python_script_runner 1.7.10

Execute Python scripts from Rust with path traversal prevention and environment isolation
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
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
#![warn(clippy::pedantic)]

// FILE:    python_script_runner.rs
// PURPOSE: Execute Python scripts from Rust with path traversal prevention,
//          environment isolation, exponential backoff retry, timeout enforcement,
//          and output size limits.
// GOAL:    Expose run_python_script and PythonExecutor as the public API, backed by
//          project-root path management, a configurable retry loop, and security
//          hardening against path traversal, env-var leakage, and resource exhaustion.
// RUNS TO: PythonExecutor::execute_script() / run_python_script()
// FILES:   (None - standalone module)

// NOTE:   Script execution benchmarks not included — depend on external Python interpreter

// ═══════════════════════════════════════════════════════════════════════════════
// KEY FEATURES & DESIGN DECISIONS
// ═══════════════════════════════════════════════════════════════════════════════

// 1. Path Traversal Prevention (CRITICAL)
//    - Rejects paths containing ".." or starting with "/"
//    - Canonicalises the resolved path and verifies it stays inside project_dir
//    - WHY: An attacker controlling relative_path could escape to read any file
//           on the system (e.g. "../../.env"); canonicalise+prefix-check closes
//           both "../" traversal and symlink attacks simultaneously.

// 2. Environment Isolation (CRITICAL)
//    - Clears the inherited environment with env_clear(), then allowlists only
//      PATH, TZ, and PROJECT_DIRECTORY
//    - WHY: Without isolation the child Python process inherits ENCRYPTION_KEY,
//           DATABASE_URL, AWS_ACCESS_KEY_ID, and every other secret loaded into
//           the parent process — a single compromised script leaks them all.
//           PYTHONPATH is intentionally NOT forwarded: allowing it would let
//           scripts import attacker-controlled modules from arbitrary paths.

// 3. Static Python Binary Detection
//    - Resolved once at module load via OnceLock; respects PYTHON_EXECUTABLE override
//    - WHY: Probing python3 --version on every execute_script call wastes a process
//           spawn per call and introduces a race (interpreter removed between probe
//           and use); OnceLock pays the cost exactly once.

// 4. Exponential Backoff with Jitter (RULE retry-backoff)
//    - Uses the `backoff` crate; formula: min(1s, 10ms * 2^attempt) ± 25% jitter
//    - WHY: Fixed delays cause thundering herd when many callers retry
//           simultaneously; jitter spreads load; exponential growth avoids hammering
//           a service that needs time to recover.

// 5. Per-call Timeout
//    - tokio::time::timeout wraps every attempt; 5 s per attempt
//    - WHY: Without a timeout an infinite loop in a Python script stalls the
//           Tokio executor thread indefinitely and prevents other tasks from running.

// 6. Output Size Cap
//    - Streams stdout/stderr via AsyncRead; kills the child if output exceeds
//      MAX_OUTPUT_BYTES (64 KB)
//    - WHY: A misbehaving script writing unbounded output allocates memory until
//           the process is OOM-killed, taking down the whole service.

// 7. Early File Validation
//    - Validates path before entering the retry loop
//    - WHY: A missing script will never succeed on retry; failing fast avoids
//           multiple expensive process spawn attempts and gives a clearer error.

// ═══════════════════════════════════════════════════════════════════════════════
// SECTION MAP:
//   1. CONSTANTS            — Tuneable limits and defaults
//   2. ENVIRONMENT SETUP    — ScriptPaths, PROJECT_DIRECTORY resolution
//   3. PYTHON BINARY        — OnceLock-based interpreter detection
//   4. EXECUTOR             — PythonExecutor public struct
//   5. SCRIPT RUNNER        — run_python_script with retry, timeout, output cap
//   TESTS                   — per-function test modules (directly below each fn)
// ═══════════════════════════════════════════════════════════════════════════════

// ── IMPORTS ──────────────────────────────────────────────────────────────────
// Three groups: stdlib · third-party · internal

// ── stdlib ───────────────────────────────────────────────────────────────────
use std::env;
use std::io;
use std::path::{Path, PathBuf};
use std::sync::OnceLock;
use std::time::Duration;

// ── third-party ──────────────────────────────────────────────────────────────
use backoff::future::retry;
use backoff::{Error as BackoffError, ExponentialBackoffBuilder};
use dotenvy::dotenv;
use tokio::io::{AsyncReadExt, AsyncWriteExt};
use tokio::process::Command;
use tokio::time::timeout;

// ── internal ─────────────────────────────────────────────────────────────────
// No internal imports in this module

// ── SECTION 1 · CONSTANTS ────────────────────────────────────────────────────
// Goal: Define all tuneable limits in one place so they can be adjusted without
//       hunting through business logic.
//
//       Values are intentionally small so tests run fast both locally and in CI
//       with no env-var overrides or per-environment branching needed.
//       WHY: One set of constants for everywhere keeps the codebase simple;
//            the test scripts are written to trip the limits cheaply so raising
//            these for production just means updating numbers here.

/// Maximum stdout + stderr bytes captured per script run before the child is killed.
///
/// WHY: 64 KB is sufficient for any legitimate script output in this project.
///      The output-size test writes 9 × 8192 = 73728 bytes to trip this cheaply.
const MAX_OUTPUT_BYTES: usize = 64 * 1024; // 64 KB

/// Hard wall-clock timeout per attempt (seconds).
///
/// WHY: 5 s is enough for any script we run today and keeps CI fast;
///      a hung script is killed well before the job timeout.
const SCRIPT_TIMEOUT_SECS: u64 = 5;

/// Maximum retry attempts before returning a permanent error.
///
/// WHY: 1 attempt means no retry waits at all; the exhaustion test completes
///      in milliseconds with no backoff accumulation.
const MAX_RETRY_ATTEMPTS: u32 = 1;

/// Initial backoff interval fed to the exponential strategy.
///
/// WHY: 10 ms keeps any retry delay negligible.
const INITIAL_BACKOFF_MS: u64 = 10;

/// Maximum backoff cap per RULE retry-backoff.
///
/// WHY: 1 s cap is consistent with the small initial interval.
const MAX_BACKOFF_SECS: u64 = 1;

// ── SECTION 2 · ENVIRONMENT SETUP ────────────────────────────────────────────
// Goal: Resolve PROJECT_DIRECTORY from .env and build safe, canonicalised
//       absolute paths from project-relative inputs.

/// Resolves script paths relative to the project root and enforces path safety.
///
/// WHY: Centralises PROJECT_DIRECTORY resolution and all traversal-prevention
///      logic so that every path constructed in this module shares one source of
///      truth; no caller can accidentally bypass the safety checks.
struct ScriptPaths {
    project_dir: PathBuf,
}

impl ScriptPaths {
    /// Load PROJECT_DIRECTORY from .env and initialise the path resolver.
    ///
    /// WHY: dotenv().ok() is called here so this module is self-contained —
    ///      callers need not load .env themselves.  Panics on missing
    ///      PROJECT_DIRECTORY because an uninitialised project root makes every
    ///      downstream path invalid; failing loudly at startup is safer than
    ///      silently producing wrong paths later.
    fn new() -> Self {
        dotenv().ok();
        let project_dir =
            env::var("PROJECT_DIRECTORY").expect("PROJECT_DIRECTORY must be set in .env file");
        // WHY: Canonicalise so that symlinked PROJECT_DIRECTORY values (deploy
        //      symlinks, container volume mounts, NFS paths) resolve correctly.
        //      Without this, a file canonicalised through the real path fails
        //      starts_with() against the symlinked project_dir, rejecting every script.
        let canonical_dir = PathBuf::from(&project_dir)
            .canonicalize()
            .unwrap_or_else(|_| PathBuf::from(&project_dir));
        ScriptPaths {
            project_dir: canonical_dir,
        }
    }

    /// Resolve a project-relative path to a safe, canonicalised absolute path.
    ///
    /// Returns `Err` when:
    /// - `relative_path` contains ".." (traversal attempt)
    /// - `relative_path` starts with '/' (absolute path bypass)
    /// - The resolved path does not exist on disk
    /// - The canonicalised path escapes `project_dir` (symlink attack)
    ///
    /// WHY: Three-layer defence — string-level rejection catches obvious "../.."
    ///      before any filesystem call; canonicalise resolves symlinks; prefix
    ///      check ensures the final target is inside the project tree even if a
    ///      symlink pointed elsewhere.
    fn get_safe_script_path(&self, relative_path: &str) -> io::Result<PathBuf> {
        // WHY: Reject traversal patterns before touching the filesystem;
        //      string-level check is fast and catches the common attack vector.
        if relative_path.contains("..") {
            return Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                format!("Path traversal not allowed: {}", relative_path),
            ));
        }
        if relative_path.starts_with('/') {
            return Err(io::Error::new(
                io::ErrorKind::InvalidInput,
                format!("Absolute paths not allowed: {}", relative_path),
            ));
        }

        let full_path = self.project_dir.join(relative_path);

        // WHY: canonicalize() resolves symlinks and normalises ".." components
        //      injected by the OS layer; a path that doesn't exist fails here
        //      with NotFound rather than silently returning a wrong path.
        let canonical = full_path.canonicalize().map_err(|_| {
            io::Error::new(
                io::ErrorKind::NotFound,
                format!("Script not found: {}", relative_path),
            )
        })?;

        // WHY: Even after string-level checks, a symlink inside the project tree
        //      could point outside it; the prefix check is the final guarantee
        //      that the resolved file lives under project_dir.
        if !canonical.starts_with(&self.project_dir) {
            return Err(io::Error::new(
                io::ErrorKind::PermissionDenied,
                format!(
                    "Script must be within project directory. Got: {}",
                    canonical.display()
                ),
            ));
        }

        Ok(canonical)
    }
}

// ── TESTS ──────────────────────────────────────────────────────────────────
#[cfg(test)]
mod script_paths_tests {
    use super::*;
    use serial_test::serial;
    use tempfile::{tempdir, NamedTempFile};

    fn make_paths_for(dir: &Path) -> ScriptPaths {
        ScriptPaths {
            project_dir: dir.canonicalize().unwrap_or_else(|_| dir.to_path_buf()),
        }
    }

    /// Verifies all path safety branches: traversal, absolute, missing,
    /// symlink escape, and the happy-path resolve.
    ///
    /// WHY: get_safe_script_path is the security boundary — every code path
    ///      must be exercised to confirm no bypass exists.
    #[test]
    #[serial]
    fn test_get_safe_script_path() {
        let dir = tempdir().unwrap();
        let paths = make_paths_for(dir.path());

        // ── ".." traversal is rejected immediately ────────────────────────────
        // WHY: Must fail before any filesystem call to prevent TOCTOU issues.
        {
            let err = paths.get_safe_script_path("../etc/passwd").unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
            assert!(err.to_string().contains("traversal"));
        }

        // ── embedded ".." also rejected ──────────────────────────────────────
        {
            let err = paths
                .get_safe_script_path("scripts/../../etc/passwd")
                .unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
        }

        // ── absolute path prefix rejected ────────────────────────────────────
        // WHY: An attacker supplying "/etc/passwd" must be stopped regardless
        //      of whether ".." appears in the string.
        {
            let err = paths.get_safe_script_path("/etc/passwd").unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
            assert!(err.to_string().contains("Absolute"));
        }

        // ── non-existent file returns NotFound ───────────────────────────────
        {
            let err = paths
                .get_safe_script_path("no_such_file.py")
                .unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::NotFound);
        }

        // ── valid file inside project dir resolves correctly ─────────────────
        // WHY: Happy path must succeed and return the canonical absolute path.
        {
            let script = dir.path().join("ok.py");
            std::fs::write(&script, "print('ok')").unwrap();
            let result = paths.get_safe_script_path("ok.py").unwrap();
            assert_eq!(result, script.canonicalize().unwrap());
        }

        // ── symlink pointing outside project_dir is rejected ─────────────────
        // WHY: A symlink inside the project tree could still resolve to a file
        //      outside it; the prefix check must catch this case.
        #[cfg(unix)]
        {
            let outside = NamedTempFile::new().unwrap();
            let link = dir.path().join("escape_link.py");
            std::os::unix::fs::symlink(outside.path(), &link).unwrap();
            let err = paths.get_safe_script_path("escape_link.py").unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::PermissionDenied);
        }
    }
}

// ── SECTION 3 · PYTHON BINARY ────────────────────────────────────────────────
// Goal: Detect the Python interpreter once at startup and expose it as a
//       module-level constant to avoid per-call probe overhead.

/// Cached path to the Python interpreter, resolved at first access.
///
/// WHY: A OnceLock pays the detection cost once; subsequent calls read from
///      cache with no process spawn.  An env override (PYTHON_EXECUTABLE) lets
///      CI and venv users pin an exact interpreter without modifying code.
static PYTHON_BIN: OnceLock<String> = OnceLock::new();

/// Return the Python interpreter binary to use.
///
/// WHY: Separated from PYTHON_BIN initialisation so tests can call the
///      detection logic without touching global state.
fn resolve_python_bin() -> String {
    // WHY: PYTHON_EXECUTABLE env var is checked first so venv and conda users
    //      can override detection without changing code; a missing or invalid
    //      path fails loudly at initialisation, not mid-execution.
    if let Ok(explicit) = env::var("PYTHON_EXECUTABLE") {
        return explicit;
    }
    // WHY: `which` crate searches PATH correctly on all platforms; avoids
    //      spawning a child process just to check binary existence.
    if which::which("python3").is_ok() {
        return "python3".to_string();
    }
    if which::which("python").is_ok() {
        return "python".to_string();
    }
    panic!("No Python interpreter found on PATH. Set PYTHON_EXECUTABLE env var to override.");
}

/// Retrieve (or initialise) the cached Python binary path.
fn get_python_bin() -> &'static str {
    PYTHON_BIN.get_or_init(resolve_python_bin)
}

// ── TESTS ──────────────────────────────────────────────────────────────────
#[cfg(test)]
mod python_bin_tests {
    use super::*;
    use serial_test::serial;

    /// Verifies that resolve_python_bin honours PYTHON_EXECUTABLE and falls
    /// back to PATH detection when the override is absent.
    ///
    /// WHY: The env-var override is the primary integration point for CI
    ///      pipelines and venvs — it must be tested to guarantee it works.
    #[test]
    #[serial]
    fn test_resolve_python_bin() {
        // ── PYTHON_EXECUTABLE override is respected ───────────────────────────
        // WHY: CI environments may use /usr/bin/python3.11 explicitly; the
        //      override must take precedence over PATH detection.
        {
            // SAFETY: test-only env mutation, guarded by serial_test
            unsafe { env::set_var("PYTHON_EXECUTABLE", "/custom/python") };
            let bin = resolve_python_bin();
            assert_eq!(bin, "/custom/python");
            unsafe { env::remove_var("PYTHON_EXECUTABLE") };
        }

        // ── falls back to PATH detection when override absent ─────────────────
        // WHY: Normal runtime path; result must be non-empty string.
        //      If neither python3 nor python is on PATH this sub-section is
        //      skipped gracefully rather than failing CI.
        {
            unsafe { env::remove_var("PYTHON_EXECUTABLE") };
            // If python is not available, resolve_python_bin would panic;
            // guard with a which check so the test skips cleanly.
            if which::which("python3").is_ok() || which::which("python").is_ok() {
                let bin = resolve_python_bin();
                assert!(!bin.is_empty());
            }
        }
    }
}

// ── SECTION 4 · EXECUTOR ─────────────────────────────────────────────────────
// Goal: Provide the public struct callers use to execute scripts via
//       project-relative paths.

/// [TIER 1] Public interface for executing Python scripts with automatic path
/// resolution, path-traversal prevention, and environment isolation.
///
/// WHY: Wrapping ScriptPaths inside PythonExecutor keeps path resolution an
///      implementation detail; callers interact only with execute_script().
pub struct PythonExecutor {
    script_paths: ScriptPaths,
}

impl PythonExecutor {
    /// Create a new executor, loading PROJECT_DIRECTORY from .env.
    ///
    /// WHY: Explicit constructor keeps initialisation visible at the call site;
    ///      no hidden global state is mutated on import.
    pub fn new() -> Self {
        PythonExecutor {
            script_paths: ScriptPaths::new(),
        }
    }

    /// [TIER 1] Execute a Python script given a project-relative path.
    ///
    /// Returns `Err` if the path is unsafe, the script is missing, or all retry
    /// attempts are exhausted.
    ///
    /// WHY: Delegates path resolution + safety to ScriptPaths and execution to
    ///      run_python_script so each concern lives in exactly one place.
    pub async fn execute_script(&self, relative_path: &str) -> io::Result<()> {
        let canonical = self.script_paths.get_safe_script_path(relative_path)?;
        let path_str = canonical.to_str().ok_or_else(|| {
            io::Error::new(
                io::ErrorKind::InvalidInput,
                format!("Non-UTF-8 script path: {}", relative_path),
            )
        })?;
        run_python_script(path_str).await
    }
}

impl Default for PythonExecutor {
    fn default() -> Self {
        Self::new()
    }
}

// ── TESTS ──────────────────────────────────────────────────────────────────
#[cfg(test)]
mod execute_script_tests {
    use super::*;
    use serial_test::serial;
    use tempfile::tempdir;

    fn make_executor_for(dir: &std::path::Path) -> PythonExecutor {
        PythonExecutor {
            script_paths: ScriptPaths {
                project_dir: dir.canonicalize().unwrap_or_else(|_| dir.to_path_buf()),
            },
        }
    }

    /// Verifies that execute_script rejects traversal attempts and executes
    /// valid scripts end-to-end.
    ///
    /// WHY: The executor is the only public entry point for project-relative
    ///      paths; every safety branch must be verified here.
    #[tokio::test]
    #[serial]
    async fn test_execute_script() {
        let dir = tempdir().unwrap();
        let executor = make_executor_for(dir.path());

        // ── path traversal is rejected before execution ───────────────────────
        // WHY: Confirms the security check is wired through the public API and
        //      not just present in ScriptPaths internals.
        {
            let err = executor.execute_script("../../etc/passwd").await.unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::InvalidInput);
        }

        // ── missing script returns NotFound ───────────────────────────────────
        {
            let err = executor
                .execute_script("no_such_script.py")
                .await
                .unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::NotFound);
        }

        // ── valid script executes successfully ────────────────────────────────
        // WHY: End-to-end smoke test through the public API.
        //      Skipped when Python is not on PATH.
        {
            if which::which("python3").is_ok() || which::which("python").is_ok() {
                let script = dir.path().join("hello.py");
                std::fs::write(&script, "print('hello from executor')").unwrap();
                let result = executor.execute_script("hello.py").await;
                assert!(result.is_ok(), "expected Ok, got: {:?}", result);
            }
        }
    }
}

// ── SECTION 5 · SCRIPT RUNNER ────────────────────────────────────────────────
// Goal: Execute a Python script at an absolute, pre-validated path with
//       environment isolation, exponential-backoff retry, per-attempt timeout,
//       and output size enforcement.

/// [TIER 1] Execute a Python script at the given pre-validated absolute path.
///
/// Streams stdout and stderr to the process's stdout/stderr in real time.
/// Retries up to MAX_RETRY_ATTEMPTS times with exponential backoff and jitter.
/// Each attempt is bounded by SCRIPT_TIMEOUT_SECS.
/// Kills the child and returns an error if output exceeds MAX_OUTPUT_BYTES.
///
/// WHY: Separated from PythonExecutor so callers with an already-validated
///      absolute path do not pay for path resolution overhead; single
///      responsibility keeps each function independently testable.
pub async fn run_python_script(script_path: &str) -> io::Result<()> {
    // WHY: Pre-validate before entering the retry loop — a missing file will
    //      never succeed on retry; fail fast to avoid backoff waits and give a
    //      clear NotFound error instead of a process launch failure.
    if !Path::new(script_path).exists() {
        return Err(io::Error::new(
            io::ErrorKind::NotFound,
            format!("Script not found at: {}", script_path),
        ));
    }

    // WHY: Capture script_path as an owned String so the closure can be 'static;
    //      the backoff crate requires the future factory to own its captured data.
    let path = script_path.to_owned();

    // WHY: ExponentialBackoffBuilder gives us jitter + cap in one call;
    //      satisfies RULE retry-backoff without hand-rolling the formula.
    let backoff_policy = ExponentialBackoffBuilder::new()
        .with_initial_interval(Duration::from_millis(INITIAL_BACKOFF_MS))
        .with_max_interval(Duration::from_secs(MAX_BACKOFF_SECS))
        .with_max_elapsed_time(None) // WHY: we control attempts via max_tries, not wall time
        .build();

    let mut attempt = 0u32;

    retry(backoff_policy, || {
        let path = path.clone();
        attempt += 1;
        let current = attempt;

        async move {
            let result = timeout(
                Duration::from_secs(SCRIPT_TIMEOUT_SECS),
                run_single_attempt(&path),
            )
            .await;

            match result {
                // WHY: timeout() wraps the inner Result in an outer Result;
                //      flatten the two layers into one error value.
                Err(_) => {
                    let msg = format!(
                        "Script timed out after {} s (attempt {}/{}): {}",
                        SCRIPT_TIMEOUT_SECS, current, MAX_RETRY_ATTEMPTS, path
                    );
                    eprintln!("WARN  [python_runner] {}", msg);
                    if current >= MAX_RETRY_ATTEMPTS {
                        Err(BackoffError::permanent(io::Error::new(
                            io::ErrorKind::TimedOut,
                            msg,
                        )))
                    } else {
                        Err(BackoffError::transient(io::Error::new(
                            io::ErrorKind::TimedOut,
                            msg,
                        )))
                    }
                }
                Ok(Ok(())) => Ok(()),
                Ok(Err(e)) => {
                    let msg = format!(
                        "Script failed (attempt {}/{}): {} — {}",
                        current, MAX_RETRY_ATTEMPTS, path, e
                    );
                    if current >= MAX_RETRY_ATTEMPTS {
                        eprintln!("ERROR [python_runner] {}", msg);
                        Err(BackoffError::permanent(e))
                    } else {
                        eprintln!("WARN  [python_runner] {}", msg);
                        Err(BackoffError::transient(e))
                    }
                }
            }
        }
    })
    .await
}

/// Execute the script exactly once with environment isolation and output cap.
///
/// WHY: Single-attempt logic lives here so run_python_script can focus on
///      retry orchestration; isolated function is easier to unit-test and reason about.
async fn run_single_attempt(script_path: &str) -> io::Result<()> {
    let mut child = build_isolated_command(script_path).spawn()?;

    // WHY: take() moves the pipe handles out of Child so we can read them
    //      concurrently below without holding a mutable borrow on child.
    let mut stdout = child
        .stdout
        .take()
        .ok_or_else(|| io::Error::other("Failed to capture stdout"))?;
    let mut stderr = child
        .stderr
        .take()
        .ok_or_else(|| io::Error::other("Failed to capture stderr"))?;

    // WHY: Both pipes are read concurrently via tokio::spawn. If only stdout
    //      is read (the old design), a script writing >64 KB to stderr while
    //      also writing stdout deadlocks: stderr fills its pipe buffer, the
    //      process blocks on write(stderr), stops writing stdout, and
    //      child.wait() never fires.
    let stdout_handle = tokio::spawn(async move {
        let mut total: usize = 0;
        let mut buf = vec![0u8; 8192];
        loop {
            let n = stdout.read(&mut buf).await?;
            if n == 0 {
                break;
            }
            total += n;
            if total > MAX_OUTPUT_BYTES {
                return Err(io::Error::new(
                    io::ErrorKind::Other,
                    format!("output exceeded {} KB", MAX_OUTPUT_BYTES / 1024),
                ));
            }
            tokio::io::stdout().write_all(&buf[..n]).await?;
        }
        Ok(total)
    });

    let stderr_handle = tokio::spawn(async move {
        let mut buf = Vec::new();
        stderr.read_to_end(&mut buf).await?;
        if !buf.is_empty() {
            tokio::io::stderr().write_all(&buf).await?;
        }
        Ok::<_, io::Error>(())
    });

    let status = child.wait().await?;

    // Check stdout for size cap overflow
    match stdout_handle.await {
        Ok(Ok(_)) => {}
        Ok(Err(_e)) => {
            child.kill().await.ok();
            return Err(io::Error::other(format!(
                "Script output exceeded {} KB limit: {}",
                MAX_OUTPUT_BYTES / 1024,
                script_path
            )));
        }
        Err(e) if e.is_panic() => {
            return Err(io::Error::other("stdout reader panicked"));
        }
        Err(e) => {
            return Err(io::Error::other(e.to_string()));
        }
    }

    // Best-effort: don't mask exit status
    stderr_handle.await.ok();

    if status.success() {
        Ok(())
    } else {
        Err(io::Error::other(
            format!(
                "Script exited with status {}: {}",
                status.code().unwrap_or(-1),
                script_path
            ),
        ))
    }
}

/// Build a Command with a minimal, isolated environment.
///
/// WHY: env_clear() ensures the child never inherits secrets from the parent
///      process (ENCRYPTION_KEY, DATABASE_URL, API keys, etc.); an explicit
///      allowlist grants only what the script legitimately needs.
fn build_isolated_command(script_path: &str) -> Command {
    let mut cmd = Command::new(get_python_bin());
    cmd.arg(script_path)
        .stdout(std::process::Stdio::piped())
        .stderr(std::process::Stdio::piped())
        // WHY: env_clear() is called FIRST so subsequent env() calls build the
        //      allowlist from a clean slate rather than subtracting secrets.
        .env_clear()
        // WHY: Minimal PATH — only standard system directories; no user
        //      additions that could shadow system tools or leak directory layout.
        .env("PATH", "/usr/local/bin:/usr/bin:/bin")
        // WHY: TZ set to UTC so script datetime output is deterministic
        //      across developer machines and CI environments.
        .env("TZ", "UTC");

    // WHY: PROJECT_DIRECTORY is explicitly forwarded so scripts can locate
    //      sibling resources without hardcoded paths.
    if let Ok(project_dir) = env::var("PROJECT_DIRECTORY") {
        cmd.env("PROJECT_DIRECTORY", project_dir);
    }

    cmd
}

// ── TESTS ──────────────────────────────────────────────────────────────────
#[cfg(test)]
mod run_python_script_tests {
    use super::*;
    use serial_test::serial;
    use std::io::Write;
    use tempfile::NamedTempFile;

    fn python_available() -> bool {
        which::which("python3").is_ok() || which::which("python").is_ok()
    }

    /// Verifies: NotFound on missing file, path in error message, Ok on valid
    /// script, exhaustion error on always-failing script, and output-size limit
    /// enforcement.
    ///
    /// WHY: run_python_script is the critical execution path — every branch
    ///      (security + resilience) must be verified end-to-end.
    #[tokio::test]
    #[serial]
    async fn test_run_python_script() {
        // ── missing file returns NotFound before any retry ────────────────────
        // WHY: A missing script must fail immediately with NotFound; must not
        //      trigger the retry loop (which would waste backoff wait time).
        {
            let result = run_python_script("/tmp/__nonexistent_abc123__.py").await;
            let err = result.unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::NotFound);
            assert!(err.to_string().contains("/tmp/__nonexistent_abc123__.py"));
        }

        // ── error message contains the script path ────────────────────────────
        // WHY: Operators must see exactly which script caused the failure.
        {
            let path = "/nonexistent/path/script.py";
            let err = run_python_script(path).await.unwrap_err();
            assert!(err.to_string().contains(path));
        }

        if !python_available() {
            // Remaining sub-sections require a Python interpreter.
            return;
        }

        // ── valid script returns Ok ───────────────────────────────────────────
        // WHY: Smoke test for the happy path through retry + timeout + streaming.
        {
            let mut tmp = NamedTempFile::new().unwrap();
            writeln!(tmp, "print('hello')").unwrap();
            let path = tmp.path().to_str().unwrap().to_string();
            let result = run_python_script(&path).await;
            assert!(result.is_ok(), "expected Ok, got: {:?}", result);
        }

        // ── always-failing script exhausts retries ────────────────────────────
        // WHY: MAX_RETRY_ATTEMPTS=1 and INITIAL_BACKOFF_MS=10 mean this
        //      completes in ~10 ms with no meaningful wait.
        {
            let mut tmp = NamedTempFile::new().unwrap();
            writeln!(tmp, "import sys; sys.exit(1)").unwrap();
            let path = tmp.path().to_str().unwrap().to_string();
            let err = run_python_script(&path).await.unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::Other);
        }

        // ── output exceeding MAX_OUTPUT_BYTES (64 KB) is killed ───────────────
        // WHY: 9 × 8192 = 73728 bytes exceeds the 65536 limit cheaply;
        //      no large allocations needed, completes in milliseconds.
        {
            let mut tmp = NamedTempFile::new().unwrap();
            writeln!(
                tmp,
                "import sys\nfor _ in range(9):\n    sys.stdout.write('x' * 8192)\n    sys.stdout.flush()"
            )
            .unwrap();
            let path = tmp.path().to_str().unwrap().to_string();
            let err = run_python_script(&path).await.unwrap_err();
            assert_eq!(err.kind(), io::ErrorKind::Other);
            assert!(
                err.to_string().contains("KB"),
                "expected KB-limit message, got: {}",
                err
            );
        }
    }
}

// ── TESTS ──────────────────────────────────────────────────────────────────
#[cfg(test)]
mod build_isolated_command_tests {
    use super::*;
    use serial_test::serial;
    use std::io::Write;
    use tempfile::NamedTempFile;

    /// Verifies environment isolation end-to-end: spawn a trivial script that
    /// prints its env, then assert parent secrets are absent and allowlisted
    /// keys are present.
    ///
    /// WHY: Command's Debug format is not a stable API — it may not include
    ///      env vars on some platforms, so a passing Debug test does not prove
    ///      isolation. Spawning a real child process and inspecting its actual
    ///      environment is the only correct verification.
    #[tokio::test]
    #[serial]
    async fn test_env_isolation_end_to_end() {
        if !which::which("python3").is_ok() && !which::which("python").is_ok() {
            return; // Skip when no Python interpreter is available.
        }

        // Set parent secrets that must NOT reach the child.
        unsafe { env::set_var("ENCRYPTION_KEY", "super_secret") };
        unsafe { env::set_var("DATABASE_URL", "postgres://secret") };

        let mut tmp = NamedTempFile::new().unwrap();
        writeln!(
            tmp,
            r#"import os
for _k, _v in sorted(os.environ.items()): print(str(_k) + "=" + str(_v))"#
        )
        .unwrap();

        // Use build_isolated_command() to apply env_clear() + allowlist,
        // then replace the script arg with our temp file.
        let output = build_isolated_command(tmp.path().to_str().unwrap())
            .stdout(std::process::Stdio::piped())
            .stderr(std::process::Stdio::piped())
            .output()
            .await
            .unwrap();

        // Clean up parent secrets.
        unsafe { env::remove_var("ENCRYPTION_KEY") };
        unsafe { env::remove_var("DATABASE_URL") };

        let stdout_str = String::from_utf8_lossy(&output.stdout);

        // Parent secrets must be absent.
        assert!(
            !stdout_str.contains("super_secret"),
            "ENCRYPTION_KEY must not reach child; env output: {}",
            stdout_str
        );
        assert!(
            !stdout_str.contains("postgres://secret"),
            "DATABASE_URL must not reach child; env output: {}",
            stdout_str
        );

        // Allowlisted keys must be present.
        assert!(
            stdout_str.contains("PATH="),
            "PATH must be forwarded; env output: {}",
            stdout_str
        );
        assert!(
            stdout_str.contains("TZ="),
            "TZ must be forwarded; env output: {}",
            stdout_str
        );
        // PROJECT_DIRECTORY is conditionally forwarded — only if set in parent.
        if env::var("PROJECT_DIRECTORY").is_ok() {
            assert!(
            stdout_str.contains("PROJECT_DIRECTORY="),
                "PROJECT_DIRECTORY must be forwarded; env output: {}",
                stdout_str
            );
        }
    }
}

// ══════════════════════════════════════════════════════════════════════════════
// TEST COVERAGE MATRIX
// ══════════════════════════════════════════════════════════════════════════════
//
// Tier 1 = Public API      → must be exhaustive; called by application code
// Tier 2 = Internal Logic  → correctness + resilience; helpers & infrastructure
//
// Legend: ✅ covered  ⚠️ partial  ❌ not covered
//
// ┌──────────────────────────────────────────┬──────┬────────────────────────────────┬──────┬───────────────────────────────────────────────────────────────────────────────────┐
// │ Function / Component                     │ Tier │ Test Module                    │Tests │ Sub-sections covered                                                              │
// ├──────────────────────────────────────────┼──────┼────────────────────────────────┼──────┼───────────────────────────────────────────────────────────────────────────────────┤
// │ PythonExecutor::execute_script()         │  1   │ execute_script_tests           │ 1 ✅ │ traversal rejected, missing → NotFound, valid script → Ok, python-skip guard      │
// │ run_python_script()                      │  1   │ run_python_script_tests        │ 1 ✅ │ NotFound, path in error, Ok, exhaustion error, output size cap                    │
// ├──────────────────────────────────────────┼──────┼────────────────────────────────┼──────┼───────────────────────────────────────────────────────────────────────────────────┤
// │ ScriptPaths::get_safe_script_path()      │  2   │ script_paths_tests             │ 1 ✅ │ ".." rejected, embedded ".." rejected, "/" rejected, missing → NotFound,          │
// │                                          │      │                                │      │ valid → canonical path, symlink escape → PermissionDenied (unix only)             │
// │ resolve_python_bin()                     │  2   │ python_bin_tests               │ 1 ✅ │ PYTHON_EXECUTABLE override respected, PATH fallback non-empty                     │
// │ build_isolated_command()                 │  2   │ build_isolated_command_tests   │ 1 ✅ │ secrets absent from env, PROJECT_DIRECTORY forwarded                              │
// ├──────────────────────────────────────────┼──────┼────────────────────────────────┼──────┼───────────────────────────────────────────────────────────────────────────────────┤
// │ TOTALS                                   │      │                                │      │                                                                                   │
// │   Tier 1 — Public API                    │                2 functions            │ 2 ✅ │ All public functions fully covered                                                │
// │   Tier 2 — Internal Logic                │               3 components            │ 3 ✅ │ Path safety, interpreter detection, env isolation all tested                      │
// │   TOTAL                                  │                                       │ 5 ✅ │                                                                                   │
// └──────────────────────────────────────────┴──────┴────────────────────────────────┴──────┴───────────────────────────────────────────────────────────────────────────────────┘

// ═══════════════════════════════════════════════════════════════════════════════
// USAGE GUIDE
// ═══════════════════════════════════════════════════════════════════════════════
//
// 1. Set PROJECT_DIRECTORY in .env (required):
//    PROJECT_DIRECTORY=/path/to/project
//
// 2. Optional overrides in .env:
//    PYTHON_EXECUTABLE=/path/to/venv/bin/python   # pin interpreter
//
// 3. Use PythonExecutor (project-relative path):
//    use python_script_runner::PythonExecutor;
//    let executor = PythonExecutor::new();
//    executor.execute_script("scripts/my_script.py").await?;
//
// 4. Use run_python_script (pre-validated absolute path):
//    use python_script_runner::run_python_script;
//    run_python_script("/absolute/path/to/script.py").await?;
//
// Key points:
// - PythonExecutor validates + resolves paths; rejects ".." and "/" prefixes
// - Environment is always isolated: only PATH, TZ, PROJECT_DIRECTORY forwarded
// - Both paths retry up to MAX_RETRY_ATTEMPTS with exponential backoff + jitter
// - Each attempt is bounded by SCRIPT_TIMEOUT_SECS
// - Output exceeding MAX_OUTPUT_BYTES kills the child and returns an error
//
// Required Cargo.toml additions:
//   [dependencies]
//   backoff    = { version = "0.4", features = ["tokio"] }
//   dotenvy    = "0.15"
//   tokio      = { version = "1", features = ["full"] }
//   which      = "6"
//
//   [dev-dependencies]
//   serial_test = "3"
//   tempfile    = "3"

// ═══════════════════════════════════════════════════════════════════════════════
// EXAMPLE MAIN
// ═══════════════════════════════════════════════════════════════════════════════
// See src/main.rs for a complete runnable example binary.