Use when a debugging thread needs to be compressed into a reusable investigation ledger. Capture the target, evidence, attempted fixes, ruled-out hypotheses, viable hypotheses, and next experiments. Good triggers include "compact this debugging session", "summarize what we've tried", and "turn this into a debugging ledger".
99
100%
Does it follow best practices?
Impact
99%
3.66xAverage score across 8 eval scenarios
Passed
No known issues
{
"context": "Tests whether the agent omits dead conversational detail (latecomers being caught up, lunch breaks, 'any updates?', departures) and preserves only investigation-relevant content.",
"type": "weighted_checklist",
"checklist": [
{
"name": "No catch-up repetition",
"description": "Output does NOT reproduce repeated catch-up explanations (e.g. 'Hi what's going on?', 'Can someone catch me up?', 'What did I miss?', the repeated re-summarizing of the situation for Jordan and Sam)",
"max_score": 15
},
{
"name": "No absence/return mentions",
"description": "Output does NOT mention team members leaving for calls, lunch, or returning — these have no bearing on investigation state",
"max_score": 12
},
{
"name": "Evidence preserved",
"description": "Output contains an '### Evidence' section that retains the key investigative facts: record count (~3.8M), UiPath/RPA automation, Cloudflare IP, logins only from UiPath range, Acme Corp half-admission",
"max_score": 15
},
{
"name": "Attempts section with labels",
"description": "Output contains an '### Attempts' section where each attempt has a worked/failed/inconclusive label",
"max_score": 10
},
{
"name": "Ruled Out section",
"description": "Output contains a '### Ruled Out' section",
"max_score": 8
},
{
"name": "Still Plausible section",
"description": "Output contains a '### Still Plausible' section with hypotheses still under consideration (e.g. rogue employee vs deliberate exfiltration, TechFlow connection)",
"max_score": 10
},
{
"name": "Next Experiments section",
"description": "Output contains a '### Next Experiments' section with 1-3 items",
"max_score": 8
},
{
"name": "Debug Target one sentence",
"description": "Output contains a '### Debug Target' section with exactly one sentence",
"max_score": 8
},
{
"name": "No repeated context",
"description": "Key facts are stated ONCE in the output — the same information is not repeated in multiple sections",
"max_score": 7
},
{
"name": "File saved",
"description": "A file named security_investigation.md exists in the workspace",
"max_score": 7
}
]
}evals
scenario-1
scenario-2
scenario-3
scenario-4
scenario-5
scenario-6
scenario-7
scenario-8
skills
compact-debug-ledger