개인적으로 만든 워크플로우 목적의 스킬 모음입니다.
생각날때 업데이트합니다.
이 섹션은 각 스킬이 어떤 입력을 읽고, 어떤 판단을 하며, 어떤 산출물을 만들고, 어느 다음 스킬로 넘어가는지 확인하기 위한 흐름도이다.
- Mermaid 블록은 렌더링 가능한 다이어그램이다.
- 순수 텍스트 블록은 Mermaid 렌더링 없이도 같은 논리를 읽을 수 있는 흐름도이다.
.agent_goal/*.md는 장기 목표/권한/규칙의 GT로 취급한다..agent_advice/*는 비-GT 방향성 문서로만 취급한다..task/*,.agent_log/*,.issue/*,.schema/*,.contract/*,.validation/*는 워크플로우 증거와 추적 상태이다.- 완료 판정은
$validate-task-completion이 담당하며, 실행 성공/로그/대시보드/인덱스만으로 완료를 선언하지 않는다. - adapter나 caller가 verifier contract를 요구하는 measurable acceptance는 live verifier가 pass해야 완전하다. required verifier의
not_evaluated는 pass가 아니며, full close 대신 verifier follow-up, explicit descope, terminal blocker, user escalation 중 하나로 보존한다. - acceptance가 참조하는 gate의 required hook 부재,
pass_with_unobserved_axes, generation-dependent count key, below-policy residual value per cycle cost는 pass/advance/close 근거가 아니다. hook supply, axis supply, effective key/terminal-outcome fallback, residual descope plus next rung, terminal blocker, user escalation 중 하나로 보존한다. - acceptance scenario coverage, full body-free
command_argv, actionable blocker relation, stochastic feasibility는 해당 증거가 등장한 cycle에서 completion/advance 소비 전에 gate로 재확인한다.scenario_uncovered,acceptance_inversion,command_provenance_missing, repeatedblocker_opacity,predetermined_unreachable,floor_edge_envelope는 같은 결함 재시도가 아니라 scenario/argv/blocker/contract repair, descope, terminal blocker, user escalation 중 하나로 보존한다. - 구조 진전은 어댑터가
structure_metrics.global_*전역 불변량을 제공하면 per-scope 감소가 아니라 global high-water 이동으로 판정한다. - 장기 실행은 새 canonical phase가 아니라
step: run의 분기이다.event_kind: long_run_launch|long_run_monitor|long_run_harvest|long_run_finalize와long_run_role: launch|monitor|harvest|finalize를 기록하며,running과completed_pending_validation은 성공이 아니라 남은 harvest/validation의 증거이다. audit-cycle-loopback/scripts/anti_loop_gate_provider.py와orchestrate-task-cycle/scripts/detect_progress_loop.py는 기존 호출 호환 shim이고, 실제 구현은 각각anti_loop_provider/,progress_loop_detection/패키지 안의 모듈로 분리되어 있다.
flowchart TD
Start([사용자 요청 또는 cycle 시작])
Context["context 수집<br/>README, task.md, .agent_goal, .task, .issue, .schema, .contract, .validation"]
LedgerInit["$maintain-cycle-ledger<br/>cycle-id, stage.jsonl, current_stage.json, packets/ 초기화"]
Authority["$manage-agent-authority<br/>권한/외부 호출/검증 우선순위 정책 요약"]
Acceptance["$normalize-acceptance-and-demo<br/>acceptance, non-goals, demo, validation commands 정규화<br/>measurable -> verifier contract when mapped<br/>scenario coverage + required gate hook completeness"]
AdapterScan["repo-local skill adapter scan<br/>code_convention_contract, domain adapter, output-delta hook,<br/>target_required_verifier, goal_axis_map,<br/>count-key collapse, structure_metrics.global_* 탐색"]
RoutePlan["route_plan<br/>task.md 존재 여부와 cycle 경로 결정"]
NoTask{"task.md 없음?"}
DeriveInitial["$derive-improvement-task initial_init<br/>초기 task.md 생성"]
SchemaPreInit["$manage-schema-contracts<br/>초기 task가 schema/contract 영향이면 계약 정렬"]
ValPlan["$plan-validation-scope<br/>current_only / affected_chain / full_chain 결정"]
ValSetPlan["$build-validation-set-with-agents plan<br/>검증셋 필요 여부, oracle/split/leakage 정책"]
Governance["$task-md-agent-governance<br/>task.md 구현, worker 위임, repo audit, task_miss 기록"]
ResultContract1["$validate-subskill-result-contract<br/>governance/result fields 검사"]
CodeStructure["orchestrate scripts/code_structure_audit.py<br/>구조/컨벤션/semantic modularity audit packet"]
Run["$run-task-code-and-log<br/>명령 실행, full command_argv, 실패 autopsy,<br/>observed_producer_claim downgrade, .agent_log 기록<br/>long_run_launch 가능"]
Running{"run status = running?"}
Monitor["$monitor-running-execution<br/>canonical step=run + event_kind long_run_*<br/>PID/log/heartbeat/stop command, expected artifacts,<br/>remaining_validation 추적"]
Quality["$review-cycle-output-quality<br/>단일 read-only xhigh 출력 품질 리뷰<br/>zero mapped goal axes -> pass_with_unobserved_axes"]
Loopback["$audit-cycle-loopback<br/>semantic_progress, same-family loop, adapter metrics,<br/>3-state gates, verifier contract, count-key hygiene,<br/>goal-axis completeness, residual cost ratio,<br/>scenario/argv/blocker/stochastic findings,<br/>anti_loop_provider packet + root-cause ledger"]
ValSetBuild["$build-validation-set-with-agents build/consume<br/>validation assets 또는 oracle 결과 산출"]
SchemaPreDerive["$manage-schema-contracts pre-derive<br/>schema/contract 영향과 stale contract 확인"]
Visible["$record-visible-increment<br/>보이는 변화 기록; not_validation_evidence=true"]
GapAnalysis["repo_skill_gap_analysis<br/>adapter/skill gap 또는 skill-creator 후보"]
Profile["$profile-cycle-efficiency<br/>중복 로그, metadata-only 반복, command surface budget 감지"]
Slice["$optimize-task-slice<br/>state_transition, batch, evidence_supply, verifier_completion,<br/>scenario_supply, command_provenance_repair,<br/>blocker_contract_repair, consolidation 등 advisory"]
DeriveNext["$derive-improvement-task<br/>다음 task.md 또는 task_pack/terminal blocker 도출"]
SchemaPost["$manage-schema-contracts post-derive<br/>새 task/schema/contract 링크 정리"]
Index["$manage-task-state-index<br/>scan, link, audit; task/candidate/miss/log/schema/issue IDs"]
Validate["$validate-task-completion<br/>complete / partial / failed + progress_verdict<br/>required verifier pass + structure global effect 확인<br/>long-run pending, scenario, command provenance,<br/>blocker actionability, stochastic feasibility gate"]
Issue["$manage-implementation-issues<br/>issue open/update/resolve, .issue mirror, GitHub fallback"]
Commit["$repo-change-commit<br/>coherent implementation/checkpoint commit"]
Dashboard["$render-cycle-dashboard<br/>한국어 dashboard.md"]
Report["$maintain-cycle-ledger / assemble_cycle_report.py<br/>한국어 final_report.md"]
Closeout["$repo-change-commit closeout<br/>report/dashboard/ledger closeout commit"]
End([cycle 결과 보고])
Start --> Context --> LedgerInit --> Authority --> Acceptance --> AdapterScan --> RoutePlan --> NoTask
NoTask -- yes --> DeriveInitial --> SchemaPreInit --> ValPlan
NoTask -- no --> ValPlan
ValPlan --> ValSetPlan --> Governance --> ResultContract1 --> CodeStructure --> Run --> Running
Running -- yes --> Monitor --> Validate
Running -- no --> Quality --> Loopback --> ValSetBuild --> SchemaPreDerive --> Visible --> GapAnalysis --> Profile --> Slice --> DeriveNext --> SchemaPost --> Index --> Validate
Validate --> Issue --> Commit --> Dashboard --> Report --> Closeout --> End
flowchart TD
RawPrompt([raw user goal prompt])
Shape["$shape-agent-goal-prompt<br/>raw prompt 보존, draft 생성, 3명 이상 critic review"]
Goal["$manage-agent-goal<br/>.agent_goal/final_goal.md<br/>.agent_goal/conventions.md"]
DeepGate{"base goal files<br/>실내용 존재?"}
Deep["$deep-interview-goal-context<br/>stateful single-question interview"]
Questions[".interview/questions.md<br/>.interview/answers.md<br/>.interview/state.md"]
Drafts[".interview/drafts/<br/>architecture, theory, schema_contract, authority draft"]
EvidenceReview{"3 evidence reviewers<br/>모두 CONFIRM?"}
AuditReview{"3 critical auditors<br/>모두 CONFIRM?"}
UserConfirm{"user final confirmation<br/>있음?"}
FinalReview{"3-6 final reviewers<br/>모두 safe_to_write?"}
Architecture["$manage-goal-architecture<br/>.agent_goal/goal_architecture.md"]
Theory["$manage-goal-theory<br/>.agent_goal/goal_theory.md"]
SchemaGoal["goal_schema_contract write<br/>.agent_goal/goal_schema_contract.md"]
Authority["$manage-agent-authority<br/>.agent_goal/agent_authority.md"]
SchemaRegistry["$manage-schema-contracts<br/>.schema/.contract registry 정렬"]
AdviceIn["외부 조언 파일 또는 본문"]
Advice["$manage-external-advice<br/>raw -> active/deferred/applied/rejected<br/>Part E advice는 기존 판정 의미/키 개정으로 정규화"]
AdvicePacket["active advice packet<br/>not_goal_truth=true"]
Index["$manage-task-state-index<br/>goal-*, int-*, adv-*, schema-* 링크"]
Consumers["orchestrate / derive / governance / validate<br/>GT와 비-GT를 분리 소비<br/>additive signal 대신 acceptance/gate/progress key에 반영"]
RawPrompt --> Shape --> Goal
Goal --> DeepGate
DeepGate -- no --> Goal
DeepGate -- yes --> Deep --> Questions --> Drafts --> EvidenceReview
EvidenceReview -- no --> Questions
EvidenceReview -- yes --> AuditReview
AuditReview -- no --> Questions
AuditReview -- yes --> UserConfirm
UserConfirm -- no --> Questions
UserConfirm -- yes --> FinalReview
FinalReview -- no --> Questions
FinalReview -- yes --> Architecture --> Index
FinalReview -- yes --> Theory --> Index
FinalReview -- yes --> SchemaGoal --> SchemaRegistry --> Index
FinalReview -- yes --> Authority --> Index
AdviceIn --> Advice --> AdvicePacket --> Consumers
Index --> Consumers
flowchart TD
Trigger([next task 필요 또는 명시적 task doctor 요청])
ExplicitDoctor{"명시적 doctor/replace/pack 지시?"}
TaskDoctor["$task-doctor<br/>task.md retarget/replace 또는 task_pack proposal"]
DoctorRead["읽기: task.md, .agent_goal, named advice, .task, .issue, .schema"]
DoctorArchive["$record-agent-work-log<br/>old task를 past_task로 보존"]
DoctorWrite["새 task.md 또는 .task/task_pack/*.json + *.md 작성"]
DoctorIndex["$manage-task-state-index<br/>supersedes, promoted_from_pack, advice links"]
DoctorCommit["$repo-change-commit<br/>task-direction change commit"]
NormalDerive["$derive-improvement-task"]
Inputs["입력 수집<br/>.agent_goal, authority, active advice, .issue,<br/>task_miss, candidates, task_pack, schema/contract,<br/>qualitative_review, loopback, validation/profile packets"]
Alignment["$inspect-repo-with-agents<br/>goal/convention/schema alignment"]
MissAgents["2-4 task_miss analysis agents"]
CandidateScan[".task/candidate_task 분류<br/>now/blocked/obsolete/duplicate"]
PackScan[".task/task_pack 분류<br/>promote/insert/reorder/skip/supersede/terminal"]
IssueFit["$manage-implementation-issues issue-fit agent<br/>fit/partial/misfit/unknown"]
ImproveAgents["3 parallel improvement-analysis agents<br/>goal fit, architecture fit, miss/risk fit"]
Synthesis["1 xhigh synthesis agent<br/>한 개 task 또는 terminal blocker 선택"]
Decision{"선택 결과"}
ArchivePast["$record-agent-work-log<br/>기존 task.md past_task archive"]
WriteTask["task.md 작성<br/>Execution Environment 최상단 포함"]
MutatePack["task_pack transaction<br/>create/promote/insert/reorder/skip/supersede/terminal"]
Candidates["unselected proposals -> .task/candidate_task/<br/>selected candidate는 task 작성 후 삭제"]
Terminal["terminal_blocker / user_escalation<br/>sealed family와 missing input 기록"]
DeriveIndex["$manage-task-state-index scan + audit"]
LoopInputs["run + quality review + output-delta artifacts<br/>failure autopsy, runner validation, gate states,<br/>long-run history, scenario/argv/blocker/stochastic fields"]
ProgressDetect["detect_progress_loop.py<br/>progress_loop_detection analyzer<br/>evidence -> gates/findings/terminal packet"]
Loopback["$audit-cycle-loopback<br/>anti_loop_gate_provider.py legacy shim<br/>-> anti_loop_provider.api/evaluator"]
LoopGate["anti_loop_progress_gate<br/>effective_allowed_dispositions, allowed_task_kinds,<br/>adapter_wiring_defect, adapter_mandate,<br/>cumulative_goal_distance_stalled,<br/>failure_surface_stage_gate, verification_source_separation,<br/>root_cause_hypothesis_gate, hypothesis_exhausted,<br/>acceptance_unreachable, unverifiable_acceptance_contract,<br/>pass_with_unobserved_axes, count_key_hygiene,<br/>scenario_uncovered, acceptance_inversion,<br/>command_provenance_missing, blocker_opacity,<br/>predetermined_unreachable, floor_edge_envelope,<br/>residual value/cycle cost, metric_verifier_not_evaluated,<br/>structure_global_invariant_metrics"]
LongRunDebt{"active long_run_branch<br/>pending final output?"}
LongRunRoute["derive constraint<br/>monitor/harvest/finalize same run_id<br/>or terminal/user escalation"]
VerifierDebt{"required verifier<br/>not_evaluated?"}
VerifierRoute["derive constraint<br/>verifier hook/metric correction/descope/<br/>terminal blocker/user escalation"]
EvidenceDebt{"scenario/argv/blocker/stochastic<br/>repair required?"}
EvidenceRoute["derive constraint<br/>scenario_supply / command_provenance_repair /<br/>blocker_contract_repair / contract revision"]
GlobalInvariant{"structure global invariant<br/>metrics present?"}
GlobalMoved{"global high-water<br/>moved?"}
GlobalRoute["global invariant present + local-only reduction<br/>cannot consume global structure target"]
Slice["$optimize-task-slice<br/>state_transition/batch/evidence/verifier_completion/<br/>scenario_supply/command_provenance_repair/consolidation advisory"]
Profile["$profile-cycle-efficiency<br/>sprawl, duplicate evidence, safety_only loops"]
Trigger --> ExplicitDoctor
ExplicitDoctor -- yes --> DoctorRead --> TaskDoctor --> DoctorArchive --> DoctorWrite --> DoctorIndex --> DoctorCommit
ExplicitDoctor -- no --> NormalDerive
LoopInputs --> ProgressDetect --> Loopback --> LoopGate --> LongRunDebt
LongRunDebt -- yes --> LongRunRoute --> NormalDerive
LongRunDebt -- no --> VerifierDebt
VerifierDebt -- yes --> VerifierRoute --> NormalDerive
VerifierDebt -- no --> EvidenceDebt
EvidenceDebt -- yes --> EvidenceRoute --> NormalDerive
EvidenceDebt -- no --> GlobalInvariant
GlobalInvariant -- no --> Slice
GlobalInvariant -- yes --> GlobalMoved
GlobalMoved -- no --> GlobalRoute --> NormalDerive
GlobalMoved -- yes --> Slice --> NormalDerive
Profile --> NormalDerive
NormalDerive --> Inputs --> Alignment --> MissAgents --> CandidateScan --> PackScan --> IssueFit --> ImproveAgents --> Synthesis --> Decision
Decision -- standalone task --> ArchivePast --> WriteTask --> Candidates --> DeriveIndex
Decision -- pack mutation --> ArchivePast --> MutatePack --> WriteTask --> DeriveIndex
Decision -- no viable task --> Terminal --> DeriveIndex
flowchart TD
Task([active task.md])
Acceptance["$normalize-acceptance-and-demo<br/>acceptance packet<br/>envelope + acceptance_verifier_contract<br/>scenario coverage + required hook completeness"]
ValScope["$plan-validation-scope<br/>changed surfaces -> current_only/affected_chain/full_chain"]
EnvNeed{"Python/dependency constrained?"}
FindEnv["$find-local-python-envs<br/>local env inventory, ranked run commands"]
DepMissing{"필수 dependency/cache 없음?"}
Install["$install-deps-with-agent<br/>find-local-python-envs 후 exactly one install agent"]
ValSetNeed{"validation set 필요?"}
ValSet["$build-validation-set-with-agents<br/>plan/build/consume, .validation assets"]
Governance["$task-md-agent-governance"]
ReadTask["task.md, authority, advice, schema, code_convention_contract 읽기"]
MapRepo["repo map<br/>git status, rg --files, manifests, tests"]
Workers["gpt-5.5 worker agents<br/>disjoint write scopes, authority/advice/convention 포함"]
Integrate["worker changes 통합<br/>format/tests, convention_conformance 확인"]
Inspect["$inspect-repo-with-agents<br/>3-6 read-only code/schema/authority/generalization audit"]
TaskMiss[".task/task_miss report<br/>miss/resolved/deleted cleanup evidence"]
Run["$run-task-code-and-log"]
Execute["정해진 command 실행<br/>validation_profile 준수<br/>full body-free command_argv 보존"]
Failure{"실패 또는 gate unsatisfiable?"}
Autopsy["safe_failure_autopsy.py<br/>execution stage ladder 정규화,<br/>safe scalar diagnostics,<br/>gate selfcheck -> self_inflicted_gate_defect"]
Log["$record-agent-work-log<br/>.agent_log + run evidence"]
Running{"long-running authorized?"}
Monitor["$monitor-running-execution<br/>step=run long_run_* event<br/>run_id/owner_task_id/log/output/artifacts/status<br/>completed_pending_validation != success"]
Quality["$review-cycle-output-quality<br/>single reviewer, output quality/delta/no-overclaim<br/>goal-axis completeness"]
Completion["$validate-task-completion"]
Gates["gates: env, execution, repo audit, OOM if relevant,<br/>task_miss, issue, advice, schema, acceptance,<br/>required verifier/hook pass, goal axes,<br/>scenario coverage, command provenance,<br/>blocker actionability, stochastic feasibility,<br/>count-key hygiene, residual cost ratio,<br/>structure global invariant, behavior-change live evidence, ID"]
Verdict{"validation_verdict<br/>complete / partial / failed"}
Progress{"progress_verdict<br/>advanced / safety_only / no_progress / regressed"}
Task --> Acceptance --> ValScope --> EnvNeed
EnvNeed -- yes --> FindEnv --> DepMissing
DepMissing -- yes --> Install --> ValSetNeed
DepMissing -- no --> ValSetNeed
EnvNeed -- no --> ValSetNeed
ValSetNeed -- yes --> ValSet --> Governance
ValSetNeed -- no --> Governance
Governance --> ReadTask --> MapRepo --> Workers --> Integrate --> Inspect --> TaskMiss --> Run
Run --> Execute --> Failure
Failure -- yes --> Autopsy --> Log
Failure -- no --> Log
Log --> Running
Running -- yes --> Monitor --> Completion
Running -- no --> Quality --> Completion
Completion --> Gates --> Verdict --> Progress
flowchart TD
Evidence([어떤 스킬이 artifact 생성])
Cache["$manage-evidence-cache<br/>fingerprint -> reuse/fresh_required/stale/unsafe_to_reuse"]
Log["$record-agent-work-log<br/>.agent_log/YYYY-MM-DD/*.md + index.jsonl"]
Ledger["$maintain-cycle-ledger<br/>stage.jsonl/current_stage.json/packets<br/>long-run events remain step=run"]
Contract["$validate-subskill-result-contract<br/>stage handoff required fields 검사"]
Index["$manage-task-state-index<br/>task/log/run/audit/val/miss/issue/goal/adv/schema IDs"]
Visible["$record-visible-increment<br/>visible delta, not_validation_evidence=true"]
Issue["$manage-implementation-issues<br/>GitHub 또는 .issue fallback, issue lifecycle"]
Commit["$repo-change-commit<br/>diff 분류, gitignore 정리, validation context, exact staging"]
Dashboard["$render-cycle-dashboard<br/>cycle dashboard.md 한국어"]
Report["$maintain-cycle-ledger / assemble_cycle_report.py<br/>final_report.md 한국어"]
Closeout["$repo-change-commit closeout<br/>dashboard/report/ledger artifacts commit"]
User([사용자에게 결과 보고])
Evidence --> Cache
Evidence --> Log
Evidence --> Contract --> Ledger
Log --> Index
Ledger --> Index
Evidence --> Visible --> Ledger
Index --> Issue
Issue --> Commit
Index --> Commit
Ledger --> Dashboard --> Report --> Closeout --> User
flowchart TD
Need([지원/진단 필요])
PythonNeed{"Python 실행환경 또는 import 문제?"}
Env["$find-local-python-envs<br/>manifest/import/env inventory, ranked commands"]
InstallNeed{"기존 env/cache로 부족?"}
Install["$install-deps-with-agent<br/>cache-first 후 one install/download agent"]
OOMNeed{"large data/model/batch/concurrency/OOM risk?"}
OOM["$inspect-oom-risk<br/>static risk map, scaling variable, memory domain, mitigation"]
RepoReviewNeed{"multi-agent repo audit/review 필요?"}
RepoReview["$inspect-repo-with-agents<br/>3-6 perspectives, file/line findings"]
RunNeed{"long-running execution 상태 확인?"}
Monitor["$monitor-running-execution<br/>canonical step=run monitor event<br/>running/completed_pending_validation != success"]
EvidenceNeed{"이전 evidence 재사용 가능성?"}
Cache["$manage-evidence-cache<br/>reuse 후보만 제공; pass로 변환하지 않음"]
Output([지원 결과를 caller packet 또는 사용자 보고로 반환])
Need --> PythonNeed
PythonNeed -- yes --> Env --> InstallNeed
InstallNeed -- yes --> Install --> Output
InstallNeed -- no --> Output
PythonNeed -- no --> OOMNeed
OOMNeed -- yes --> OOM --> Output
OOMNeed -- no --> RepoReviewNeed
RepoReviewNeed -- yes --> RepoReview --> Output
RepoReviewNeed -- no --> RunNeed
RunNeed -- yes --> Monitor --> Output
RunNeed -- no --> EvidenceNeed
EvidenceNeed -- yes --> Cache --> Output
EvidenceNeed -- no --> Output
flowchart TD
Inputs["inputs<br/>registry, artifact paths, changed files,<br/>runner validation, output delta,<br/>failure autopsies, gate states,<br/>scenario/command/blocker/stochastic fields,<br/>long-run event history"]
LoopbackCLI["anti_loop_gate_provider.py<br/>legacy CLI/API shim"]
LoopbackAPI["anti_loop_provider/api.py<br/>export bridge + runtime cache sync"]
LoopbackEval["anti_loop_provider/evaluator.py<br/>LoopbackEvaluator / evaluate"]
AdapterLayer["adapters.py + domain.py + quality.py<br/>domain adapter, quality vector,<br/>facet/root-family normalization"]
LoopGates["gates + acceptance + verification + blockers<br/>coverage/substance, verifier status,<br/>scenario coverage, command provenance,<br/>blocker actionability, stochastic feasibility,<br/>failure surface stage, source separation,<br/>adapter and chain stalls"]
RootCause["root_cause.py + root_cause_runtime.py<br/>hypotheses, repo-owned actionability,<br/>exhaustion and untried repair"]
LoopRegistry["registry.py<br/>anti-loop registry, root-cause ledger,<br/>sealed blocker families"]
LoopPacket["packet.py + assembly.py + outcome.py<br/>anti_loop_progress_gate packet<br/>effective dispositions + allowed task kinds"]
DetectCLI["detect_progress_loop.py<br/>legacy CLI shim"]
DetectAnalyzer["progress_loop_detection/cli.py + analysis.py<br/>ProgressLoopAnalyzer"]
DetectEvidence["evidence.py + fingerprints.py + normalizers.py<br/>structured evidence, observed output class,<br/>root axis/key, feature symbol"]
DetectGates["output_delta_gate.py + input_delta_gate.py + validator_gate.py<br/>declared vs observed delta,<br/>supplied input delta, validator integrity"]
DetectTerminal["analysis_gates.py + terminal.py + provider.py<br/>safety-only streak, metadata-only loops,<br/>terminal quiescence/escalation,<br/>provider reattempt gates"]
DetectRegistry["progress_loop_detection/registry.py<br/>feature symbol registry + terminal history"]
DetectPacket["analysis_result.py<br/>loop-breaker packet"]
Derive["$derive-improvement-task<br/>next task / terminal blocker / user escalation constraints"]
Inputs --> LoopbackCLI --> LoopbackAPI --> LoopbackEval
LoopbackEval --> AdapterLayer --> LoopGates --> LoopPacket
LoopbackEval --> RootCause --> LoopRegistry --> LoopPacket
LoopPacket --> Derive
Inputs --> DetectCLI --> DetectAnalyzer --> DetectEvidence --> DetectGates --> DetectTerminal --> DetectRegistry --> DetectPacket --> Derive
아래 블록은 Mermaid 렌더링 없이도 구조가 보이도록 박스, 분기, 화살표만으로 그린 텍스트 도형이다.
+--------------------------------------------------------------------------------+
| USER REQUEST / CYCLE START |
+--------------------------------------------------------------------------------+
|
v
+--------------------------------------------------------------------------------+
| CONTEXT |
| - README, task.md, .agent_goal, .task, .issue, .schema, .contract 확인 |
+--------------------------------------------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $maintain-cycle-ledger | --> | .task/cycle/<cycle-id>/ |
| cycle id / stage ledger | | stage.jsonl, current_stage.json, packets/ |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $manage-agent-authority | --> | authority_policy |
| 권한/외부호출/검증 우선순위 | | agent_authority.md or default permissions |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $normalize-acceptance | --> | acceptance packet |
| acceptance/non-goal/demo | | envelope_floor / deficit_axis when present |
| measurable target contract | | acceptance_verifier_contract when mapped |
| | | required verifier/hook not_evaluated != pass|
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| repo-local adapter scan | --> | code_convention_contract |
| adapters and hooks | | domain adapter / output-delta / gate hooks |
| | | target_required_verifier, goal_axis_map, |
| | | count-key collapse, global_* metrics |
+----------------------------+ +---------------------------------------------+
|
v
+--------------------+
| task.md exists ? |
+--------------------+
| yes | no
v v
+----------------------------+ +-------------------------------------------+
| continue active task | | $derive-improvement-task initial_init |
| 기존 task.md로 진행 | | initial task.md 생성 |
+----------------------------+ +-------------------------------------------+
| |
+----------------------+-------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $plan-validation-scope | --> | validation_profile |
| changed surfaces classify | | current_only / affected_chain / full_chain |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $build-validation-set | --> | validation-set plan |
| plan mode | | oracle / split / leakage / label policy |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $task-md-agent-governance | --> | implementation + audit + task_miss |
| worker implementation | | worker outputs, repo audit, miss reports |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| result contract + structure| --> | validated handoff + structure packet |
| field check / code audit | | required fields, shard/coupling/depth |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $run-task-code-and-log | --> | run result + .agent_log |
| command execution | | success/running/partial/failed/not_run |
| full body-free argv | | long_run_launch may only create handoff |
+----------------------------+ +---------------------------------------------+
|
v
+--------------------+
| run is running ? |
+--------------------+
| yes | no
v v
+----------------------------+ +-------------------------------------------+
| $monitor-running-execution | | $review-cycle-output-quality |
| step=run long_run_* event | | single read-only quality reviewer |
| PID/log/heartbeat/artifacts| +-------------------------------------------+
| running/pending != success |
+----------------------------+ |
| v
| +---------------------------------------+
| | $audit-cycle-loopback |
| | anti_loop_provider package packet |
| | semantic_progress / anti-loop gates |
| | pass/fail/not_evaluated gate meaning |
| | verifier debt + count-key hygiene |
| | failure stage + root-cause ledger |
| | goal-axis/residual/global invariant |
| +---------------------------------------+
| |
| v
| +---------------------------------------+
| | validation build + schema + increment |
| | $build-validation-set build/consume |
| | $manage-schema-contracts pre-derive |
| | $record-visible-increment |
| +---------------------------------------+
| |
| v
| +---------------------------------------+
| | derive preparation |
| | $profile-cycle-efficiency |
| | $optimize-task-slice |
| | verifier_completion when required |
| | $derive-improvement-task |
| | $manage-schema-contracts post-derive |
| | $manage-task-state-index |
| +---------------------------------------+
| |
+----------------------+-----------------------+
|
v
+----------------------------+ +---------------------------------------------+
| $validate-task-completion | --> | validation_verdict + progress_verdict |
| final completion gate | | complete/partial/failed + progress class |
| | | required verifier pass before full close |
| | | scenario/command/blocker/stochastic gates |
| | | long-run pending blocks completion |
| | | global structure target not consumed local |
+----------------------------+ +---------------------------------------------+
|
v
+----------------------------+ +---------------------------------------------+
| issue -> commit -> report | --> | $manage-implementation-issues |
| final workflow closeout | | $repo-change-commit |
| | | $render-cycle-dashboard / final_report.md |
| | | closeout commit |
+----------------------------+ +---------------------------------------------+
|
v
+--------------------------------------------------------------------------------+
| USER REPORT |
+--------------------------------------------------------------------------------+
+-------------------------------+
| raw user goal prompt |
+-------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $shape-agent-goal-prompt | ---> | draft final_goal.md / conventions.md |
| preserve raw prompt | | 3+ critics: intent/overreach/risk |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $manage-agent-goal | ---> | .agent_goal/final_goal.md |
| merge supported goal truth | | .agent_goal/conventions.md |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+
| base goal files complete ? |
| final_goal + conventions |
+-------------------------------+
| yes | no
v v
+-------------------------------+ +----------------------------------------+
| $deep-interview-goal-context | | return to $manage-agent-goal |
| stateful one-question loop | | prerequisite objective/conventions |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| .interview/questions/answers | ---> | one pending question per invocation |
| state.md tracks active batch | | answers become interview evidence |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| .interview/drafts | ---> | draft architecture/theory/schema/ |
| not final goal truth | | authority files |
+-------------------------------+ +----------------------------------------+
|
v
+-----------------------------+
| 3 evidence reviewers CONFIRM?|
+-----------------------------+
| yes | no
v v
+-----------------------------+ +-------------------------------+
| 3 critical auditors CONFIRM?| | revise drafts or ask next |
+-----------------------------+ | targeted question |
| yes +-------------------------------+
v
+-----------------------------+
| user final confirmation ? |
+-----------------------------+
| yes | no
v v
+-----------------------------+ +-------------------------------+
| 3-6 final reviewers safe ? | | hold writes; ask/record |
+-----------------------------+ | missing confirmation |
| yes +-------------------------------+
v
+-------------------------------+ +----------------------------------------+
| final .agent_goal writes | ---> | goal_architecture.md |
| all-or-nothing after confirm | | goal_theory.md |
| | | goal_schema_contract.md |
| | | agent_authority.md |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $manage-schema-contracts | ---> | .schema/.contract aligned to |
| $manage-task-state-index | | goal_schema_contract; goal/int IDs |
+-------------------------------+ +----------------------------------------+
External advice side path:
+-------------------------------+ +----------------------------------------+
| external advice Markdown | ---> | $manage-external-advice |
+-------------------------------+ | raw -> active/deferred/applied/rejected|
| workflow advice becomes in-place |
| acceptance/gate/progress-key revision |
+----------------------------------------+
|
v
+----------------------------------------+
| active advice packet |
| not_goal_truth=true |
| consumed by orchestrate/derive/ |
| governance/validate only as non-GT |
+----------------------------------------+
+-------------------------------+
| next task needed |
| or explicit task doctor input |
+-------------------------------+
|
v
+-----------------------------+
| explicit doctor/pack/retarget?|
+-----------------------------+
| yes | no
v v
+-------------------------------+ +---------------------------------------+
| $task-doctor | | $derive-improvement-task |
| pre-cycle intervention only | | normal next-task selection |
+-------------------------------+ +---------------------------------------+
| |
v v
+-------------------------------+ +---------------------------------------+
| read direction sources | | load planning context |
| user instruction, task.md, | | .agent_goal, authority, advice, |
| .agent_goal, named advice, | | .issue, task_miss, candidates, |
| .task, .issue, schema | | task_pack, schema, review, loopback |
+-------------------------------+ +---------------------------------------+
| |
v v
+-------------------------------+ +---------------------------------------+
| archive old task | | analysis fanout |
| $record-agent-work-log | | - goal/schema alignment audit |
| past_task before overwrite | | - 2-4 task_miss agents |
+-------------------------------+ | - candidate scan |
| | - task_pack scan |
v | - 1 issue-fit agent |
+-------------------------------+ | - 3 improvement agents |
| write task.md or task_pack | | - 1 synthesis agent |
| preserve scope_fidelity, | +---------------------------------------+
| envelope, verifier contract, | |
| terminal/global residuals, | |
| G-axis/cost residuals | |
+-------------------------------+ v
| +---------------------------------------+
v | apply hard selection gates |
+-------------------------------+ | anti_loop effective dispositions, |
| index / schema / issue / | | allowed_task_kinds, adapter defects, |
| advice reconciliation | | chain stalls, sealed families, |
+-------------------------------+ | verifier debt, count-key hygiene, |
| goal-axis, residual cost, global keys |
| | no producer self-report truth |
| +---------------------------------------+
v |
+-------------------------------+ v
| optional task-direction commit| +---------------------------------------+
| $repo-change-commit | | synthesis decision |
+-------------------------------+ +---------------------------------------+
| | standalone task
| v
| +----------------------------------+
| | archive old task -> write task.md |
| | keep/delete candidates safely |
| +----------------------------------+
| |
| | pack mutation
| v
| +----------------------------------+
| | create/promote/insert/reorder/ |
| | skip/supersede task_pack + task |
| +----------------------------------+
| |
| | no viable task
| v
| +----------------------------------+
| | terminal_blocker / user_escalation|
| | sealed family + missing input |
| +----------------------------------+
| |
+---------------------------+----------+
|
v
+----------------------------------+
| $manage-task-state-index |
| scan + link + audit |
+----------------------------------+
Anti-loop and efficiency inputs into derive:
+-------------------------------+ +----------------------------------------+
| run + quality + output-delta | ---> | detect_progress_loop.py |
| failure autopsy + gate states | | progress_loop_detection package |
+-------------------------------+ | evidence/fingerprint/root feature |
| output/input/validator gates |
| terminal quiescence/escalation packet |
+----------------------------------------+
|
v
+----------------------------------------+
| $audit-cycle-loopback |
| anti_loop_provider package |
| semantic_progress, root family, |
| adapter metrics, effective dispositions|
| evaluation_status pass/fail/not_eval |
| failure_surface_stage_gate |
| root-cause ledger + sealed families |
| verifier, count-key, residual, global |
+----------------------------------------+
|
v
+----------------------------------------+
| $optimize-task-slice |
| state_transition / batch / evidence / |
| verifier_completion / consolidation / |
| stop advisory |
+----------------------------------------+
|
v
+----------------------------------------+
| $profile-cycle-efficiency |
| duplicate evidence, metadata-only, |
| safety_only loops, sprawl budgets |
+----------------------------------------+
+-------------------------------+
| active task.md |
+-------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $normalize-acceptance | ---> | acceptance packet |
| criteria / non-goals / demo | | validation commands + envelope |
| measurable -> verifiable | | acceptance_verifier_contract |
| scenario coverage | | required hook completeness |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $plan-validation-scope | ---> | validation_profile |
| changed surface classification| | current_only / affected_chain / full |
+-------------------------------+ +----------------------------------------+
|
v
+-----------------------------+
| Python/env dependency need? |
+-----------------------------+
| yes | no
v v
+-------------------------------+ +---------------------------------------+
| $find-local-python-envs | | skip env discovery |
| rank existing env commands | +---------------------------------------+
+-------------------------------+ |
| |
v |
+-----------------------------+ |
| missing dependency/cache ? | |
+-----------------------------+ |
| yes | no |
v v |
+-------------------+ +-------------------+ |
| $install-deps | | use ranked env | |
| one install agent | | no install | |
+-------------------+ +-------------------+ |
| | |
+---------+---------+----------------------------+
|
v
+-----------------------------+
| validation set needed ? |
+-----------------------------+
| yes | no
v v
+-------------------------------+ +---------------------------------------+
| $build-validation-set | | continue to governance |
| plan/build/consume assets | +---------------------------------------+
+-------------------------------+ |
| |
+--------------------------+---------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $task-md-agent-governance | ---> | implementation outputs |
| read task/authority/advice/ | | worker changes + integration |
| schema/convention contract | | repo audit + .task/task_miss |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $run-task-code-and-log | ---> | run evidence + .agent_log |
| execute specified command | | status: success/running/partial/failed|
| preserve full argv | | command_provenance_missing blocks |
| | | baseline/reproduction/comparison |
+-------------------------------+ +----------------------------------------+
|
v
+-----------------------------+
| failure or gate unsatisfiable?|
+-----------------------------+
| yes | no
v v
+-------------------------------+ +---------------------------------------+
| safe_failure_autopsy.py | | keep normal run evidence |
| execution stage ladder | +---------------------------------------+
| safe scalar diagnostics only | |
| gate selfcheck defect class | |
+-------------------------------+ |
| |
+--------------------------+---------------------+
|
v
+-----------------------------+
| long-running authorized ? |
+-----------------------------+
| yes | no
v v
+-------------------------------+ +---------------------------------------+
| $monitor-running-execution | | $review-cycle-output-quality |
| step=run long_run_* event | | single read-only reviewer |
| running/pending != success | | |
+-------------------------------+ +---------------------------------------+
| |
+--------------------------+---------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $validate-task-completion | ---> | validation_verdict |
| final gate matrix | | complete / partial / failed |
| env/run/repo/OOM/miss/issue/ | | progress_verdict: advanced / |
| advice/schema/acceptance/ID | | safety_only / no_progress / regressed |
| verifier/structure global | | not_evaluated verifier blocks |
| scenario/argv/blocker/stoch | | complete; pending long-run blocks |
+-------------------------------+ +----------------------------------------+
+-------------------------------+
| support / diagnostic need |
+-------------------------------+
|
v
+-----------------------------+
| Python import/env problem ? |
+-----------------------------+
| yes | no
v v
+-------------------------------+ +---------------------------------------+
| $find-local-python-envs | | check OOM risk need |
| manifests + env inventory | +---------------------------------------+
| output: ranked run commands | |
+-------------------------------+ v
| +-----------------------------+
v | OOM/memory risk surface ? |
+-----------------------------+ +-----------------------------+
| existing env/cache enough ? | | yes | no
+-----------------------------+ v v
| yes | no +--------------------+ +------------------+
v v | $inspect-oom-risk | | repo audit need ?|
+-------------------+ +-------------------+ +------------------+
| return command | | $install-deps | |
| no install | | one setup agent | v
+-------------------+ +-------------------+ +---------------------+
| | | $inspect-repo-with |
+---------+--------+ | -agents |
| | 3-6 perspectives |
| +---------------------+
| |
| v
| +----------------------+
| | running run check ? |
| +----------------------+
| | yes | no
| v v
| +-------------------+ +------------------+
| | $monitor-running | | evidence reuse ? |
| | step=run event | +------------------+
| | pending != pass |
| +-------------------+ | yes | no
| | v v
| | +-------------------+ +---------+
| | | $manage-evidence | | return |
| | | -cache | | support |
| | | reuse/stale only | | result |
| | +-------------------+ +---------+
| | |
+--------------------------------------+-------------+
|
v
+-----------------------------+
| caller packet or user report|
+-----------------------------+
+-------------------------------+
| any skill creates evidence |
| run/audit/validation/report |
+-------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $validate-subskill-result | ---> | contract packet |
| required fields warn/block | | task_id, step, blockers, evidence |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $maintain-cycle-ledger | ---> | .task/cycle/<cycle-id>/stage.jsonl |
| append canonical stage event | | current_stage.json, packets/ |
| long-run remains step=run | | event_kind long_run_* when applicable |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $record-agent-work-log | ---> | .agent_log/YYYY-MM-DD/*.md |
| factual intent/work/result/ | | .agent_log/index.jsonl |
| shortcomings | | log-* IDs when indexed |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $manage-task-state-index | ---> | .task/index.jsonl + index.md |
| append-only IDs and links | | task/log/run/audit/val/miss/issue/ |
| global consistency audit | | goal/adv/schema IDs |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $record-visible-increment | ---> | .task/delta/<cycle-id>-visible-delta |
| before/after user-visible | | not_validation_evidence=true |
| workflow change only | | not a completion proof |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $manage-implementation-issues | ---> | GitHub issue or .issue/open|resolved |
| issue open/update/resolve | | requires run/validation evidence |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $repo-change-commit | ---> | commit hash or skip/block reason |
| classify dirty worktree | | exact staging, validation context |
| source/workflow/noise split | | safety_only/partial blockers in msg |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| $render-cycle-dashboard | ---> | dashboard.md in Korean |
| stage/verdict/blocker summary | | malformed/running/partial visible |
+-------------------------------+ +----------------------------------------+
|
v
+-------------------------------+ +----------------------------------------+
| final_report + closeout commit| ---> | final_report.md + closeout commit |
| Korean report, canonical IDs | | report/dashboard/ledger artifacts |
+-------------------------------+ +----------------------------------------+
+----------------------------------------+
| inputs |
| registry/artifact paths/changed files |
| runner validation/output delta |
| failure autopsy/gate state JSON |
| scenario/argv/blocker/stochastic JSON |
| long-run event history |
+----------------------------------------+
| |
v v
+----------------------------------+ +----------------------------------+
| anti_loop_gate_provider.py | | detect_progress_loop.py |
| legacy CLI/API shim | | legacy CLI shim |
+----------------------------------+ +----------------------------------+
| |
v v
+----------------------------------+ +----------------------------------+
| anti_loop_provider/api.py | | progress_loop_detection/cli.py |
| export bridge/runtime caches | | argparse/constant registry check |
+----------------------------------+ +----------------------------------+
| |
v v
+----------------------------------+ +----------------------------------+
| evaluator.py | | analysis.py |
| LoopbackEvaluator/evaluate | | ProgressLoopAnalyzer mixins |
+----------------------------------+ +----------------------------------+
| |
v v
+----------------------------------+ +----------------------------------+
| adapters/domain/quality | | evidence/fingerprints/normalizers|
| quality vectors, root families | | evidence items, root axis/key |
+----------------------------------+ +----------------------------------+
| |
v v
+----------------------------------+ +----------------------------------+
| gates/acceptance/verification | | output/input/validator gates |
| scenario/argv/blocker/stochastic | | terminal/provider gates |
| blockers/chain/root_cause | | quiescence/escalation |
| failure surface stage gate | | completion repair constraints |
+----------------------------------+ +----------------------------------+
| |
v v
+----------------------------------+ +----------------------------------+
| registry/root-cause ledger | | progress symbol registry |
| sealed blocker families | | loop-breaker result packet |
+----------------------------------+ +----------------------------------+
| |
+-------------------+------------------+
|
v
+-----------------------------+
| derive-improvement-task |
| task/terminal/escalation |
| constraints |
+-----------------------------+
orchestrate-task-cycle
context -> ledger -> authority -> acceptance/verifier/scenario contracts -> adapter scan -> validation planning -> governance -> run or long-run branch -> review/loopback or monitor -> derive/index -> validate gates -> issue -> commit -> dashboard/report -> closeout
maintain-cycle-ledger
cycle init -> stage append -> packet link -> dashboard/final_report render support
validate-subskill-result-contract
subskill result -> required fields including long-run detail, command_argv, scenario/blocker/stochastic gates -> warn/block -> ledger event -> downstream warnings
manage-agent-goal
user goal/conventions -> preserve/merge -> final_goal.md + conventions.md
shape-agent-goal-prompt
raw prompt -> draft goal/conventions -> 3+ critics -> reconciled draft -> optional manage-agent-goal write
deep-interview-goal-context
base goal gate -> one-question interview -> drafts -> evidence review -> audit -> user confirm -> final review -> four .agent_goal files
manage-goal-architecture
repo map -> component responsibilities -> goal relevance -> goal_architecture.md
manage-goal-theory
technical evidence -> mechanisms/assumptions/tradeoffs/validation logic -> goal_theory.md
manage-agent-authority
context -> operation selection -> authority template -> safety validation -> authority summary/file
manage-schema-contracts
goal schema contract + source/interfaces -> contract surfaces -> versions/causal map -> .schema/.contract updates
manage-external-advice
raw advice -> normalize active advice -> in-place workflow revisions without GT upgrade -> audit stale/dead/degenerate claims -> apply/defer/reject -> adv-* links
derive-improvement-task
context + agents + gates -> respect allowed dispositions, verifier debt, long-run pending state, scenario/argv/blocker/stochastic repair, count-key hygiene, goal-axis completeness, residual cost, global invariant keys -> one next task/task_pack/terminal blocker -> archive old task -> write task.md -> index
task-doctor
explicit doctor instruction -> read rules/task/advice -> archive old task -> rewrite task/task_pack while preserving verifier/hook/axis/cost/global residuals -> reconcile schema/index/issue -> optional commit
optimize-task-slice
blockers/candidates/loop signals -> classify next slice including verifier_completion, hook_supply, axis_supply, scenario_supply, command_provenance_repair, blocker_contract_repair, cost_disproportionate_residual -> advisory packet for derive
profile-cycle-efficiency
ledger/logs/validation/issues -> detect repeated low-value cycles/sprawl and supply residual value-per-cycle-cost denominator -> recommended action for derive/report
task-md-agent-governance
task.md -> repo map -> worker implementation -> integration -> multi-agent audit -> task_miss report
inspect-repo-with-agents
repo map -> 3-6 read-only perspectives -> verify severe claims -> findings/coverage/gaps
inspect-oom-risk
repo/config/scale hints -> memory growth tracing -> severity findings -> mitigation
find-local-python-envs
imports/manifests/env inventory -> rank environments -> exact run commands
install-deps-with-agent
requirement -> find-local-python-envs -> cache check -> one install agent only if needed -> verification
plan-validation-scope
changed surfaces -> current_only/affected_chain/full_chain -> validation manifest
normalize-acceptance-and-demo
task context -> acceptance/non-goals/demo/validation packet -> envelope/verifier/scenario/hook contract and residual cost fields -> governance/validation scope
build-validation-set-with-agents
task/evidence -> plan/build/refresh/consume/block -> .validation assets and result packet
run-task-code-and-log
requested command -> execute/profile scope -> full body-free command_argv or command_provenance_missing -> safe_failure_autopsy if needed -> .agent_log and run evidence, including long_run_launch handoff when authorized
monitor-running-execution
running run evidence -> heartbeat/status/artifact check -> step=run long_run_monitor event -> running/completed_pending_validation/stale/missing_details/not_running without success promotion
review-cycle-output-quality
output artifacts -> one read-only qualitative reviewer -> quality/output-delta/no-overclaim/goal-axis packet
audit-cycle-loopback
run/review/output-delta/failure-autopsy/adapter -> anti_loop_provider evaluator -> 3-state gates, failure surface stage, root-cause ledger, target_required_verifier, scenario/argv/blocker/stochastic findings, count-key hygiene, goal-axis completeness, residual cost, global invariant high-water -> derive constraints
validate-task-completion
evidence bundle -> completion gates -> required verifier/hook pass + observed goal axes + scenario coverage + command provenance + blocker actionability + stochastic feasibility + long-run pending check + count-key hygiene + residual cost ratio + structure global effect -> validation_verdict + progress_verdict -> validation report
manage-evidence-cache
fingerprints -> reuse/fresh_required/stale/unsafe_to_reuse -> owning validator decides
manage-task-state-index
artifacts -> append-only IDs/links/audit -> .task/index.jsonl + index.md
record-agent-work-log
factual notes -> normalized log -> .agent_log entry -> optional task-state link
record-visible-increment
before/after evidence -> visible delta artifact -> not validation evidence
manage-implementation-issues
task/validation/blockers -> GitHub or .issue tracking -> issue lifecycle links
repo-change-commit
dirty worktree + validation context -> classify/stage/commit coherent changes -> commit hash or blocker
render-cycle-dashboard
cycle ledger -> Korean dashboard with canonical tokens and blockers
anti_loop_gate_provider.py / anti_loop_provider/
legacy entrypoint -> api export bridge -> evaluator -> adapters/domain/quality -> gates/root_cause/registry -> anti_loop_progress_gate packet
detect_progress_loop.py / progress_loop_detection/
legacy entrypoint -> CLI -> ProgressLoopAnalyzer -> evidence/fingerprints/normalizers -> output/input/validator gates -> terminal quiescence/escalation -> loop-breaker packet
safe_failure_autopsy.py
logs/adapters -> execution_stage_ladder + post_failure_diagnostics + gate_selfcheck -> failure class, next_failure_stage, safe scalar diagnostics