Tighten agent token brief stop gate

53bff77→main

Sessions

1051aa7485b2 View transcript

Changes

2

• cmd/entire/cli

• Msession_tokens.go +1/-1

• Msessions_test.go +5/-1

527 unmodified lines

case hasTokenRecommendation(report, "api-call-amplification"):
    return agentBriefCostProxyBatchAction, true
case hasTokenRecommendation(report, "context-replay-hotspot"):
    return "Use at most 2 focused reads after summarizing known findings, then answer. Avoid broad grep, broad diffs, and broad tests.", true
    return "Use at most 2 focused reads only if a named file or test can change the answer; otherwise answer now. Avoid broad grep, broad diffs, and broad tests.", true
case hasTokenRecommendation(report, "subagent-heavy"):
    return "Do not launch broad subagents. Use one narrowly scoped check with a concrete expected output.", true
case hasTokenRecommendation(report, "high-context-pressure"):

Mcmd/entire/cli/session_tokens.go +1/-1

1327 unmodified lines

"Token usage: 6213.6k total; 97.4% cache/context replay; 70 API calls."
"Next best action:"
"Use at most 3 batched reads before answering."
"Continue only if a named file or test can change the verdict; otherwise answer now."
"Avoid broad grep, broad diffs, broad tests, and repeated token diagnostics; keep the answer tight."
"Signals:"
"- Cache/context replay dominates token volume."

out := stdout.String() checks := []string{ "Token usage: 637.7k total; 95.5% cache/context replay; 3 API calls.", "Use at most 2 focused reads after summarizing known findings, then answer.", "Use at most 2 focused reads only if a named file or test can change the answer; otherwise answer now.", "Avoid broad grep, broad diffs, and broad tests.", "- Cache/context replay dominates token volume.", }

"Session: test-cost-proxy-brief" "Use at most 3 batched reads" "Avoid broad grep, broad diffs, broad tests" "otherwise answer now" "keep the answer tight" "- Cache write/new context pressure is elevated." "- Output pressure is elevated."

"Checkpoint: c05e500cafe0" "Use at most 3 batched reads" "Avoid broad grep, broad diffs, broad tests" "otherwise answer now" "keep the answer tight" "- Cache write/new context pressure is elevated." "- Output pressure is elevated."

"Token usage: 6213.6k total; 97.4% cache/context replay; 70 API calls." "Next best action:" "Use at most 3 batched reads before answering." "Continue only if a named file or test can change the verdict; otherwise answer now." "Avoid broad grep, broad diffs, broad tests, and repeated token diagnostics; keep the answer tight." "Signals:" "- Cache/context replay dominates token volume."