TaskJudge

Solution Trace 수집·검증 · taskjudge.com
tj_20260820T111907_b416c82a-forgedUnverified Generic CLI Agent · poc/negative-tests · 제출자 익명
서버 판정
unverified
grade-policy-2026-08-16
클라이언트 주장
certified
서명 검증
실패
서명 불일치 (manifest 변조 의심)
체인 연결
정합
이벤트 14건
토큰 (in / out / cache)
0 / 0 / 0
· source=
비용
미산출 ()
가격표
실행 시간
0.30s
exit=0
CPU / 최대 메모리
0ms / 3MB
프로세스 2개
capability
cost_usageunavailable
file_traceos
tool_traceos
token_usagenative
prompt_tracepartial
network_traceos
등급 판정 항목
file_trace충족
tool_trace미충족
chain_sealed미충족
prompt_trace미충족
usage_verified미충족
runner_boundary충족
process_evidence충족
no_unexplained_gap충족
workspace_evidence충족
capability downgrade
token_usageAgent native telemetry 없음, Local Gateway 미사용
cost_usagetoken_usage 부재로 비용 산출 불가
암호학적 기록
trace_root_hashsha256:267073aa23e91044632203330c12aa91bb607cb9810e0fbc49c8f87090c75dfe
artifact_root_hashsha256:4f53cda18c2baa0c0354bb5f9a3ecbe5ed12ab4d8e11ba873c2f11161202b945
manifest_digestsha256:577dea91474ce8375bcc62fd9406b6e79025076097ca649215581ca8936beb55
runner public key-----BEGIN PUBLIC KEY----- MCowBQYDK2VwAyEAXr2INUktoFQ+wuP9rTwJkPZSWv9Cj2zTBwJOQTEYon0= -----END PUBLIC KEY-----…
관측 정밀도와 알려진 공백
process 50ms · network 200ms · filesystem 2000ms
타임라인 (원문 미포함)
원본
서버가 받은 제출 payload 전부. 여기 없는 것은 서버에 없다.
manifest
argv: 1 개
./agent.sh
task: poc/negative-tests
agent: generic
usage: {}
checks: 9 항목
file_trace: true
tool_trace: false
chain_sealed: true
prompt_trace: false
usage_verified: false
runner_boundary: true
process_evidence: true
no_unexplained_gap: true
workspace_evidence: true
runner: taskjudge-poc/0.1.0
network: 2 항목
hosts: []
endpoint_count: 0
wall_ms: 304
exit_code: 0
resources: 10 항목
cpu_ms: 0
source: wait4/rusage
cpu_sys_ms: 2
max_rss_kb: 3072
cpu_user_ms: 0
io_read_bytes: 0
process_count: 2
io_write_bytes: 4096
voluntary_ctx_switches: 4
involuntary_ctx_switches: 2
sealed_at: 2026-08-20T02:19:07.690Z
signature: 3 항목
alg: ed25519
sig_b64: V3NBsu4SqQpw+k+nPwOL7hBHalW6J6VTwk7gfe1jsPsohndImWoxI6LjdBRkuJqMhwGd7rS998XqcoRe5DboBg==
public_key_pem: -----BEGIN PUBLIC KEY----- MCowBQYDK2VwAyEAXr2INUktoFQ+wuP9rTwJkPZSWv9Cj2zTBwJOQTEYon0= -----END PUBLIC KEY-----
workspace: /data/ais/bot/catcheri/workspace/taskjudge/poc/verify-workspace
collection: 7 항목
pty: true
network: os
process: true
filesystem: true
local_gateway: false
tls_intercept: false
native_telemetry: false
downgrades: 2 개
reason: Agent native telemetry 없음, Local Gateway 미사용
capability: token_usage
reason: token_usage 부재로 비용 산출 불가
capability: cost_usage
session_id: tj_20260820T111907_b416c82a-forged
started_at: 2026-08-20T02:19:07.690Z
event_count: 14
trust_grade: certified
capabilities: 6 항목
cost_usage: unavailable
file_trace: os
tool_trace: os
token_usage: native
prompt_trace: partial
network_trace: os
event_counts: 11 항목
agent_exit: 1
file_change: 1
process_exit: 2
agent_started: 1
process_start: 2
session_start: 1
pty_agent_output: 1
resource_summary: 1
workspace_snapshot: 1
workspace_final_diff: 1
capability_negotiated: 1
agent_display: Generic CLI Agent
failed_checks: 3 개
prompt_trace
tool_trace
usage_verified
trace_version: 1.0
policy_version: grade-policy-2026-08-16
redaction_hits: 2 항목
github: 1
openai: 1
stream_digests: 3 항목
files.jsonl: 3 항목
bytes: 425
lines: 2
digest: sha256:a2dc6195b182817228576a2d4f6176b9c60d592c49cc98a322288e4fb45fd859
usage.jsonl: 3 항목
bytes: 0
lines: 0
digest: sha256:e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
events.jsonl: 3 항목
bytes: 6591
lines: 14
digest: sha256:84f5ce58d67bce44b484b8de5b4bd1e2b40d770cc32971c8f45998c1db05cfb1
trace_root_hash: sha256:267073aa23e91044632203330c12aa91bb607cb9810e0fbc49c8f87090c75dfe
artifact_root_hash: sha256:4f53cda18c2baa0c0354bb5f9a3ecbe5ed12ab4d8e11ba873c2f11161202b945
file_change_counts: 3 항목
created: 1
deleted: 0
modified: 0
capability_downgrades: 2 개
reason: Agent native telemetry 없음, Local Gateway 미사용
capability: token_usage
reason: token_usage 부재로 비용 산출 불가
capability: cost_usage
observation_precision: 4 항목
known_gaps: 3 개
샘플링 주기보다 짧게 생존한 프로세스와 소켓은 누락될 수 있다
파일은 주기 스캔 + 시작/종료 스냅샷 diff 로 관측하므로 중간 변경 순서가 합쳐질 수 있다
TLS 를 종단하지 않으므로 요청 본문과 토큰은 네트워크 계층에서 관측 불가
network_sampling_ms: 200
process_sampling_ms: 50
filesystem_sampling_ms: 2000
서버 판정 (verdict)
checks: 9 항목
file_trace: true
tool_trace: false
chain_sealed: false
prompt_trace: false
usage_verified: false
runner_boundary: true
process_evidence: true
no_unexplained_gap: true
workspace_evidence: true
trust_grade: unverified
claimed_grade: certified
failed_checks: 4 개
chain_sealed
prompt_trace
tool_trace
usage_verified
chain_problems: []
policy_version: grade-policy-2026-08-16
grade_downgraded: true
signature_reason: 서명 불일치 (manifest 변조 의심)
timeline (14건)
0.000s 세션 시작 seq 1
seq: 1
hash: sha256:53a44ba55527836fd56aae2ed0cb894cf4553216a6effdbff2828eee1493e3eb
host: 4 항목
cwd: /data/ais/bot/catcheri/workspace/taskjudge
user: rootsall
python: 3.13.2
platform: linux
t_ms: 0
type: session_start
source: runner
0.000s capability 확정 seq 2
seq: 2
hash: sha256:ad8140c340934739b23a467d81192c18e5fc114f3048ae87f866d3ef2b9fa8a0
t_ms: 0
type: capability_negotiated
source: runner
0.001s workspace 스냅샷 seq 3
seq: 3
hash: sha256:8c2c65248114ce0ac07e55577c96261d659bff00d155f1e0f972a682fd0722f4
t_ms: 1
type: workspace_snapshot
source: os
0.002s Agent 실행 seq 4
seq: 4
hash: sha256:d079cb103f165a29dad224a7858331e41706c15116fb40befdd2636b09858fa6
t_ms: 2
type: agent_started
source: runner
0.045s 프로세스 시작 seq 5
seq: 5
comm: bash
hash: sha256:3b3d7225d25d35a401a1d45dc18e15c3ed2a3243884faca3890b377ad71ae5fc
t_ms: 45
type: process_start
source: os
0.045s 프로세스 시작 seq 6
seq: 6
comm: sleep
hash: sha256:362c92f598dbab5ee432e04ef387e72942e7cbc6fddb3cb812ce508f6df8f51f
t_ms: 45
type: process_start
source: os
0.305s Agent 출력(PTY) seq 7
seq: 7
hash: sha256:e5aecd030c684ab8a4b2de39cda6632540046f465726b90673ad047289cbac0a
t_ms: 305
type: pty_agent_output
chars: 132
source: pty
0.305s Agent 종료 seq 8
seq: 8
hash: sha256:200c670298408103ce68615e34c16d3c10a89efd082a87f4cc468021982bb793
t_ms: 305
type: agent_exit
source: runner
exit_code: 0
0.321s 프로세스 종료 seq 9
seq: 9
comm: sleep
hash: sha256:92e72ac29d8038b53701a4547ba3518aea74d4e2764b34f7470308bb458231d6
t_ms: 321
type: process_exit
source: os
0.321s 프로세스 종료 seq 10
seq: 10
comm: bash
hash: sha256:e2ae546bd096ed06cc2b9fe5497eca80a00bf9220b0a24ca27b1d67561fc38d9
t_ms: 321
type: process_exit
source: os
0.606s 파일 변경 seq 11
seq: 11
hash: sha256:aef99a7106d181e5a27ab213d04a526e70a29ae82129a71b6826ffdfc2caefac
path: out.py
t_ms: 606
type: file_change
change: created
source: os
0.606s 최종 diff seq 12
seq: 12
hash: sha256:2e5f8ac2d523b7840e3db67b427b2b5f8ec19b53824b44882fa068ad1addaf44
t_ms: 606
type: workspace_final_diff
source: os
0.606s 자원 요약 seq 13
seq: 13
hash: sha256:f6748bd0087f4f01a0ec335644b9c0428655694b4cd375d2ad6cd0624516fbc7
t_ms: 606
type: resource_summary
source: os
0.606s Trace 봉인 seq 14
seq: 14
hash: sha256:267073aa23e91044632203330c12aa91bb607cb9810e0fbc49c8f87090c75dfe
t_ms: 606
type: session_sealed
source: runner
API credentials never leave this device. TaskJudge servers receive only sanitized usage metadata, approved solution content, and cryptographic trace records.