openclaw/reporting-governance-plugin
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: require concrete evidence for progress claims

Browse Source
This commit is contained in:
Eve
2026-04-23 15:22:02 +08:00
parent 17449fa519
commit 180619cf87
3 changed files with 59 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
scripts/test_force_recall_long_task_preflight.mjs
View File

@@ -66,11 +66,12 @@ async function main() {
'reason=silent long-task cannot continue without externalized checkpoint path',
'reason=claimed execution requires evidence of a concrete next action',
'reason=owner decision flow must end in button-path, not plain text',
'reason=claimed progression without concrete execution evidence is forbidden',
'requiredEvidence=executionEvidence',
'reason=claimed progression without concrete progress evidence is forbidden',
'requiredEvidence=progressEvidence',
'requiredValue=sessionKey, runId, modified_files, verification result, or equivalent concrete progress evidence',
'HARD_GATE: Block any plain-text handoff or silent-continuation claim when externalized checkpoint evidence is missing.',
'HARD_GATE: Block any reply path that says you already moved into the next task or are advancing the next step without concrete execution evidence.',
'HARD_GATE: Do not say you are already on the next task, already dispatched follow-up work, or already progressing in background unless you can point to actual tool execution, file changes, emitted messages, or checkpoint records.',
'HARD_GATE: Block any reply path that says you already moved into the next task or are advancing the next step without concrete progress evidence.',
'HARD_GATE: Do not say you are already on the next task, already dispatched follow-up work, or already progressing in background unless you can point to a sessionKey, runId, modified_files record, verification result, actual tool execution, file changes, emitted messages, or checkpoint records.',
'ENFORCEMENT: Forbidden path: plain-text handoff that pretends the long task is already continuing without an externalized checkpoint.',
'ENFORCEMENT: Forbidden path: stating you have already entered the next task/step when the record only contains planning language and no concrete execution evidence.',
];
@@ -98,7 +99,7 @@ async function main() {
'[LONG_TASK_GATE_LOCK]',
'gateStatus=degraded',
'gateRequired=unknown',
'HARD_GATE: Evaluator unavailable is not permission to claim silent continuation or next-task progression without verifiable evidence.',
'HARD_GATE: Evaluator unavailable is not permission to claim silent continuation or next-task progression without verifiable progress evidence.',
'HARD_GATE: Fall back to a non-silent, evidence-preserving follow-up if you cannot prove checkpoint state or concrete execution.',
];