sdd-flow 2.0.0 → 2.0.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,11 @@
1
1
  # Changelog
2
2
 
3
+ ## 2.0.1 — 2026-11-20
4
+
5
+ - Default new projects to step; follow explicit multi-task/all-task authorization as a bounded batch without resetting existing project preferences.
6
+ - Redesign README and update usage, migration, format and validation documentation accordingly.
7
+ - Add default engineering rules: KISS/Clean Code/YAGNI/DRY/SOLID guidance and no redundant comments or docblocks.
8
+
3
9
  ## 2.0.0 — 2026-11-20
4
10
 
5
11
  Breaking: Node.js 22+, schema 2 artifacts, canonical YAML task state and version-bound scope/evidence.
package/README.md CHANGED
@@ -1,113 +1,402 @@
1
+ <div align="center">
2
+
1
3
  # sdd-flow
2
4
 
3
- A portable development methodology for AI-assisted software work. Organize intent, load relevant
4
- task context and retain traceable execution. Harness orchestration is outside the core's scope.
5
+ ### A clear workflow for developing with AI.
6
+
7
+ Relevant context for the agent.<br>
8
+ Well-defined changes for you.<br>
9
+ A traceable history for the project.
10
+
11
+ ![Node.js 22 or newer](https://img.shields.io/badge/Node.js-%E2%89%A5%2022-417E38)
12
+ ![MIT license](https://img.shields.io/badge/license-MIT-2563EB)
13
+ ![Portable methodology](https://img.shields.io/badge/methodology-stack%20independent-475569)
14
+
15
+ [Português brasileiro](README.pt-BR.md) · [Get started](#get-started-in-3-steps) · [Workflow](#how-it-works) · [Context](#how-context-stays-focused) · [Documentation](#documentation)
16
+
17
+ </div>
18
+
19
+ ---
20
+
21
+ **sdd-flow organizes AI-assisted development into cohesive changes that you can understand, implement and verify.**
22
+
23
+ Describe what you need. The agent consults relevant sources, prepares a spec, implements the authorized scope and records the results. Project knowledge and progress remain in the repository for the next session.
24
+
25
+ | Organization | Context | Traceability |
26
+ |---|---|---|
27
+ | One spec per cohesive change | Sources selected for the phase and task | Connected requirements, tasks, decisions and verification |
28
+
29
+ **Your daily focus stays on the spec.** The agent and CLI maintain supporting records. Configuration, skills and general documents are consulted when relevant.
30
+
31
+ ## How it works
32
+
33
+ ```mermaid
34
+ flowchart TD
35
+ A["Describe the change"] --> B["Agent researches and plans"]
36
+ B --> C["Review the scope"]
37
+ C --> D["Agent implements and verifies"]
38
+ D --> E{"Result"}
39
+ E -->|"Needs adjustment"| D
40
+ E -->|"Ready to close"| F["Consolidate and record the outcome"]
41
+ D -. "Requirement changed" .-> B
42
+ ```
43
+
44
+ **Research → plan → implement → validate → close.** These are responsibilities, not mandatory separate conversations.
45
+
46
+ - **Before:** clarify the goal, define acceptance and consider affected consumers.
47
+ - **During:** execute cohesive tasks and check regressions.
48
+ - **After:** record observations and consolidate current knowledge.
5
49
 
6
- This checkout targets **2.0.0**. It is a breaking evolution from 1.x; publishing this source is a separate release action.
7
- Requires Node.js 22 or newer.
50
+ Use the method for **new projects** or **existing applications**, with any application stack. Node.js 22+ is required for the CLI.
8
51
 
9
- Portuguese guides: [methodology](docs/methodology.pt-BR.md), [local testing](docs/local-testing.pt-BR.md),
10
- [usage](docs/usage.pt-BR.md) and [GitHub/npm release](docs/releases.pt-BR.md).
52
+ ## Get started in 3 steps
11
53
 
12
- ## Try this checkout
54
+ > **This README describes 2.x**, a breaking evolution from 1.x (see [migration](docs/migration.md)). Confirm your CLI with `sdd --version`.
13
55
 
14
- ~~~sh
56
+ ### 1. Install the CLI
57
+
58
+ With Node.js 22+:
59
+
60
+ ```sh
61
+ npm install -g sdd-flow
62
+ sdd --version
63
+ ```
64
+
65
+ <details>
66
+ <summary>Other options: run without installing, or from source</summary>
67
+
68
+ Run without a global install:
69
+
70
+ ```sh
71
+ npx sdd-flow@latest --help
72
+ ```
73
+
74
+ From the **sdd-flow source checkout** (for contributors):
75
+
76
+ ```sh
15
77
  npm ci
16
78
  npm link
79
+ sdd --version
80
+ ```
81
+
82
+ Installing the package provides the CLI. Project files are installed by running `sdd init` inside each application.
83
+
84
+ </details>
85
+
86
+ ### 2. Initialize your project
87
+
88
+ ```sh
17
89
  cd /path/to/your-project
18
- sdd init --harness claude-code
19
- # Alternatively: sdd init --harness opencode
20
- ~~~
21
-
22
- Complete agents/SETUP.md and the short project map. Configure actual checks, then run sdd doctor.
23
- For projects already using 1.x, read [migration](docs/migration.md) first.
24
-
25
- ## A change from plan to evidence
26
-
27
- ~~~sh
28
- sdd new export-csv --mode standard
29
- sdd context 001-export-csv --phase plan
30
- # Author SPEC.md and TASK.md using the canonical formats.
31
- sdd check 001-export-csv
32
- # After the user authorizes this exact scope:
33
- sdd approve 001-export-csv --by "Moisés"
34
- sdd context 001-export-csv --phase implement --task T01
35
- # Implement and verify the task, then set its canonical status to done.
36
- sdd verify 001-export-csv --scope partial
37
- # Consolidate lasting capability contracts before final verification.
38
- sdd verify 001-export-csv
39
- sdd close 001-export-csv --by "Moisés" --note "Local correction; no lasting contract changed" --archive
40
- ~~~
41
-
42
- The approval command records authorization; it does not prove who typed it or grant runtime permissions.
43
- Agents must use actual authorization already given by the user and must not fabricate it.
44
-
45
- ## What changes in 2.x
46
-
47
- - **Short entry point:** a managed block preserves existing root instructions.
48
- - **Structured metadata:** real YAML parsing, schema versions, stable IDs and explicit references.
49
- - **Bounded context:** select the phase/task, current capabilities, active decisions and relevant skills.
50
- Overflow fails with file sizes; documents are never silently truncated.
51
- - **One task state:** TASK.md front matter is canonical. STATE.md is an optional short resumption note.
52
- - **Adaptive rigor:** light, standard and rigorous modes; step or batch autonomy.
53
- - **Executable checks:** approvals bind to scope; receipts bind to source files, selected contracts and configuration.
54
- - **Manual evidence:** manual and inspection cases can pass when observed and recorded for current inputs.
55
- - **Consolidation:** close only with current full verification and an explicit current-knowledge update or explanation.
56
- - **Safe maintenance:** dry-run, ownership checks, backups, legacy preservation and active adapter entry points.
57
- - **Portable distribution:** export Agent Skills bundles, a Claude Code plugin directory, or an OpenCode skill layout.
58
-
59
- ## Commands
60
-
61
- | Command | Purpose |
90
+ sdd init
91
+ ```
92
+
93
+ Optional native entry points:
94
+
95
+ ```sh
96
+ sdd harness claude-code
97
+ # Alternative: sdd harness opencode
98
+ ```
99
+
100
+ For other environments, direct the agent to the project's `AGENTS.md`. The core uses files and commands; a native adapter is optional.
101
+
102
+ ### 3. Keep using short prompts
103
+
104
+ For initial setup:
105
+
106
+ ```text
107
+ Read and follow agents/SETUP.md to configure the workflow.
108
+ ```
109
+
110
+ Then describe the change and select the phase:
111
+
112
+ | Phase | Prompt |
113
+ |---|---|
114
+ | Research | Read and follow agents/prompts/rpi-research.md. |
115
+ | Plan | Read and follow agents/prompts/rpi-plan.md. |
116
+ | Implement | Read and follow agents/prompts/rpi-implement.md. |
117
+ | Validate | Read and follow agents/prompts/rpi-validate.md. |
118
+ | Close | Read and follow agents/prompts/rpi-close.md. |
119
+
120
+ The procedures contain the workflow instructions. You do not need to repeat them in chat.
121
+ The request and active spec must be identifiable. If several changes exist or a new session lacks context,
122
+ provide the spec ID or let the agent clarify which one to use.
123
+
124
+ Setup is initial adoption, not a repeated step for every spec. Once complete, `SETUP.md` may be removed.
125
+ Project language and conventions belong in configuration and policies.
126
+
127
+ **Follow the spec, tasks and verification outcome.** The agent uses the CLI and maintains supporting records.
128
+ Phases may run in the same conversation.
129
+
130
+ For an existing application, setup records an observed baseline without refactoring the application
131
+ or inventing historical specs.
132
+
133
+ ## What is added to your repository
134
+
135
+ | Files | Purpose |
136
+ |---|---|
137
+ | `AGENTS.md` and `QUICKSTART.md` | Entry point and getting started |
138
+ | `agents/PROJECT.md` | Short application map and source references |
139
+ | `agents/POLICIES.md` and `agents/QUALITY.md` | Project constraints and quality criteria |
140
+ | `agents/config.yaml` | Language, autonomy, rigor and actual check commands |
141
+ | `agents/RULES.md`, `agents/AGENTS.md` and `agents/prompts/` | Method and phase procedures |
142
+ | `agents/specs/` | Active changes |
143
+ | `agents/current/` and active decisions | Current knowledge |
144
+ | `agents/archive/` | Closed changes |
145
+ | `agents/skills/` | Reusable project procedures |
146
+
147
+ Installation includes templates and maintenance metadata. **It does not generate features or fictional specs.**
148
+
149
+ Each change uses only the applicable artifacts:
150
+
151
+ | Artifact | Role |
62
152
  |---|---|
63
- | init | Install missing project files and preserve project-owned configuration |
64
- | docs --types prd,sd,ds,adr,roadmap | Create only selected missing reference drafts; --dry-run previews and existing documents are preserved |
65
- | new NAME | Reserve an ID and create a change in the chosen mode |
66
- | context ID | Compose the bounded packet; --inspect diagnoses size without emitting bodies; headings are selected in the manifest |
67
- | skills list / show NAME | Discover metadata or read a selected body |
68
- | check ID | Validate structure, references, task dependencies and phase gates |
69
- | approve ID --by NAME | Record actual authorization for the current normative scope |
70
- | revise ID --by NAME --reason TEXT | Preserve revision history, reopen impacted work and invalidate the prior scope approval |
71
- | evidence ID | Create a template bound to current inputs for manual/inspection cases |
72
- | handoff ID | Prepare version-bound external limitations and safe user verification instructions |
73
- | verify ID | Run local checks; --handoff records limitations; --run-external opts into authorized external commands |
74
- | close ID --by NAME | Require current full verification and consolidation; optionally --archive |
75
- | reopen / cancel ID | Record explicit lifecycle decisions without deleting historical IDs |
76
- | status / doctor | Inspect active/archive changes or diagnose project setup |
77
- | update --dry-run | Preview installation changes; customized files are preserved unless explicitly replaced |
78
- | migrate ID --dry-run | Preview conservative conversion of legacy artifacts |
79
- | harness NAME | Install active native command/skill wrappers; --remove removes only unmodified owned files |
80
- | export FORMAT --out DIR | Build a new directory for claude-code, opencode or agent-skills distribution |
81
-
82
- All commands support --json. Exit codes: 0 completed (passed or limited), 1 invalid/failing, 2 partial verification.
83
- Limited closes implementation with pending external verification; it never claims those tests passed or authorizes deployment.
84
- Required checks block on failure; advisory checks remain visible warnings. QUALITY.md is loaded only for planning/review.
85
- With --json, success and error objects use stdout; runtime/environment diagnostics remain on stderr.
86
- CLI and documentation use English; project code, UI and conversation languages are configured by the project/user.
153
+ | `change.yaml` | Identity, references, dependencies and lifecycle |
154
+ | `SPEC.md` | Requirements and acceptance |
155
+ | `TASK.md` | Plan and canonical task state, when applicable |
156
+ | `RESEARCH.md` | Research when useful or required by the mode |
157
+ | Evidence and run records | What was checked, for which inputs, with which result |
158
+
159
+ The agent maintains these as work progresses. Research, evidence and history are not all created by `sdd new`.
160
+
161
+ ## How context stays focused
162
+
163
+ **The repository stores knowledge. The agent receives a relevant selection for the current task.**
164
+
165
+ ```mermaid
166
+ flowchart TD
167
+ A["Active spec and task"] --> D["Select context"]
168
+ B["Contracts, decisions and skills"] --> D
169
+ C["Relevant docs/ references"] --> D
170
+ D --> E{"Within budget?"}
171
+ E -->|"Yes"| F["Provide context to the agent"]
172
+ E -->|"No"| G["Review sources, sections and scope"]
173
+ G --> D
174
+ ```
175
+
176
+ - A short entry point maps the project.
177
+ - Phase and task guide selection.
178
+ - Skills are discovered by name and description; selected bodies are loaded.
179
+ - Large references can be selected by heading.
180
+ - Full historical logs are not loaded automatically.
181
+ - `TASK.md` holds task state; `STATE.md` is an optional short resumption note.
182
+
183
+ ```sh
184
+ sdd context 001-my-change --phase implement --task T01 --inspect
185
+ ```
186
+
187
+ Inspection reports sizes without emitting document bodies. The default budget is **48,000 UTF-8 bytes**,
188
+ configurable. It is not a token count or the complete chat context window.
189
+
190
+ Generating an oversized packet reports the problem **without silently truncating content**.
191
+ The agent can select sections, review references or split independent deliverables.
192
+
193
+ Selection depends on maintained references. The CLI cannot infer every semantic dependency or control chat history.
194
+
195
+ ## Keep references organized in docs/
196
+
197
+ PRDs, business rules, system design, ADRs, design systems, contracts and external materials can stay
198
+ in `docs/`, with or without subdirectories.
199
+
200
+ **Centralize sources; select what to read.** The agent consults relevant documents and preserves their origin.
201
+
202
+ Example in `change.yaml`:
203
+
204
+ ```yaml
205
+ sources:
206
+ - path: docs/product/prd.md
207
+ section: Data export
208
+ context:
209
+ - path: docs/design/design-system.md
210
+ section: States and interactions
211
+ ```
212
+
213
+ There is no mandatory document bundle. A clear description can be enough to start.
214
+
215
+ Optional reference drafts:
216
+
217
+ ```sh
218
+ sdd docs --types prd,sd,ds,adr,roadmap --dry-run
219
+ sdd docs --types prd,sd,ds,adr,roadmap
220
+ ```
221
+
222
+ Existing files are preserved. New files are drafts for review.
223
+ See [setup and reference documents (Português)](docs/setup-documents.pt-BR.md).
224
+
225
+ ## Choose proportional rigor
226
+
227
+ | Mode | Typical use |
228
+ |---|---|
229
+ | `light` | Small change; tasks and research may be omitted |
230
+ | `standard` | Bounded feature with verifiable tasks |
231
+ | `rigorous` | Sensitive or uncertain work with research and explicit approval |
232
+
233
+ New installations default to **step**: implement and verify one task, then stop for review.
234
+ Authorize named tasks or all tasks to continue as a bounded batch without repeated confirmations.
235
+ Plan approval alone does not override step. **Batch** can also be the project's standing preference;
236
+ updates preserve an existing choice. Each task is still checked and dependencies respected.
237
+
238
+ ## Verify honestly, including unavailable environments
239
+
240
+ The CLI runs configured checks and records commands, results and inputs.
241
+ Approval binds to scope; verification binds to the inputs examined.
242
+
243
+ | Result | Meaning |
244
+ |---|---|
245
+ | `passed` | Full verification passed for current inputs |
246
+ | `limited` | Implementation complete with explicit external limitations and pending verification |
247
+ | `partial` | Conditions for final closure are not yet met |
248
+ | `failed` | Required failure observed or inputs changed during execution |
249
+
250
+ **Unavailable staging or production access need not block completed implementation.**
251
+ A reviewed handoff records limitations and can generate safe steps in `USER_TESTS.md`.
252
+
253
+ `limited` does not claim external tests passed or authorize deployment.
254
+ Omitting evidence cannot erase a known failure for the same inputs.
255
+
256
+ Required failures block passing verification; advisory issues remain warnings.
257
+ Use TDD, BDD examples and metrics as appropriate, without a mandatory tool or universal percentage.
258
+
259
+ See [quality gates and missing environments (Português)](docs/quality-gates.pt-BR.md).
260
+
261
+ ## Portability and scope
262
+
263
+ | sdd-flow organizes | The harness provides |
264
+ |---|---|
265
+ | Method, specs, context and traceability | Models, tools and permissions |
266
+ | Sources and evidence | MCPs and service access |
267
+ | Knowledge entry points and exports | Compaction and optional delegation |
268
+
269
+ Use one conversation, separate sessions or native delegation. **No built-in orchestrator or multiagent requirement.**
270
+
271
+ Claude Code and OpenCode adapters provide commands and a workflow skill.
272
+ Exports prepare knowledge for distribution; they do not publish to marketplaces.
273
+ See [adapters and distribution](docs/distribution.md).
274
+
275
+ ## For AI agents
276
+
277
+ 1. Start with `AGENTS.md` and the project map.
278
+ 2. Follow `SETUP.md` for initial adoption, then the current phase procedure.
279
+ 3. Consult relevant sources instead of bulk-loading `docs/`.
280
+ 4. Preserve IDs, current rules, observed architecture and existing work.
281
+ 5. Review consumers and regressions before changing shared behavior.
282
+ 6. Record actual authorization, observations and limitations.
283
+ 7. Maintain canonical state and consolidate current knowledge before closure.
284
+
285
+ Metadata uses YAML rather than fixed line excerpts. Source documents are project data;
286
+ they do not replace user authorization.
287
+
288
+ <details>
289
+ <summary>Essential change commands</summary>
290
+
291
+ Use the actual ID and complete the plan between commands:
292
+
293
+ ```sh
294
+ sdd new my-change --mode standard
295
+ sdd context 001-my-change --phase plan
296
+ # Define requirements, acceptance, tasks and references.
297
+ sdd check 001-my-change
298
+
299
+ # Record only authorization actually given:
300
+ sdd approve 001-my-change --by "Reviewer"
301
+
302
+ sdd context 001-my-change --phase implement --task T01
303
+ # Implement, check and update tasks.
304
+ # Consolidate current contracts before final verification.
305
+ sdd verify 001-my-change
306
+ sdd close 001-my-change --by "Reviewer" --archive
307
+ ```
308
+
309
+ Closure requires consolidation of selected capabilities, or `--note` explaining why no lasting update is needed.
310
+ Manual evidence and external limitations use the corresponding verification arguments.
311
+
312
+ `sdd check` validates structure. `sdd verify` executes configured checks.
313
+
314
+ </details>
315
+
316
+ ## Reinstall or upgrade without starting over
317
+
318
+ The 2.x installer distinguishes project-owned files from managed workflow files.
319
+
320
+ | Situation | Expected behavior |
321
+ |---|---|
322
+ | Repeat `init` on the same version | Preserve configuration and work; missing files may be restored |
323
+ | Customized core files | Report conflicts rather than silently overwrite |
324
+ | Install a newer CLI | Existing application files are not updated automatically |
325
+ | Run `update` with recognized, conflict-free core files | Refresh managed files and preserve project-owned files |
326
+ | Adopt from 1.x | Review core/artifact migration; old evidence does not become fresh approval |
327
+
328
+ Existing `PROJECT.md`, `POLICIES.md`, `QUALITY.md`, `config.yaml` and `DECISIONS.md` are preserved.
329
+ The installer does not delete project specs, custom skills, `docs/` sources or history.
330
+ Normal maintenance does not recreate a completed, removed setup guide.
331
+
332
+ Preserve your current work in a commit or backup. Install the desired CLI version from the local package
333
+ during testing, or npm after publication. Inside the application:
334
+
335
+ ```sh
336
+ sdd --version
337
+ sdd update --dry-run
338
+ # Review the plan and resolve conflicts before applying.
339
+ sdd update
340
+ sdd doctor
341
+ ```
342
+
343
+ Refresh native entry points with `sdd harness claude-code` or `sdd harness opencode` when used.
344
+ Native settings are not overwritten; old permissions/hooks require their own review.
345
+
346
+ <details>
347
+ <summary>Migrating from 1.x or handling a conflict</summary>
348
+
349
+ The updater recognizes unchanged templates from the original 1.3.3 baseline.
350
+ Customized versions, including third-party adaptations, may need manual review.
351
+
352
+ For a legacy spec that you will resume or normalize:
353
+
354
+ ```sh
355
+ sdd migrate 001-my-change --dry-run
356
+ sdd migrate 001-my-change
357
+ ```
358
+
359
+ Use the real ID. Migration preserves old bodies and creates structures for review.
360
+ Review requirements, acceptance, tasks and references before normalization.
361
+ You do not need to convert all closed historical work just to adopt the flow for new changes.
362
+
363
+ Customized core files cause a conflict before planned core writes are applied.
364
+ Move project-specific rules into policies, contracts or skills before deciding to replace core files.
365
+ `--replace-managed` is explicit and should follow review; replaced content is backed up.
366
+
367
+ These protections describe the 2.x installer. Do not use a 1.x installer to downgrade
368
+ or assume automatic compatibility with future versions.
369
+ Read the [migration guide](docs/migration.md).
370
+
371
+ </details>
87
372
 
88
373
  ## Documentation
89
374
 
90
- - [Português](README.pt-BR.md)
91
- - [Quality gates and missing environments (Português)](docs/quality-gates.pt-BR.md)
92
- - [Usage scenarios and project-specific policies (Português)](docs/usage.pt-BR.md)
93
- - [Updating the existing npm package (Português)](docs/releases.pt-BR.md)
94
- - [Formats and a complete example](docs/formats.md)
95
- - [Context and lifecycle design](docs/architecture.md)
96
- - [Upgrading existing projects](docs/migration.md)
97
- - [Adapters and distribution](docs/distribution.md)
98
- - [Quality checks and evaluation](docs/quality.md)
99
- - [Observed implementation validation](docs/validation.md)
100
- - [Contributing](CONTRIBUTING.md)
101
- - [Changelog](CHANGELOG.md)
102
-
103
- ## Limits
104
-
105
- The CLI validates declared relationships and records actual executions. It cannot prove that a passing test
106
- semantically covers every requirement, authenticate an approval author, or sandbox a configured command.
107
- Review assertions and use your tool's native permissions plus isolated test environments.
108
- Source fingerprints use project files outside agents/ and dependency internals plus explicitly selected
109
- workflow contracts. Configure verification.exclude for generated artifacts; exclusions affect evidence validity.
375
+ | Goal | Guide |
376
+ |---|---|
377
+ | Use the workflow end to end | [Usage (Português)](docs/usage.pt-BR.md) |
378
+ | Understand setup and documents | [Setup (Português)](docs/setup-documents.pt-BR.md) |
379
+ | Understand the method and changes | [Methodology (Português)](docs/methodology.pt-BR.md) |
380
+ | Configure quality and external tests | [Quality gates (Português)](docs/quality-gates.pt-BR.md) |
381
+ | Understand artifact formats | [Formats and examples](docs/formats.md) |
382
+ | Test locally | [Local testing (Português)](docs/local-testing.pt-BR.md) |
383
+ | Upgrade legacy projects | [Migration](docs/migration.md) |
384
+ | Prepare plugins and skills | [Distribution](docs/distribution.md) |
385
+ | Review observed validation and limits | [Implementation validation](docs/validation.md) |
386
+ | Publish a version | [GitHub and npm (Português)](docs/releases.pt-BR.md) |
387
+ | Contribute | [Contributing](CONTRIBUTING.md) · [Changelog](CHANGELOG.md) |
388
+
389
+ ## Important limits
390
+
391
+ The CLI validates declared relationships and records runs. Passing tests do not prove semantic coverage.
392
+ Approval is a recorded assertion, not identity authentication.
393
+
394
+ Configured commands run with current account permissions; the CLI is not a sandbox.
395
+ Review settings, use native harness controls and choose safe environments.
396
+
397
+ Context selection avoids unnecessary reading; no universal cost or generated-code improvement is claimed.
398
+ Interactive adapter compatibility must be checked against the actual tool versions in use.
110
399
 
111
400
  ## License
112
401
 
113
- MIT. Workflow knowledge and runtime integrations are maintained together in this repository.
402
+ [MIT](LICENSE).