append(
|
|
491
|
+
f"- {len(unrelated)} stored claims do not mention this task's terms ("
|
|
492
|
+
+ ", ".join(f"`{term}`" for term in focus)
|
|
493
|
+
+ "); look them up by business if one applies: "
|
|
494
|
+
+ ", ".join(f"`{name}` ({count})" for name, count in counts.most_common())
|
|
495
|
+
)
|
|
430
496
|
lines.extend(f"- Knowledge read error: {error}" for error in result["errors"])
|
|
431
497
|
lines.extend(
|
|
432
498
|
[
|
|
@@ -1592,6 +1592,61 @@ def _recorded_results(
|
|
|
1592
1592
|
return result_args
|
|
1593
1593
|
|
|
1594
1594
|
|
|
1595
|
+
# Verdicts that accept the cited evidence. `_grounded_vote` also checks a
|
|
1596
|
+
# burden-not-met refusal, which re-inspected it; counter-evidence stands on what
|
|
1597
|
+
# the verifier found instead.
|
|
1598
|
+
_GROUNDED_VERDICTS = ("agree", "supplement")
|
|
1599
|
+
|
|
1600
|
+
|
|
1601
|
+
def _origin_evidence_paths(plan_path: Path) -> dict[str, frozenset[str]]:
|
|
1602
|
+
"""Each queued finding's cited files, from the working state beside the round plan.
|
|
1603
|
+
|
|
1604
|
+
``plan-round`` writes `convergence-round-<N>-plan-<suffix>.json` next to the
|
|
1605
|
+
`convergence-work-<suffix>.json` it planned from. A plan named otherwise has
|
|
1606
|
+
no working state to read, and its votes are collected as given.
|
|
1607
|
+
"""
|
|
1608
|
+
match = re.fullmatch(r"convergence-round-\d+-plan-(?P<suffix>.+)\.json", Path(plan_path).name)
|
|
1609
|
+
path = Path(plan_path).with_name(f"convergence-work-{match['suffix']}.json") if match else None
|
|
1610
|
+
if path is None or not path.is_file():
|
|
1611
|
+
return {}
|
|
1612
|
+
from .worker_audit_ledger import named_files
|
|
1613
|
+
|
|
1614
|
+
findings = load_owned_json_object(path).get("findings") or []
|
|
1615
|
+
return {
|
|
1616
|
+
str(row["findingId"]): frozenset(named_files(str(row.get("originEvidence") or "")))
|
|
1617
|
+
for row in findings if isinstance(row, Mapping) and row.get("findingId")
|
|
1618
|
+
}
|
|
1619
|
+
|
|
1620
|
+
|
|
1621
|
+
def _grounded_vote(vote: FindingVote, origin: frozenset[str] | None) -> dict[str, Any]:
|
|
1622
|
+
"""The vote as collected; one that opened none of the cited files counts as unverifiable.
|
|
1623
|
+
|
|
1624
|
+
Evidence that names no file (a query result, prose) leaves nothing to compare,
|
|
1625
|
+
so such a vote is collected as given.
|
|
1626
|
+
"""
|
|
1627
|
+
from .worker_audit_ledger import recorded_as_opened
|
|
1628
|
+
|
|
1629
|
+
row = {
|
|
1630
|
+
"verdict": vote.verdict,
|
|
1631
|
+
"disagreeBasis": vote.disagree_basis,
|
|
1632
|
+
"explanation": vote.explanation,
|
|
1633
|
+
}
|
|
1634
|
+
leans_on_origin = vote.verdict in _GROUNDED_VERDICTS or vote.disagree_basis == "burden-not-met"
|
|
1635
|
+
if not origin or not leans_on_origin:
|
|
1636
|
+
return row
|
|
1637
|
+
if any(recorded_as_opened(cited, frozenset(vote.opened)) for cited in origin):
|
|
1638
|
+
return row
|
|
1639
|
+
return {
|
|
1640
|
+
"verdict": "unverifiable",
|
|
1641
|
+
"disagreeBasis": None,
|
|
1642
|
+
"explanation": (
|
|
1643
|
+
"Collected as unverifiable: the Opened line names none of the cited files ("
|
|
1644
|
+
+ ", ".join(sorted(origin)) + "). " + vote.explanation
|
|
1645
|
+
),
|
|
1646
|
+
}
|
|
1647
|
+
|
|
1648
|
+
|
|
1649
|
+
|
|
1595
1650
|
def _collect_results(args: argparse.Namespace) -> tuple[str, Path]:
|
|
1596
1651
|
plan = load_owned_json_object(args.plan)
|
|
1597
1652
|
planned = _planned_finding_ids(plan)
|
|
@@ -1624,14 +1679,13 @@ def _collect_results(args: argparse.Namespace) -> tuple[str, Path]:
|
|
|
1624
1679
|
"close their attempt as ok — a discarded or failed attempt's file is not "
|
|
1625
1680
|
"a vote"
|
|
1626
1681
|
)
|
|
1682
|
+
origins = _origin_evidence_paths(args.plan)
|
|
1627
1683
|
by_finding: dict[str, dict[str, Any]] = {}
|
|
1628
1684
|
for worker, parsed in votes.items():
|
|
1629
1685
|
for finding_id, vote in parsed.items():
|
|
1630
|
-
by_finding.setdefault(finding_id, {})[worker] =
|
|
1631
|
-
|
|
1632
|
-
|
|
1633
|
-
"explanation": vote.explanation,
|
|
1634
|
-
}
|
|
1686
|
+
by_finding.setdefault(finding_id, {})[worker] = _grounded_vote(
|
|
1687
|
+
vote, origins.get(finding_id),
|
|
1688
|
+
)
|
|
1635
1689
|
write_json_atomic(args.output, {
|
|
1636
1690
|
"schemaVersion": "1.0",
|
|
1637
1691
|
"round": plan.get("round"),
|
|
@@ -2123,6 +2177,36 @@ def _read_if_present(path: Path) -> str | None:
|
|
|
2123
2177
|
return path.read_text(encoding="utf-8") if path.is_file() else None
|
|
2124
2178
|
|
|
2125
2179
|
|
|
2180
|
+
def _evidence_ledger(result_path: Path) -> frozenset[str] | None:
|
|
2181
|
+
"""The Evidence read paths of a result's audit sidecar.
|
|
2182
|
+
|
|
2183
|
+
``None`` when the result's prompt asked for no ledger — the rule
|
|
2184
|
+
validate-run applies — so a result of such a prompt is not checked. A prompt
|
|
2185
|
+
that asked for one and a sidecar without rows give an empty ledger.
|
|
2186
|
+
"""
|
|
2187
|
+
from .worker_artifact_paths import WorkerArtifactPathError, audit_sidecar_rel
|
|
2188
|
+
from .worker_audit_ledger import audit_evidence_read_paths, ledger_required
|
|
2189
|
+
|
|
2190
|
+
prefix, separator, rest = result_path.name.partition("-worker-")
|
|
2191
|
+
prompt = result_path.parent.parent / "prompts" / f"{prefix}-worker-prompt-{rest}"
|
|
2192
|
+
if not separator or not ledger_required(prompt):
|
|
2193
|
+
return None
|
|
2194
|
+
try:
|
|
2195
|
+
sidecar = result_path.with_name(audit_sidecar_rel(result_path.name))
|
|
2196
|
+
except WorkerArtifactPathError:
|
|
2197
|
+
return None
|
|
2198
|
+
return frozenset(audit_evidence_read_paths(_read_if_present(sidecar) or ""))
|
|
2199
|
+
|
|
2200
|
+
|
|
2201
|
+
def _joined_ledger(
|
|
2202
|
+
first: frozenset[str] | None, gapfill: frozenset[str] | None,
|
|
2203
|
+
) -> frozenset[str] | None:
|
|
2204
|
+
"""A gap-fill without its own ledger is not checked: it may cite files it opened anew."""
|
|
2205
|
+
if gapfill is None:
|
|
2206
|
+
return None
|
|
2207
|
+
return gapfill | (first or frozenset())
|
|
2208
|
+
|
|
2209
|
+
|
|
2126
2210
|
def _census_audit(args: argparse.Namespace) -> dict[str, Any]:
|
|
2127
2211
|
"""Audit the census verdicts. Unjudged cells are data, never an error."""
|
|
2128
2212
|
authority = validated_run_authority(args.run_manifest)
|
|
@@ -2141,10 +2225,16 @@ def _census_audit(args: argparse.Namespace) -> dict[str, Any]:
|
|
|
2141
2225
|
raise ConvergenceContractError(
|
|
2142
2226
|
f"--gapfill names {unrostered}, which no --result names"
|
|
2143
2227
|
)
|
|
2228
|
+
first_ledgers = {worker: _evidence_ledger(path) for worker, path in result_paths.items()}
|
|
2144
2229
|
audit = audit_census(
|
|
2145
2230
|
census,
|
|
2146
2231
|
{worker: _read_if_present(path) for worker, path in result_paths.items()},
|
|
2147
2232
|
{worker: _read_if_present(path) for worker, path in gapfill_paths.items()},
|
|
2233
|
+
opened=first_ledgers,
|
|
2234
|
+
gapfill_opened={
|
|
2235
|
+
worker: _joined_ledger(first_ledgers.get(worker), _evidence_ledger(path))
|
|
2236
|
+
for worker, path in gapfill_paths.items()
|
|
2237
|
+
},
|
|
2148
2238
|
)
|
|
2149
2239
|
for row in audit["workers"]:
|
|
2150
2240
|
row["resultPath"] = relative_to_project_root(
|
|
@@ -9,7 +9,7 @@
|
|
|
9
9
|
|
|
10
10
|
이 모듈이 그 지시문을 만든다. 입력은 리드가 `apply-critic-gaps` 에 줄 것과 같은
|
|
11
11
|
coverage 배치(투표 전, `gaps[]` 만 채운 상태)와 Round 0 그룹(분석 로스터)이다.
|
|
12
|
-
배정은 엔진과 같은 규칙(`critic_gap_assignees
|
|
12
|
+
배정은 엔진과 같은 규칙(`critic_gap_assignees`: all analysers up to two, otherwise round-robin)이라
|
|
13
13
|
`apply-critic-gaps` 의 커버리지 검사와 어긋나지 않는다. gap 마다 critic 결과
|
|
14
14
|
파일의 `### [<gapId>]` 절과 critic 감사 사이드카를 실어, 검증자가 리드의 전사가
|
|
15
15
|
아니라 critic 이 실제로 인용한 것을 판단하게 한다. 응답 형식은 번호 라운드와
|
|
@@ -131,10 +131,11 @@ def critic_verify_gaps(
|
|
|
131
131
|
) -> list[CriticGap]:
|
|
132
132
|
"""이 워커에게 배정된 gap 을 배치 순서대로.
|
|
133
133
|
|
|
134
|
-
|
|
135
|
-
|
|
136
|
-
|
|
137
|
-
|
|
134
|
+
The assignment is `critic_gap_assignees`: every analyser verifies every gap
|
|
135
|
+
on a roster of `FULL_GAP_VERIFICATION_ROSTER` or fewer, otherwise gap *i*
|
|
136
|
+
goes to `roster[i % len(roster)]`. The roster is the Round 0 groups'
|
|
137
|
+
analysis workers in order, the order `apply-critic-gaps` records as
|
|
138
|
+
`analyserRoster`.
|
|
138
139
|
"""
|
|
139
140
|
gaps_value = batch.get("gaps")
|
|
140
141
|
if not isinstance(gaps_value, list) or not gaps_value:
|
|
@@ -153,17 +154,17 @@ def critic_verify_gaps(
|
|
|
153
154
|
gap for gap in gaps_value
|
|
154
155
|
if isinstance(gap, Mapping) and not gap.get("duplicateOf")
|
|
155
156
|
]
|
|
156
|
-
|
|
157
|
+
verifiers_by_gap = critic_gap_assignees(roster, verifiable)
|
|
157
158
|
assigned: list[CriticGap] = []
|
|
158
159
|
seen: set[str] = set()
|
|
159
|
-
for gap,
|
|
160
|
+
for gap, verifiers in zip(verifiable, verifiers_by_gap):
|
|
160
161
|
gap_id = _nonempty_string(gap.get("gapId"))
|
|
161
162
|
if not gap_id:
|
|
162
163
|
raise CriticVerifyPromptError("coverage batch gap has no gapId")
|
|
163
164
|
if gap_id in seen:
|
|
164
165
|
raise CriticVerifyPromptError(f"duplicate critic gapId: {gap_id}")
|
|
165
166
|
seen.add(gap_id)
|
|
166
|
-
if
|
|
167
|
+
if worker_id not in verifiers:
|
|
167
168
|
continue
|
|
168
169
|
ticket_ids = gap.get("ticketIds")
|
|
169
170
|
assigned.append(CriticGap(
|
|
@@ -177,8 +178,8 @@ def critic_verify_gaps(
|
|
|
177
178
|
))
|
|
178
179
|
if not assigned:
|
|
179
180
|
raise CriticVerifyPromptError(
|
|
180
|
-
f"
|
|
181
|
-
|
|
181
|
+
f"no gap is assigned to `{worker_id}`; verifiers in batch order: "
|
|
182
|
+
+ (", ".join("+".join(verifiers) for verifiers in verifiers_by_gap) or "none")
|
|
182
183
|
)
|
|
183
184
|
return assigned
|
|
184
185
|
|
|
@@ -1136,26 +1136,34 @@ def _parse_critic_dispatches(
|
|
|
1136
1136
|
return dispatches, completed
|
|
1137
1137
|
|
|
1138
1138
|
|
|
1139
|
+
# With this many analysers or fewer, every gap goes to all of them: one
|
|
1140
|
+
# verifier per gap lets a single shallow pass decide what merges into the report.
|
|
1141
|
+
FULL_GAP_VERIFICATION_ROSTER = 2
|
|
1142
|
+
|
|
1143
|
+
|
|
1139
1144
|
def critic_gap_assignees(
|
|
1140
1145
|
roster_order: list[str],
|
|
1141
1146
|
gaps: list[Mapping[str, Any]],
|
|
1142
|
-
) -> list[str]:
|
|
1143
|
-
"""
|
|
1147
|
+
) -> list[tuple[str, ...]]:
|
|
1148
|
+
"""The verifiers of each gap, in batch order.
|
|
1144
1149
|
|
|
1145
|
-
|
|
1146
|
-
|
|
1147
|
-
|
|
1148
|
-
|
|
1150
|
+
A roster of `FULL_GAP_VERIFICATION_ROSTER` or fewer verifies every gap in
|
|
1151
|
+
full. A larger roster is walked round-robin, one verifier per gap: a
|
|
1152
|
+
rejection costs the same as an acceptance in a stretch recorded as
|
|
1153
|
+
`rounds: []`, so each analyser reads only its share.
|
|
1149
1154
|
|
|
1150
|
-
|
|
1155
|
+
Declared duplicates (`duplicateOf`) are not assigned; they are not verified.
|
|
1151
1156
|
"""
|
|
1152
1157
|
verifiable = [gap for gap in gaps if not gap.get("duplicateOf")]
|
|
1153
1158
|
if not roster_order:
|
|
1154
1159
|
return []
|
|
1155
|
-
|
|
1156
|
-
|
|
1157
|
-
|
|
1158
|
-
|
|
1160
|
+
if len(roster_order) <= FULL_GAP_VERIFICATION_ROSTER:
|
|
1161
|
+
return [tuple(roster_order) for _ in verifiable]
|
|
1162
|
+
return [(worker,) for worker in _round_robin_assignees(roster_order, len(verifiable))]
|
|
1163
|
+
|
|
1164
|
+
|
|
1165
|
+
def _round_robin_assignees(roster_order: list[str], count: int) -> list[str]:
|
|
1166
|
+
return [roster_order[index % len(roster_order)] for index in range(count)]
|
|
1159
1167
|
|
|
1160
1168
|
|
|
1161
1169
|
def _critic_gap_coverage_errors(
|
|
@@ -1174,13 +1182,20 @@ def _critic_gap_coverage_errors(
|
|
|
1174
1182
|
계약 위반이 아니라 더 본 것이고, 그 사이 방치된 gap 은 표가 없어
|
|
1175
1183
|
`unverifiedGaps` 로 스스로 드러난다.
|
|
1176
1184
|
"""
|
|
1177
|
-
assignees =
|
|
1178
|
-
|
|
1185
|
+
assignees = {
|
|
1186
|
+
worker for verifiers in critic_gap_assignees(roster_order, gaps) for worker in verifiers
|
|
1187
|
+
}
|
|
1188
|
+
# A run finished under the earlier one-verifier rule dispatched the round-robin set.
|
|
1189
|
+
earlier = set(_round_robin_assignees(
|
|
1190
|
+
roster_order, len([gap for gap in gaps if not gap.get("duplicateOf")]),
|
|
1191
|
+
)) if roster_order else set()
|
|
1192
|
+
if dispatched in (set(roster_order), assignees, earlier):
|
|
1179
1193
|
return []
|
|
1180
1194
|
missing = sorted(assignees - dispatched)
|
|
1181
1195
|
return [
|
|
1182
|
-
"critic dispatches must cover every gap
|
|
1183
|
-
"
|
|
1196
|
+
"critic dispatches must cover every gap's assigned verifiers (all analysers "
|
|
1197
|
+
f"up to {FULL_GAP_VERIFICATION_ROSTER}, otherwise round-robin) or account "
|
|
1198
|
+
"for every non-critic analyser exactly "
|
|
1184
1199
|
f"once; missing: {', '.join(missing) or 'none'}"
|
|
1185
1200
|
]
|
|
1186
1201
|
|
|
@@ -94,7 +94,14 @@ ADVERSARIAL_RESPONSE = """### <finding-id>
|
|
|
94
94
|
**Basis** (only if REFUTED): counter-evidence | burden-not-met
|
|
95
95
|
**Explanation** (required for every verdict, SURVIVES included): <2-3 sentences; for
|
|
96
96
|
SURVIVES say what you attacked and why the attack failed; for counter-evidence include
|
|
97
|
-
the file:line you found>
|
|
97
|
+
the file:line you found>
|
|
98
|
+
**Opened**: <every file you opened for this finding, as path:line, comma-separated>"""
|
|
99
|
+
|
|
100
|
+
# `okstra convergence collect-results` enforces this for round votes
|
|
101
|
+
# (`convergence._grounded_vote`); critic and acceptance votes are not collected there.
|
|
102
|
+
OPENED_RULE = """A SURVIVES, SURVIVES-WITH-CAVEAT or burden-not-met verdict whose `**Opened**` line
|
|
103
|
+
names none of the files the finding cites is collected as UNVERIFIABLE: a verdict
|
|
104
|
+
on evidence you did not open is not a vote."""
|
|
98
105
|
|
|
99
106
|
def _nonempty_string(value: Any) -> str:
|
|
100
107
|
return value if isinstance(value, str) and value.strip() else ""
|
|
@@ -222,7 +229,7 @@ def reverify_prompt_body(
|
|
|
222
229
|
"collector also reads `**Verdict:** …` and `- Verdict: …` as the same field.\n\n"
|
|
223
230
|
)
|
|
224
231
|
rows.append(ADVERSARIAL_RESPONSE.replace("<finding-id>", findings[0].finding_id))
|
|
225
|
-
rows.append("\n")
|
|
232
|
+
rows.append("\n\n" + OPENED_RULE + "\n")
|
|
226
233
|
if len(findings) > 1:
|
|
227
234
|
rows.append(f"\n### {findings[1].finding_id}\n**Verdict**: ...\n")
|
|
228
235
|
return "".join(rows)
|