llmlint 0.4.4

LLM-as-judge linter: enforce code-quality checks deterministic linters can't express, by driving real coding harnesses through oneharness.
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
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
//! `llmlint history`: inspect the results logged by past runs.
//!
//! With no id it lists recent runs; with an id (or `latest`) it shows that run's
//! full results — the complete per-rule verdicts, votes, per-judge breakdown,
//! violations, and rationales the terminal report omits. `--status`/`--rule`
//! drill into part of a run, `--path` prints just the JSON record's location for
//! scripting, and `--format json` emits the raw record (or a JSON array when
//! listing).
//!
//! The history *directory* is resolved with the same precedence writing uses
//! (`--dir` > `LLMLINT_HISTORY_DIR` > config `history.dir` > platform default),
//! so inspection and logging always agree on where records live. Config loading
//! is best-effort — history is readable even outside a configured project.

use std::path::{Path, PathBuf};

use serde_json::{json, Value};

use crate::cli::{HistoryArgs, OutputFormat};
use crate::domain::labels::{self, Label};
use crate::errors::{Error, Result};
use crate::io::{configfs, history};

/// Rule outcomes that `--status` accepts, matching the `outcome` field in a
/// stored record (see [`crate::domain::verdict::Outcome`]).
const VALID_STATUSES: &[&str] = &["pass", "fail", "skipped", "ignored", "not_relevant"];

pub fn run(args: HistoryArgs) -> Result<i32> {
    let cwd = match &args.cwd {
        Some(d) => d.clone(),
        None => std::env::current_dir().map_err(|e| Error::Io(e.to_string()))?,
    };

    validate_filters(&args)?;
    // Parsed through the same grammar `lint --label` records with, so a filter
    // that could never match a recorded label is a usage error, not a silent
    // empty result.
    let filter = labels::parse_flags(&args.label)?;

    let dir = resolve_dir(&args, &cwd)?;

    match &args.id {
        Some(id) => show(&dir, id, &args, &filter),
        None => list(&dir, &args, &filter),
    }
}

/// The label filter as the user spelled it (`--label k=v --label k2=v2`), for
/// naming it in a no-match error.
fn describe_filter(filter: &[Label]) -> String {
    filter
        .iter()
        .map(|l| format!("{} {l}", labels::FLAG))
        .collect::<Vec<_>>()
        .join(" ")
}

/// Resolve `id` (or `latest`) within the label filter. With no filter this is
/// exactly [`history::load`]; with one, `latest` is the newest matching run and
/// an explicit id is shown only when it matches — otherwise the history error,
/// naming the filter.
fn load_matching(dir: &Path, id: &str, filter: &[Label]) -> Result<history::Record> {
    if filter.is_empty() {
        return history::load(dir, id);
    }
    let described = describe_filter(filter);
    if id == "latest" {
        return history::all(dir)?
            .into_iter()
            .find(|r| r.labels().matches(filter))
            .ok_or_else(|| {
                Error::History(format!("no run matching {described} in {}", dir.display()))
            });
    }
    let record = history::load(dir, id)?;
    if record.labels().matches(filter) {
        Ok(record)
    } else {
        Err(Error::History(format!(
            "no run with id {id:?} matching {described} in {} \
             (run `llmlint history {described}` to list matching runs)",
            dir.display()
        )))
    }
}

/// Resolve the history directory: the `--dir` override, else the same
/// config/env/default resolution logging uses. Config load is best-effort (a
/// missing or broken config falls back to env + platform default), since reading
/// history must work outside a configured project.
fn resolve_dir(args: &HistoryArgs, cwd: &Path) -> Result<PathBuf> {
    if let Some(dir) = &args.dir {
        return Ok(dir.clone());
    }
    let config = configfs::load(&[], cwd)
        .map(|l| l.config)
        .unwrap_or_default();
    // `resolve` reads the env + config + default; `force_off` is irrelevant to
    // where records live, so pass `false`.
    history::resolve(&config, false).dir.ok_or_else(|| {
        Error::History(
            "no history directory could be determined (set history.dir, \
             LLMLINT_HISTORY_DIR, or pass --dir)"
                .to_string(),
        )
    })
}

/// `--status`/`--rule` only make sense against a single run, and each `--status`
/// must name a real outcome. Reject early with an actionable message.
fn validate_filters(args: &HistoryArgs) -> Result<()> {
    if args.id.is_none() && (!args.status.is_empty() || !args.rule.is_empty()) {
        return Err(Error::History(
            "--status/--rule filter a single run; pass a run id (or `latest`)".to_string(),
        ));
    }
    for s in &args.status {
        if !VALID_STATUSES.contains(&s.as_str()) {
            return Err(Error::History(format!(
                "unknown --status {s:?}; valid statuses: {}",
                VALID_STATUSES.join(", ")
            )));
        }
    }
    Ok(())
}

/// Show one run. `--path` prints just the record's file path; otherwise the run's
/// results, optionally narrowed by `--status`/`--rule`, as human text or JSON.
fn show(dir: &Path, id: &str, args: &HistoryArgs, filter: &[Label]) -> Result<i32> {
    let record = load_matching(dir, id, filter)?;
    if args.path {
        println!("{}", record.path.display());
        return Ok(0);
    }

    // Narrow the `rules` array in place when filters are given; everything else
    // (metadata, summary, errors) is left as recorded.
    let mut value = record.value.clone();
    if !args.status.is_empty() || !args.rule.is_empty() {
        if let Some(rules) = value.get("rules").and_then(Value::as_array) {
            let filtered = filter_rules(rules, &args.status, &args.rule);
            value["rules"] = Value::Array(filtered);
        }
    }

    match args.format {
        OutputFormat::Json => println!(
            "{}",
            serde_json::to_string_pretty(&value).map_err(|e| Error::Io(e.to_string()))?
        ),
        OutputFormat::Human => print!("{}", render_run(&record, &value)),
    }
    Ok(0)
}

/// List recent runs, newest first, keeping only those matching the label filter
/// and then capping at `--limit` (default 20) — so non-matching runs never count
/// against the limit. A filter that matches nothing is the history error naming
/// it (an unfiltered empty history stays a friendly note).
fn list(dir: &Path, args: &HistoryArgs, filter: &[Label]) -> Result<i32> {
    if args.path {
        // `--path` with no id: the directory itself, for scripting.
        println!("{}", dir.display());
        return Ok(0);
    }
    let limit = args.limit.unwrap_or(20);
    let records: Vec<history::Record> = history::all(dir)?
        .into_iter()
        .filter(|r| r.labels().matches(filter))
        .take(limit)
        .collect();
    if records.is_empty() && !filter.is_empty() {
        return Err(Error::History(format!(
            "no run matching {} in {}",
            describe_filter(filter),
            dir.display()
        )));
    }

    match args.format {
        OutputFormat::Json => {
            let arr: Vec<Value> = records.iter().map(list_entry).collect();
            println!(
                "{}",
                serde_json::to_string_pretty(&Value::Array(arr))
                    .map_err(|e| Error::Io(e.to_string()))?
            );
        }
        OutputFormat::Human => {
            if records.is_empty() {
                println!("No runs recorded in {}", dir.display());
                return Ok(0);
            }
            for r in &records {
                println!("{}", render_list_line(r));
            }
        }
    }
    Ok(0)
}

/// Keep only the rules matching every active filter: an `--status` set (any of
/// its outcomes) and/or an `--rule` set (any of its names).
fn filter_rules(rules: &[Value], statuses: &[String], names: &[String]) -> Vec<Value> {
    rules
        .iter()
        .filter(|r| {
            let status_ok = statuses.is_empty()
                || r.get("outcome")
                    .and_then(Value::as_str)
                    .is_some_and(|o| statuses.iter().any(|s| s == o));
            let name_ok = names.is_empty()
                || r.get("name")
                    .and_then(Value::as_str)
                    .is_some_and(|n| names.iter().any(|x| x == n));
            status_ok && name_ok
        })
        .cloned()
        .collect()
}

/// The compact per-run object emitted by `history --format json` when listing.
fn list_entry(r: &history::Record) -> Value {
    let g = |k: &str| r.value.get(k).cloned().unwrap_or(Value::Null);
    json!({
        "id": r.id,
        "timestamp": g("timestamp"),
        "command": g("command"),
        "exit_code": g("exit_code"),
        "summary": g("summary"),
        "cwd": g("cwd"),
        "path": r.path.display().to_string(),
        // Always present (`{}` when the run had none), so a consumer never has
        // to special-case an unlabelled run.
        "labels": r.labels(),
    })
}

/// One line in the human run listing: id, timestamp, exit code, a terse counts
/// summary, and the run's labels when it has any.
fn render_list_line(r: &history::Record) -> String {
    let s = |k: &str| r.value.get(k).and_then(Value::as_str).unwrap_or("");
    let exit = r
        .value
        .get("exit_code")
        .and_then(Value::as_i64)
        .unwrap_or(-1);
    let counts = r
        .value
        .get("summary")
        .map(counts_summary)
        .unwrap_or_default();
    let mut line = format!("{}  {}  exit {exit}  {counts}", r.id, s("timestamp"));
    let labels = r.labels();
    if !labels.is_empty() {
        line.push_str(&format!("  labels: {}", labels.render()));
    }
    line
}

/// Render one run's full results as human-readable text.
fn render_run(record: &history::Record, value: &Value) -> String {
    let id = &record.id;
    let s = |k: &str| value.get(k).and_then(Value::as_str).unwrap_or("");
    let exit = value.get("exit_code").and_then(Value::as_i64).unwrap_or(-1);
    let mut out = String::new();
    out.push_str(&format!("Run {id}  {}\n", s("timestamp")));
    out.push_str(&format!(
        "  command: {}   exit: {exit}   cwd: {}\n",
        s("command"),
        s("cwd")
    ));
    if let Some(files) = value.get("config_files").and_then(Value::as_array) {
        let joined: Vec<&str> = files.iter().filter_map(Value::as_str).collect();
        if !joined.is_empty() {
            out.push_str(&format!("  config: {}\n", joined.join(", ")));
        }
    }
    let labels = record.labels();
    if !labels.is_empty() {
        out.push_str(&format!("  labels: {}\n", labels.render()));
    }
    if let Some(summary) = value.get("summary") {
        out.push_str(&format!("  {}\n", counts_summary(summary)));
    }
    out.push('\n');

    if let Some(rules) = value.get("rules").and_then(Value::as_array) {
        if rules.is_empty() {
            out.push_str("  (no rules match)\n");
        }
        for rule in rules {
            push_rule(&mut out, rule);
        }
    }
    if let Some(errors) = value.get("errors").and_then(Value::as_array) {
        for e in errors.iter().filter_map(Value::as_str) {
            out.push_str(&format!("  ERROR {e}\n"));
        }
    }
    out
}

/// Append one rule's line(s): the status label + name, its rationale (if any),
/// each judge's breakdown (multi-judge rules), and each violation's location.
fn push_rule(out: &mut String, rule: &Value) {
    let name = rule.get("name").and_then(Value::as_str).unwrap_or("");
    let outcome = rule.get("outcome").and_then(Value::as_str).unwrap_or("");
    let label = match outcome {
        "pass" => "PASS",
        "fail" => "FAIL",
        "skipped" => "SKIP",
        "ignored" => "IGN ",
        "not_relevant" => "N/A ",
        _ => "?   ",
    };
    let votes = match (
        rule.get("votes_total").and_then(Value::as_u64),
        rule.get("votes_hold").and_then(Value::as_u64),
    ) {
        (Some(total), Some(hold)) if total > 1 => format!(" ({hold}/{total} judges held)"),
        _ => String::new(),
    };
    out.push_str(&format!("  {label} {name}{votes}\n"));
    if let Some(r) = rule.get("rationale").and_then(Value::as_str) {
        let r = r.trim();
        if !r.is_empty() {
            out.push_str(&format!("       rationale: {r}\n"));
        }
    }
    if let Some(judges) = rule.get("judges").and_then(Value::as_array) {
        for (i, j) in judges.iter().enumerate() {
            let verdict = if j.get("relevant") == Some(&Value::Bool(false)) {
                "not relevant"
            } else if j.get("holds").and_then(Value::as_bool).unwrap_or(false) {
                "held"
            } else {
                "violated"
            };
            match j.get("rationale").and_then(Value::as_str) {
                Some(r) if !r.trim().is_empty() => {
                    out.push_str(&format!("       judge {} {verdict}: {}\n", i + 1, r.trim()))
                }
                _ => out.push_str(&format!("       judge {} {verdict}\n", i + 1)),
            }
        }
    }
    if let Some(violations) = rule.get("violations").and_then(Value::as_array) {
        for v in violations {
            out.push_str(&format!("       {}\n", format_violation(v)));
        }
    }
}

/// Format a stored violation JSON as `file:line[-end]: message`, matching the
/// terminal report's spelling.
fn format_violation(v: &Value) -> String {
    let mut loc = String::new();
    if let Some(file) = v.get("file").and_then(Value::as_str) {
        loc.push_str(file);
        if let Some(line) = v.get("line").and_then(Value::as_u64) {
            loc.push_str(&format!(":{line}"));
            if let Some(end) = v.get("end_line").and_then(Value::as_u64) {
                loc.push_str(&format!("-{end}"));
            }
        }
    }
    let msg = v
        .get("message")
        .and_then(Value::as_str)
        .unwrap_or("violation");
    if loc.is_empty() {
        msg.to_string()
    } else {
        format!("{loc}: {msg}")
    }
}

/// A terse counts line from a stored `summary` object: `N rules: X passed, Y
/// failed, Z skipped[, W not relevant][, E errored]` — the same shape the
/// terminal report prints.
fn counts_summary(summary: &Value) -> String {
    let g = |k: &str| summary.get(k).and_then(Value::as_u64).unwrap_or(0);
    let mut s = format!(
        "{} rules: {} passed, {} failed, {} skipped",
        g("total"),
        g("passed"),
        g("failed"),
        g("skipped")
    );
    if g("ignored") > 0 {
        s.push_str(&format!(", {} ignored", g("ignored")));
    }
    if g("not_relevant") > 0 {
        s.push_str(&format!(", {} not relevant", g("not_relevant")));
    }
    if g("errored") > 0 {
        s.push_str(&format!(", {} errored", g("errored")));
    }
    s
}

#[cfg(test)]
mod tests {
    use super::*;

    fn record() -> Value {
        json!({
            "id": "20260704T000000Z-00001",
            "timestamp": "2026-07-04T00:00:00Z",
            "command": "lint",
            "cwd": "/proj",
            "exit_code": 1,
            "config_files": ["llmlint.yml"],
            "summary": {"total": 3, "passed": 1, "failed": 1, "skipped": 1, "not_relevant": 0, "errored": 0},
            "rules": [
                {"name": "ok_rule", "outcome": "pass", "votes_total": 1, "votes_hold": 1},
                {"name": "bad_rule", "outcome": "fail", "rationale": "raw SQL",
                 "violations": [{"file": "src/db.rs", "line": 12, "message": "inline SQL"}]},
                {"name": "no_files", "outcome": "skipped"}
            ],
            "errors": []
        })
    }

    #[test]
    fn filter_by_status_and_name() {
        let rules = record()["rules"].as_array().unwrap().clone();
        // Status filter keeps only failures.
        let fails = filter_rules(&rules, &["fail".into()], &[]);
        assert_eq!(fails.len(), 1);
        assert_eq!(fails[0]["name"], "bad_rule");
        // Name filter keeps only the named rule.
        let named = filter_rules(&rules, &[], &["ok_rule".into()]);
        assert_eq!(named.len(), 1);
        assert_eq!(named[0]["name"], "ok_rule");
        // Combined: both must match (no overlap here).
        assert!(filter_rules(&rules, &["pass".into()], &["bad_rule".into()]).is_empty());
        // No filters keeps everything.
        assert_eq!(filter_rules(&rules, &[], &[]).len(), 3);
    }

    #[test]
    fn render_run_shows_metadata_and_located_violation() {
        let text = render_run(&stored(record()), &record());
        assert!(text.contains("Run 20260704T000000Z-00001  2026-07-04T00:00:00Z"));
        assert!(text.contains("command: lint   exit: 1   cwd: /proj"));
        assert!(text.contains("config: llmlint.yml"));
        assert!(text.contains("3 rules: 1 passed, 1 failed, 1 skipped"));
        assert!(text.contains("PASS ok_rule"));
        assert!(text.contains("FAIL bad_rule"));
        assert!(text.contains("rationale: raw SQL"));
        assert!(text.contains("src/db.rs:12: inline SQL"));
        assert!(text.contains("SKIP no_files"));
        // An unlabelled run prints no labels line.
        assert!(!text.contains("labels:"));
    }

    fn stored(value: Value) -> history::Record {
        history::Record {
            id: "20260704T000000Z-00001".into(),
            path: PathBuf::from("20260704T000000Z-00001.json"),
            value,
        }
    }

    #[test]
    fn labels_render_in_both_human_views_and_every_listing_entry() {
        let mut value = record();
        value["labels"] = json!({"session": "abc", "turn": "1"});
        let labelled = stored(value.clone());
        assert!(render_run(&labelled, &value).contains("  labels: session=abc, turn=1\n"));
        assert!(render_list_line(&labelled).ends_with("  labels: session=abc, turn=1"));
        assert_eq!(
            list_entry(&labelled)["labels"],
            json!({"session": "abc", "turn": "1"})
        );
        // Absence reads as `{}` in the listing and adds nothing to the line.
        let plain = stored(record());
        assert_eq!(list_entry(&plain)["labels"], json!({}));
        assert!(!render_list_line(&plain).contains("labels"));
    }

    #[test]
    fn counts_summary_appends_optional_segments() {
        let base = json!({"total": 2, "passed": 2, "failed": 0, "skipped": 0});
        assert_eq!(
            counts_summary(&base),
            "2 rules: 2 passed, 0 failed, 0 skipped"
        );
        let extra = json!({"total": 3, "passed": 1, "failed": 0, "skipped": 0,
                           "not_relevant": 1, "errored": 1});
        let s = counts_summary(&extra);
        assert!(s.contains("1 not relevant"));
        assert!(s.contains("1 errored"));
    }

    #[test]
    fn multi_judge_breakdown_renders() {
        let rule = json!({
            "name": "voted", "outcome": "fail", "votes_total": 3, "votes_hold": 1,
            "judges": [
                {"holds": false, "rationale": "raw SQL"},
                {"holds": true, "rationale": "query layer"},
                {"relevant": false}
            ]
        });
        let mut out = String::new();
        push_rule(&mut out, &rule);
        assert!(out.contains("FAIL voted (1/3 judges held)"));
        assert!(out.contains("judge 1 violated: raw SQL"));
        assert!(out.contains("judge 2 held: query layer"));
        assert!(out.contains("judge 3 not relevant"));
    }
}