zy839971925-zyy/Agent--deep-research-Workflow-Skills / skills/reasoning-workflow
reasoning-workflow
VerifiedDefault silent workflow for everyday questions, research, writing, design and engineering. Match effort to the request, resolve costly ambiguity, find specialist resources and verify the requested outcome. Explicitly selectable with “调用我的 workflow skills”.
tests/audit_repro/adversarial_probes.py
View on GitHubfrom pathlib import Path
import sys,json,copy,importlib.util,tempfile,hashlib
HERE=Path(__file__).resolve().parent
import argparse
ap=argparse.ArgumentParser();ap.add_argument('--skill',type=Path,default=HERE/'new/Reasoning Workflow');ap.add_argument('--output',type=Path,default=HERE/'adversarial_results.json');args=ap.parse_args()
ROOT=args.skill.resolve();sys.path.insert(0,str(ROOT/'scripts'))
import runtime_core as rt
from routing_core import route_task,recheck_profile,validate_profile
from route_resources import select
from learning_core import admit_record,retrieve
from validate_workflow import fingerprint,validate_bridge
from validate_state import validate as state_valid
spec=importlib.util.spec_from_file_location('suite',ROOT/'tests/run_regression.py');suite=importlib.util.module_from_spec(spec);spec.loader.exec_module(suite)
FIX=ROOT/'tests/fixtures';load=lambda n:json.loads((FIX/n).read_text());err=lambda fs:[f.code for f in fs if f.severity=='ERROR']
results=[]
def log(id,expected,actual,**kw):
results.append(dict(id=id,expected=expected,actual=actual,**kw))
# B1: forbid parallel, no display-group label
x=load('legal_parallel_independent_input.json')
for n in x['plan']['nodes']:
n['parallel_policy']='forbidden';n['node_contract_hash']=rt.compute_node_contract_hash(n)
for a,n in zip(x['schedule']['assignments'],x['plan']['nodes']):a['node_contract_hash']=n['node_contract_hash']
log('B1_parallel_forbidden_without_group','validation refuses unsequenced parallel forbidden nodes',dict(plan_errors=err(rt.validate_plan(x['plan'],ROOT)),schedule_errors=err(rt.validate_schedule(x['plan'],x['schedule'],ROOT))))
for a in x['schedule']['assignments']:a['concurrency_group']='G'
log('C1_parallel_group_control','refuses same schedule with group',err(rt.validate_schedule(x['plan'],x['schedule'],ROOT)))
# B2: Ultra resources can remain permanently out of batch
p=load('standard_profile.json');p.update(requested_depth='ultra',task_modes=['research'],active_gap_tags=[]);p=recheck_profile(p,[]);r=route_task(p);index=json.loads((ROOT/'routing-index.json').read_text());selected=select(index,r,p,[])
log('B2_explicit_ultra_not_selected','Ultra method reachable on explicit Ultra even without extra tags',dict(profile_valid=validate_profile(p,ROOT),route=r,selected=[s['reference_id'] for s in selected],repeat_selected=[s['reference_id'] for s in select(index,r,p,[])]))
p['active_gap_tags']=['ultra'];selected=select(index,route_task(p),p,[])
log('C2_ultra_tag_control','ultra tag selects overlay',[s['reference_id'] for s in selected])
p['active_gap_tags']=['ultra','delegation','recovery','orthogonal-verification'];selected=select(index,route_task(p),p,[])
log('B2b_multi_gap_coverage','recovery control either selected or explicitly queued',[s['reference_id'] for s in selected])
# B3: tier3 required evidence not resolved
q=load('learning_intrinsic_flag_false.json');q.update(tier=3,status='promoted',source_eval_refs=['eval-good'],regression_refs=['missing-regression'],held_out_eval_refs=['missing-heldout'],baseline_comparison_ref='missing-baseline',dual_review_ref='missing-review',rollback_ref='missing-rollback');q['provenance']={'feedback_types':['deterministic_verifier'],'intrinsic_reflection_only':False}
seen=[]
def resolver(ref):seen.append(ref);return ref=='eval-good'
fs=admit_record(q,ROOT,resolver)
log('B3_tier3_unresolved_promotion_gates','must resolve required promotion evidence',dict(findings=fs,resolver_calls=seen.copy(),maintenance_retrieved=[x['learning_id'] for x in retrieve([q],['deep_research'],[],maintenance=True,validation_resolver=resolver)]))
import subprocess
with tempfile.TemporaryDirectory() as td:
d=Path(td);(d/'receipt.txt').write_text('actual one external evaluation receipt')
(d/'index.json').write_text(json.dumps({'eval-good':{'verdict':'pass','kind':'external_evaluation','path':'receipt.txt','sha256':hashlib.sha256((d/'receipt.txt').read_bytes()).hexdigest()}}))
(d/'record.json').write_text(json.dumps(q))
proc=subprocess.run([sys.executable,str(ROOT/'scripts/validate_learning.py'),str(d/'record.json'),'--validation-evidence',str(d/'index.json')],capture_output=True,text=True)
log('B3_cli_promotion_gate','missing tier3 gate references rejected even if source eval is real',dict(exit_code=proc.returncode,stdout=json.loads(proc.stdout)))
log('C3_ordinary_tier3_control','tier3 excluded from ordinary retrieval',[x['learning_id'] for x in retrieve([q],['deep_research'],[],validation_resolver=resolver)])
# B4: strict review fingerprint omits relationship endpoint payload
case=suite.Regression();case.setUp()
try:
p,r,s,m,b=case.bridge()
s['observations']=[{'id':'o1','record_type':'observation','declared_status':'current','text':'critical original observation'}]
s['relationships']=[{'id':'rel1','record_type':'relationship','declared_status':'current','relation_type':'causes','epistemic_status':'inferred','from_ref':'o1','to_ref':'e1'}]
s['answers'][0]['depends_on'].append('rel1')
fs,reg,closure=state_valid(s,ROOT)
log('B4_fixture_schema_check','valid fixture',dict(errors=err(fs),closure=closure))
for entry in b['reviews']:
entry['target_fingerprint']=fingerprint(s,entry['target_refs'])
for v in s['verification']:
if v['id']==entry['record']['verification_id']:v['metadata']['target_fingerprint']=entry['target_fingerprint']
before=validate_bridge(p,r,s,m,b,case.d)
oldfp=fingerprint(s,['a1']);s['observations'][0]['text']='opposite critical observation, not reviewed'
after=validate_bridge(p,r,s,m,b,case.d)
log('B4_singular_endpoint_fingerprint','source endpoint change invalidates old target review',dict(before=before,after=after,fingerprint_unchanged=oldfp==fingerprint(s,['a1'])))
oldfp=fingerprint(s,['a1']);s['evidence'][0]['text']='direct evidence changed'
log('C4_direct_evidence_control','direct evidence mutation invalidates review',dict(fingerprint_changed=oldfp!=fingerprint(s,['a1']),bridge_passed=validate_bridge(p,r,s,m,b,case.d)['passed']))
# B5: Question lane required acceptance incomplete
p,r,s,m,b=case.bridge()
s['requirements'][0]['closure_effect']='blocking';s['acceptance_criteria'][0]['closure_effect']='blocking'
s['requirements'][0]['declared_status']='blocked';m['requirements'][0]['declared_status']='blocked'
s['acceptance_criteria'][0]['declared_status']='blocked';m['acceptance_criteria'][0]['declared_status']='blocked'
fs,result=__import__('validate_delivery').validate(s,m,case.d)
out=validate_bridge(p,r,s,m,b,case.d)
log('B5_question_requirements_blocked','material blocking explicit requirements prevent delivery',dict(delivery_errors=err(fs),delivery_ready=result['computed_delivery_ready'],bridge=out))
s['lane']='mixed'
fs,result=__import__('validate_delivery').validate(s,m,case.d)
log('C5_mixed_lane_control','same obligations block mixed lane',dict(errors=err(fs),delivery_ready=result['computed_delivery_ready']))
finally:case.tearDown()
args.output.write_text(json.dumps(results,ensure_ascii=False,indent=2))
print(json.dumps(results,ensure_ascii=False,indent=2))
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |