prodigy 0.4.4

Turn ad-hoc Claude sessions into reproducible development pipelines with parallel AI agents
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
## Command Types

Prodigy supports several types of commands in workflows. **Each command step must specify exactly one command type** - they are mutually exclusive.

### Quick Reference

| Command Type | Primary Use Case | Key Features |
|-------------|------------------|--------------|
| [`shell:`]#shell-commands | Execute shell commands | Output capture, conditional execution, timeouts |
| [`claude:`]#claude-commands | Run Claude AI commands | Variable interpolation, commit tracking, output declarations |
| [`analyze:`]#analyze-commands | Codebase analysis with caching | Cache control, force refresh, format options |
| [`foreach:`]#foreach-commands | Parallel iteration | Process lists in parallel, item limits, error handling |
| [`write_file:`]#write-file-commands | Create files | Format validation (JSON/YAML), directory creation, permissions |
| [`validate:`]#validation-commands | Implementation validation | Threshold checking, gap detection, multi-step validation |

!!! note "Deprecated: Goal Seek Commands"
    The `goal_seek:` command type has been removed from Prodigy. Use `validate:` with `on_incomplete` handlers for iterative refinement workflows instead.

!!! tip "Command Exclusivity"
    Each workflow step must use exactly one command type. You cannot combine `shell:` and `claude:` in the same step. Use `on_success:` or `on_failure:` to chain commands together.

### Choosing the Right Command Type

```mermaid
flowchart TD
    Start{What are you<br/>trying to do?}

    Start -->|Run shell commands| Shell[shell:]
    Start -->|Execute Claude AI| Claude[claude:]
    Start -->|Analyze codebase| Analyze{Need fresh<br/>metrics?}
    Start -->|Process multiple items| Foreach[foreach:]
    Start -->|Save data to file| WriteFile[write_file:]
    Start -->|Check completeness| Validate[validate:]

    Analyze -->|Yes, after changes| AnalyzeForce[analyze:<br/>force_refresh: true]
    Analyze -->|No, use cache| AnalyzeCache[analyze:<br/>max_cache_age: 300]

    style Shell fill:#e1f5ff
    style Claude fill:#f3e5f5
    style AnalyzeForce fill:#fff3e0
    style AnalyzeCache fill:#fff3e0
    style Foreach fill:#fce4ec
    style WriteFile fill:#f1f8e9
    style Validate fill:#ede7f6
```

**Figure**: Decision tree for selecting the appropriate command type based on your workflow needs.

### Shell Commands

Execute shell commands during workflow execution.

**Source**: src/config/command.rs:328

**Syntax**:
```yaml
- shell: "command to execute"
  timeout: 300                    # (1)!
  capture_output: true            # (2)!
  capture_format: "json"          # (3)!
  capture_streams: "both"         # (4)!
  output_file: "results.txt"      # (5)!
  on_failure:                     # (6)!
    claude: "/debug-failure"
  on_success:                     # (7)!
    shell: "echo 'Success!'"
  when: "variable == 'value'"     # (8)!
```

1. Maximum execution time in seconds before the command is terminated
2. Capture command output to a variable (boolean or variable name)
3. Format for captured output: `json`, `text`, or `lines`
4. Which streams to capture: `stdout`, `stderr`, or `both`
5. File path to redirect command output to
6. Nested command to execute if the shell command fails (non-zero exit)
7. Nested command to execute if the shell command succeeds (zero exit)
8. Conditional expression - command only runs if the condition evaluates to true

**Fields**:
- `shell` (required): The shell command to execute
- `timeout` (optional): Maximum execution time in seconds
- `capture_output` (optional): Boolean or variable name to capture output to
- `capture_format` (optional): Format for captured output (json, text, lines)
- `capture_streams` (optional): Which streams to capture (stdout, stderr, both)
- `output_file` (optional): File path to redirect output to
- `on_failure` (optional): Nested command to execute if shell command fails
- `on_success` (optional): Nested command to execute if shell command succeeds
- `when` (optional): Conditional expression - command only runs if true

**Example** (from workflows/fix-files-mapreduce.yml:36):
```yaml
- shell: "cargo check --lib 2>&1 | grep -E '(error|warning)' | head -10 || echo 'No errors'"
  capture_output: true
```

**Example with on_failure** (from workflows/coverage.yml:13):
```yaml
- shell: "just test"
  on_failure:
    claude: "/prodigy-debug-test-failure --spec ${coverage.spec} --output ${shell.output}"
```

!!! example "Common Use Cases"
    - Running build commands (`cargo build`, `npm install`)
    - Executing tests (`cargo test`, `pytest`)
    - File operations (`mkdir`, `cp`, `rm`)
    - Running analysis tools (`cargo clippy`, `eslint`)
    - Generating data files

!!! tip "Error Handling Flow"
    Shell commands support error handling via `on_failure` and `on_success` hooks. The flow looks like:

    ```mermaid
    flowchart LR
        Exec[Execute shell command] --> Check{Exit code<br/>zero?}
        Check -->|Yes| Success[Run on_success<br/>if defined]
        Check -->|No| Failure[Run on_failure<br/>if defined]
        Success --> Next[Continue workflow]
        Failure --> Next

        style Check fill:#fff3e0
        style Success fill:#e8f5e9
        style Failure fill:#ffebee
    ```

### Claude Commands

Execute Claude CLI commands via Claude Code.

**Source**: src/config/command.rs:324

**Syntax**:
```yaml
- claude: "/command-name args"
  id: "command_id"                # (1)!
  commit_required: true           # (2)!
  timeout: 600                    # (3)!
  outputs:                        # (4)!
    spec:
      file_pattern: "*.md"
  when: "condition"               # (5)!
  on_failure:                     # (6)!
    claude: "/fix-issue"
  on_success:                     # (7)!
    shell: "echo 'Done'"
```

1. Unique identifier for this command - used to reference outputs with `${id.output_name}`
2. Whether the command is expected to create git commits (validates commit after execution)
3. Maximum execution time in seconds before the command is terminated
4. Declare outputs that downstream commands can reference (e.g., files created)
5. Conditional expression - command only runs if the condition evaluates to true
6. Nested command to execute if Claude command fails or returns error
7. Nested command to execute if Claude command succeeds

**Fields**:
- `claude` (required): The Claude command string with arguments
- `id` (optional): Unique identifier for this command in the workflow
- `commit_required` (optional): Whether command is expected to create git commits (default: false)
- `timeout` (optional): Maximum execution time in seconds
- `outputs` (optional): Declare outputs that can be referenced by other commands
- `when` (optional): Conditional expression for execution
- `on_failure` (optional): Nested command to execute on failure
- `on_success` (optional): Nested command to execute on success

**Example with outputs** (from workflows/coverage.yml:3):
```yaml
- claude: "/prodigy-coverage"
  id: coverage
  commit_required: true
  outputs:
    spec:
      file_pattern: "*-coverage-improvements.md"

- claude: "/prodigy-implement-spec ${coverage.spec}"
  commit_required: true
```

**Example in MapReduce** (from workflows/fix-files-mapreduce.yml:31):
```yaml
- claude: "/analyze-and-fix-file ${item.path} --complexity ${item.complexity}"
  capture_output: true
  timeout: 300
```

!!! example "Common Use Cases"
    - Running custom Claude commands (slash commands)
    - Code analysis and generation
    - Implementing specifications from markdown
    - Debugging test failures with AI assistance
    - Automated code review and linting

### Analyze Commands

Perform codebase analysis with intelligent caching to avoid redundant computation. Analysis results include both metrics and context data that can be used by subsequent Claude commands.

**Source**: src/config/command.rs:332

**Syntax**:
```yaml
# Source: workflows/analysis-workflow.yml:19-23
- analyze:
    max_cache_age: 300        # Optional: cache validity in seconds
    force_refresh: false      # Optional: bypass cache (default: false)
    save: true                # Optional: save analysis results (default: true)
    format: "summary"         # Optional: output format (summary, detailed)
```

**Fields**:
- `max_cache_age` (optional): Maximum age of cached analysis in seconds before refresh
- `force_refresh` (optional): Force fresh analysis even if cache is valid (default: false)
- `save` (optional): Save analysis results to context directory (default: true)
- `format` (optional): Output format for analysis results (summary, detailed)

**Caching Behavior**:
- If cache exists and is younger than `max_cache_age`, use cached results
- If `force_refresh: true`, always perform fresh analysis regardless of cache
- Analysis runs automatically at workflow start unless `--skip-analysis` flag is used
- Cached results are stored per-repository for reuse across workflows

**Example with cached analysis** (from workflows/analysis-workflow.yml:19-26):
```yaml
# Source: workflows/analysis-workflow.yml:19-26
- analyze:
    max_cache_age: 300        # Use cache if less than 5 minutes old
    force_refresh: false      # Don't force if cache is fresh
    save: true
    format: "summary"

- claude: "/prodigy-code-review"
```

**Example with forced refresh** (from workflows/analysis-workflow.yml:29-34):
```yaml
# Source: workflows/analysis-workflow.yml:29-34
- analyze:
    force_refresh: true       # Always get fresh analysis for accuracy
    save: true
    format: "summary"

- claude: "/prodigy-cleanup-tech-debt"
```

**Example with short cache** (from workflows/tech-debt.yml:3-5):
```yaml
# Source: workflows/tech-debt.yml:3-5
- analyze:
    max_cache_age: 300
    save: true

- claude: "/prodigy-cleanup-tech-debt"
  id: cleanup
```

!!! example "Common Use Cases"
    - Refreshing analysis after significant code changes
    - Providing fresh metrics for tech debt cleanup
    - Updating context before documentation generation
    - Caching analysis for multiple commands to save time
    - Force refresh for critical operations requiring accuracy

!!! info "When to Use Analyze"
    Use `analyze:` when you need fresh codebase metrics or context for Claude commands. The first analyze in a workflow can use cache, but subsequent analyses after modifications should use `force_refresh: true` for accuracy.

!!! tip "Cache Strategy"
    - **Development workflows**: Use `max_cache_age: 300` (5 minutes) for fast iteration
    - **After code changes**: Use `force_refresh: true` for accurate metrics
    - **Multiple commands**: Share one analysis with appropriate `max_cache_age`
    - **CI/CD pipelines**: Use `force_refresh: true` for consistency

### Foreach Commands

Iterate over a list of items, executing commands for each item in parallel or sequentially.

**Source**: src/config/command.rs:344, src/config/command.rs:188-211

**Syntax**:
```yaml
- foreach:
    items: ["item1", "item2"]       # Static list
    # OR
    items: "shell-command"          # Command that produces items
    do:                             # Commands to execute per item
      - claude: "/process ${item}"
      - shell: "test ${item}"
    parallel: 5                     # Optional: number of parallel workers
    continue_on_error: true         # Optional: continue if item fails
    max_items: 100                  # Optional: limit number of items
```

**Input Types**:
1. **Static List**: Array of strings directly in YAML
2. **Command Output**: Shell command whose output produces item list

**Fields**:
- `items` (required): Either a static list or a command to generate items
- `do` (required): List of commands to execute for each item
- `parallel` (optional): Number of parallel workers (boolean or count)
  - `true`: Default parallel count
  - `false`: Sequential execution
  - Number: Specific parallel count
- `continue_on_error` (optional): Continue processing other items if one fails (default: false)
- `max_items` (optional): Maximum number of items to process

**Variable Access**:
Inside `do` commands, use `${item}` to reference the current item.

**Parallel Execution Strategy**:

```mermaid
graph TD
    Items[Work Items:<br/>item1, item2, ..., itemN] --> Distribute{parallel setting}

    Distribute -->|false or 1| Sequential[Sequential Execution<br/>One at a time]
    Distribute -->|Number N| Parallel[Parallel Execution<br/>N workers]

    Sequential --> W1[Worker: item1]
    W1 --> W2[Worker: item2]
    W2 --> W3[Worker: item3]

    Parallel --> P1[Worker 1: item1]
    Parallel --> P2[Worker 2: item2]
    Parallel --> P3[Worker N: itemN]

    W3 --> Continue{continue_on_error}
    P1 --> Continue
    P2 --> Continue
    P3 --> Continue

    Continue -->|false| StopOnError[Stop on first error]
    Continue -->|true| ProcessAll[Process all items]

    style Sequential fill:#e1f5ff
    style Parallel fill:#e8f5e9
    style StopOnError fill:#ffebee
    style ProcessAll fill:#fff3e0
```

**Figure**: Foreach execution showing sequential vs parallel processing and error handling.

**Example** (from workflows/fix-files-mapreduce.yml - conceptual):
```yaml
- foreach:
    items: "find src -name '*.rs' -type f"
    do:
      - claude: "/analyze-file ${item}"
      - shell: "rustfmt ${item}"
    parallel: 4
    continue_on_error: true
```

!!! example "Common Use Cases"
    - Processing multiple files (e.g., formatting all `.rs` files)
    - Running operations on a list of identifiers
    - Batch operations with controlled parallelism
    - Conditional processing based on item properties

!!! tip "Parallel Execution"
    Set `parallel: 5` to process 5 items concurrently. Use `parallel: false` for sequential processing.

### Write File Commands

Write content to files with format validation and directory creation.

**Source**: src/config/command.rs:348, src/config/command.rs:279-317

**Syntax**:
```yaml
- write_file:
    path: "output/file.json"        # Required: file path (supports variables)
    content: "${map.results}"       # Required: content to write (supports variables)
    format: json                    # Optional: text, json, yaml (default: text)
    mode: "0644"                    # Optional: file permissions (default: 0644)
    create_dirs: true               # Optional: create parent directories (default: false)
```

**Fields**:
- `path` (required): File path to write to (supports variable interpolation)
- `content` (required): Content to write (supports variable interpolation)
- `format` (optional): Output format - text, json, or yaml (default: text)
  - `text`: Write content as-is
  - `json`: Validate JSON and pretty-print
  - `yaml`: Validate YAML and format
- `mode` (optional): File permissions in octal format (default: "0644")
- `create_dirs` (optional): Create parent directories if they don't exist (default: false)

**Format Types** (src/config/command.rs:301-313):
- `Text`: No processing, write content directly
- `Json`: Validate JSON syntax and pretty-print
- `Yaml`: Validate YAML syntax and format

**Example** (from workflows/debtmap-reduce.yml:105):
```yaml
- write_file:
    path: ".prodigy/map-results.json"
    content: "${map.results}"
    format: json
    create_dirs: true
```

!!! example "Common Use Cases"
    - Saving workflow results to files
    - Generating configuration files from templates
    - Writing JSON/YAML data with automatic validation
    - Creating reports and logs
    - Persisting intermediate data between workflow steps

!!! tip "Format Validation"
    Use `format: json` or `format: yaml` to automatically validate and pretty-print output. Invalid content will fail the step.

### Validation Commands

Validate implementation completeness using shell or Claude commands with threshold-based checking.

**Source**: src/cook/workflow/validation.rs:12-50

**Syntax**:
```yaml
- validate:
    shell: "validation-command"         # Either shell or claude or commands
    claude: "/validate-command"         # Either shell or claude or commands
    commands:                           # Or array of commands for multi-step validation
      - shell: "step1"
      - claude: "/step2"
    expected_schema: {...}              # Optional: JSON schema for validation output
    threshold: 100                      # Optional: completion threshold % (default: 100)
    timeout: 300                        # Optional: timeout in seconds
    result_file: "validation.json"      # Optional: where validation results are written
    on_incomplete:                      # Optional: commands to run if validation fails
      claude: "/fix-gaps ${validation.gaps}"
      max_attempts: 3
```

**Fields**:
- `shell` (optional): Shell command for validation (deprecated, use `commands`)
- `claude` (optional): Claude command for validation
- `commands` (optional): Array of commands for multi-step validation
- `expected_schema` (optional): JSON schema that validation output must match
- `threshold` (optional): Minimum completion percentage required (default: 100)
- `timeout` (optional): Timeout for validation command in seconds
- `result_file` (optional): File where validation results are written
- `on_incomplete` (optional): Configuration for handling incomplete implementations
  - Nested command to execute if validation fails
  - `max_attempts`: Maximum retry attempts

**Validation Output Format**:
Validation commands should output JSON matching this structure:
```json
{
  "complete": true,
  "score": 100,
  "gaps": [],
  "incomplete_specs": []
}
```

**Example with result_file** (from workflows/implement.yml:8):
```yaml
validate:
  claude: "/prodigy-validate-spec $ARG --output .prodigy/validation-result.json"
  result_file: ".prodigy/validation-result.json"
  threshold: 100
  on_incomplete:
    claude: "/prodigy-complete-spec $ARG --gaps ${validation.gaps}"
```

**Example with commands array** (from workflows/debtmap.yml:13):
```yaml
validate:
  commands:
    - claude: "/prodigy-validate-debtmap-plan --before .prodigy/debtmap-before.json --plan .prodigy/IMPLEMENTATION_PLAN.md --output .prodigy/plan-validation.json"
  result_file: ".prodigy/plan-validation.json"
  threshold: 75
  on_incomplete:
    claude: "/fix-plan-gaps --gaps ${validation.gaps}"
    max_attempts: 3
```

!!! example "Common Use Cases"
    - Verifying specification completeness
    - Checking implementation against requirements
    - Validating test coverage meets thresholds
    - Ensuring documentation quality standards
    - Multi-step validation pipelines

!!! tip "Iterative Validation"
    Use `validate:` with `on_incomplete` handlers for iterative refinement workflows. The `threshold` setting and `max_attempts` provide control over retry behavior.

### Common Fields

Several fields are available across all command types:

**Source**: src/config/command.rs:320-401

- `id` (string): Unique identifier for referencing command outputs
- `timeout` (number): Maximum execution time in seconds
- `when` (string): Conditional execution expression
- `on_success` (command): Command to run on successful execution
- `on_failure` (command): Command to run on failed execution
- `capture_output` (boolean or string): Capture command output (true/false or variable name)
- `capture_format` (string): Format for captured output (json, text, lines)
- `capture_streams` (string): Which streams to capture (stdout, stderr, both)
- `output_file` (string): File to redirect output to

### Command Exclusivity

Each workflow step must specify **exactly one** command type. You cannot combine multiple command types in a single step:

=== "Valid ✓"
    ```yaml
    - shell: "cargo test"        # ✓ Only shell command
    - claude: "/lint"            # ✓ Only Claude command
    - analyze:                   # ✓ Only analyze command
        max_cache_age: 300
    - validate:                  # ✓ Only validate command
        commands:
          - shell: "cargo test"
    ```

=== "Invalid ✗"
    ```yaml
    - shell: "cargo test"        # ✗ Cannot combine shell and claude
      claude: "/lint"

    - analyze: {...}             # ✗ Cannot combine analyze and claude
      claude: "/review"

    - validate: {...}            # ✗ Cannot combine validate and foreach
      foreach: {...}
    ```

**Enforcement**: src/config/command.rs:465-476

### Deprecated: Test Command

!!! warning "Deprecated Syntax"
    The `test:` command syntax is deprecated and will be removed in a future version.
    Use `shell:` with `on_failure:` instead for the same functionality.

**Source**: src/config/command.rs:446-463

**Old (deprecated)**:
```yaml
- test:
    command: "cargo test"
    on_failure:
      claude: "/debug-test"
```

**New (recommended)**:
```yaml
- shell: "cargo test"
  on_failure:
    claude: "/debug-test"
```

The migration is straightforward: replace `test:` → `command:` with `shell:` and keep all other fields unchanged.