@vexdo/cli 0.1.3 → 0.2.0

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.
Files changed (3) hide show
  1. package/README.md +79 -2
  2. package/dist/index.js +1236 -334
  3. package/package.json +5 -1
package/README.md CHANGED
@@ -197,10 +197,87 @@ Non-goals:
197
197
  - **Task escalated**
198
198
  - Review `.vexdo/logs/*` and rerun with `vexdo fix "..."`.
199
199
 
200
- ## 10) Contributing
200
+ ## 10) Roadmap
201
+
202
+ Vexdo currently runs tasks sequentially using a local Codex subprocess. The roadmap
203
+ evolves the orchestrator toward cloud-native execution, smarter context management,
204
+ and better observability.
205
+
206
+ ### Cloud execution and parallelism
207
+
208
+ - **Codex Cloud execution** — replace local `codex exec` with `codex cloud exec`. Each
209
+ step runs in an isolated cloud sandbox with direct GitHub repository access. The
210
+ orchestrator submits tasks, polls for completion, retrieves diffs, and uses
211
+ `codex cloud exec resume` for fix iterations. Eliminates local git state dependencies.
212
+
213
+ - **Parallel step execution** — once isolation is handled by Codex Cloud, steps without
214
+ `depends_on` relationships dispatch concurrently. A dependency-aware worker pool means
215
+ a three-service task completes in the time of its longest step, not the sum of all.
216
+
217
+ ### Review and verification
218
+
219
+ - **GitHub Copilot CLI as reviewer** — replace the Claude-based reviewer with
220
+ `copilot --output-format=json`. Copilot reads the local diff with full repository
221
+ context (imports, types, related files). Claude stays as the Arbiter.
222
+
223
+ - **Verification ladder** — structured must-haves in task YAML (`must_haves: [...]`).
224
+ Arbiter verifies each requirement against the diff at four tiers: static (file/export
225
+ presence), command (tests pass), behavioral (observable output), or human (escalate).
226
+ Submit is only allowed when all must-haves pass.
227
+
228
+ - **Stuck detection** — if Codex produces the same diff twice, a diagnostic retry fires
229
+ with a targeted prompt. On a second identical diff, the loop escalates with a structured
230
+ diagnostic showing exactly which review comments were not addressed.
231
+
232
+ ### Context and memory
233
+
234
+ - **Fresh context injection** — before each Codex submission, prepend summaries of
235
+ completed steps and the decisions register to the prompt. Prevents Codex from
236
+ re-implementing utilities already built by earlier steps. Capped at 2000 tokens.
237
+
238
+ - **Decisions register** — `.vexdo/decisions.md`: an append-only table of architectural
239
+ decisions made during execution (validation library, storage strategy, naming conventions).
240
+ Arbiter populates it automatically; injected into every subsequent step prompt.
241
+
242
+ - **Scout agent** — a focused Claude call before Codex submission that scans the target
243
+ service's codebase and returns relevant existing files, reuse hints, and conventions to
244
+ follow. Non-fatal: if Scout fails, execution continues without it.
245
+
246
+ - **Adaptive replanning** — after each step completes, a lightweight Claude call checks
247
+ whether remaining step specs are still accurate. Proposes updates for developer
248
+ confirmation before the next step runs.
249
+
250
+ ### Resilience
251
+
252
+ - **Continue-here protocol** — `.vexdo/continue.md` checkpoint written at every major
253
+ phase transition (codex submitted, codex done, review iteration, arbiter done).
254
+ `vexdo start --resume` reads the checkpoint and resumes from the exact saved position
255
+ rather than re-entering the step from the beginning.
256
+
257
+ ### Observability and interaction
258
+
259
+ - **Cost and token tracking** — every Claude API call captures token usage and estimated
260
+ cost. Per-step and total costs are shown in `vexdo status`. Optional budget ceiling in
261
+ `.vexdo.yml` pauses execution before overspending.
262
+
263
+ - **UAT script generation** — after all steps complete, Vexdo writes `.vexdo/uat.md`:
264
+ a human test script derived from step must-haves and Arbiter summaries. `vexdo submit`
265
+ warns if UAT items are unchecked (override with `--skip-uat`).
266
+
267
+ - **Discuss command** — `vexdo discuss <task-id>` opens an interactive Claude session
268
+ with full task context pre-loaded. Ask questions about what was built, queue spec
269
+ updates for pending steps, steer execution from a second terminal while `start` runs.
270
+
271
+ ### Task board TUI
272
+
273
+ A `vexdo board` command built with **Ink** (React for CLIs) that renders all task lanes
274
+ as a navigable terminal board. Keyboard shortcuts to start, edit, inspect, and abort
275
+ tasks without leaving the terminal.
276
+
277
+ ## 11) Contributing
201
278
 
202
279
  See [CONTRIBUTING.md](./CONTRIBUTING.md).
203
280
 
204
- ## 11) License
281
+ ## 12) License
205
282
 
206
283
  MIT.