kaizen-cli 0.1.11

Distributable agent observability: real-time-tailable sessions, agile-style retros, and repo-level improvement (Cursor, Claude Code, Codex). SQLite, redact before any sync you enable.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120

// SPDX-License-Identifier: AGPL-3.0-or-later
//! H4 — High-cost tool concentration.

use crate::retro::types::{Bet, Inputs};

const TOP_SHARE: f64 = 0.25;
const MIN_EVENTS: u64 = 15;

pub fn run(inputs: &Inputs) -> Vec<Bet> {
    let total_events: u64 = inputs.aggregates.tool_event_counts.values().sum();
    if total_events < MIN_EVENTS {
        return vec![];
    }
    let mut pairs: Vec<(String, u64, i64)> = inputs
        .aggregates
        .tool_event_counts
        .iter()
        .map(|(t, c)| {
            let cost = inputs
                .aggregates
                .tool_cost_usd_e6
                .get(t)
                .copied()
                .unwrap_or(0);
            (t.clone(), *c, cost)
        })
        .collect();
    pairs.sort_by_key(|p| std::cmp::Reverse(p.1));
    let Some((tool, count, cost_e6)) = pairs.first() else {
        return vec![];
    };
    let share = (*count as f64) / (total_events as f64);
    if share < TOP_SHARE {
        return vec![];
    }
    let span_rows: Vec<_> = inputs
        .tool_spans
        .iter()
        .filter(|span| span.tool == *tool)
        .collect();
    let mut lats: Vec<u64> = span_rows
        .iter()
        .filter_map(|span| span.lead_time_ms)
        .collect();
    lats.sort_unstable();
    let p95 = if lats.is_empty() {
        None
    } else {
        Some(lats[((lats.len() - 1) * 95) / 100])
    };
    let reasoning = span_rows
        .iter()
        .map(|span| span.reasoning_tokens.unwrap_or(0) as u64)
        .sum::<u64>();
    let est_tokens =
        (*count as f64) * 200.0 + (*cost_e6 as f64 / 10_000.0) + reasoning as f64 * 0.5;
    let id = format!("H4:{tool}");
    vec![Bet {
        id,
        heuristic_id: "H4".into(),
        title: format!("Tool `{}` dominates agent traffic", tool),
        hypothesis: format!(
            "`{}` accounts for {:.0}% of tool calls in the window — tighten rules or add cheaper shortcuts.",
            tool,
            share * 100.0
        ),
        expected_tokens_saved_per_week: est_tokens,
        effort_minutes: 30,
        evidence: vec![
            format!("Calls: {} / {} total.", count, total_events),
            format!("Attributed cost (micro-USD sum): {}", cost_e6),
            format!(
                "Tool spans: {} · p95 lead time: {} · reasoning tokens: {}",
                span_rows.len(),
                p95.map(|v| format!("{v}ms")).unwrap_or_else(|| "-".into()),
                reasoning
            ),
        ],
        apply_step: format!(
            "Review read/search patterns involving `{}`; add project-specific rules or smaller entrypoint files.",
            tool
        ),
        evidence_recency_ms: inputs.window_end_ms,
    }]
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::retro::types::RetroAggregates;
    use std::collections::{HashMap, HashSet};

    #[test]
    fn flags_dominant_tool() {
        let mut agg = RetroAggregates::default();
        agg.tool_event_counts.insert("read_file".into(), 20);
        agg.tool_event_counts.insert("grep".into(), 3);
        let inputs = Inputs {
            window_start_ms: 0,
            window_end_ms: 1,
            events: vec![],
            files_touched: vec![],
            skills_used: vec![],
            tool_spans: vec![],
            skills_used_recent_slugs: HashSet::new(),
            usage_lookback_ms: 0,
            skill_files_on_disk: vec![],
            rule_files_on_disk: vec![],
            rules_used_recent_slugs: HashSet::new(),
            file_facts: HashMap::new(),
            eval_scores: vec![],
            aggregates: agg,
            prompt_fingerprints: vec![],
            feedback: vec![],
        };
        let bets = run(&inputs);
        assert_eq!(bets.len(), 1);
        assert!(bets[0].title.contains("read_file"));
    }
}