@miller-tech/uap 1.30.1 → 1.31.0

package/README.md

@@ -91,7 +91,7 @@ uap setup -p all
 | Browser            | 1 module       | Stealth web automation via CloakBrowser (Playwright drop-in)                     |
 | MCP Router         | 11 modules     | 2-tool meta-router + expert-consultation registry (98% token savings)            |
 | Models             | 10 modules     | Multi-model routing, planning, execution, validation, 13 model profiles          |
-| Delivery Harness   | 8 modules      | `uap deliver`: convergence loop, best-of-N explorer, critic, practice recall, escalation |
+| Delivery Harness   | 11 modules     | `uap deliver`: convergence loop, best-of-N explorer, critic, practice recall, escalation, ideation seeds, HALO tracing, coordination + deploy queueing |
 | Patterns           | 23 patterns    | Battle-tested workflows from Terminal-Bench 2.0                                  |
 | Droids             | 30 experts     | Full SDLC expert stack: strategy, design, build, review, release, ops ([reference](docs/reference/EXPERT_DROIDS.md)) |
 | Expert Orchestrator | 1 module      | Adaptive droid-chain selection across plan→design→implement→review→release       |
@@ -368,8 +368,13 @@ the model's say-so.
 3. **Structured critic** (`--critic`) — turns a failed turn's gate output into a numbered, file-scoped repair plan via a gate-specific analyst persona.
 4. **Best-practice recall** (`--practices`) — injects provenance-safe practice cards learned from past successful deliveries, retrieved by semantic similarity (nomic-768 embeddings, keyword fallback).
 5. **Escalation ladder** (`--escalate`) — on stagnation, climbs cheap→expensive: widen exploration → enable the critic → switch to a stronger model.
+6. **Divergent ideation** (`--ideate`, `--ideate-project <name>`) — replaces the static strategy seeds with task-specific, deliberately diverse seeds: generated by a bisociation-style model call, or taken from an open-collider project's curated ideas (`uap ideate`). Implies best-of-N exploration.
+7. **HALO tracing** (`--halo`) — emits one AGENT span per run and one CHAIN span per turn (scores, strategies, failed gates) so `uap harness analyze` can mine systemic failure modes across runs.
+8. **Coordination** (`--coordinate`) — registers the run with the multi-agent coordination layer (`uap agent`): announces work on the project, warns about overlapping agents, heartbeats every turn, completes/deregisters on exit.
+9. **Deploy batching** (`--deploy`) — on success, queues a commit of the applied files into the deploy batcher; execute with `uap deploy flush`.
+10. **`--optimize`** — one switch for every convergence aid: 4 candidates/turn + critic + practices + escalation + ideation + HALO + coordination (deploy stays explicit).
 
-### Components (8 modules)
+### Components (11 modules)
 
 | Component         | File                                  | Purpose                                                            |
 | ----------------- | ------------------------------------- | ----------------------------------------------------------------- |
@@ -381,6 +386,9 @@ the model's say-so.
 | Critic            | `src/delivery/critic.ts`              | Gate-persona repair plans from failed turns                        |
 | Practice Store    | `src/delivery/practice.ts`            | Provenance-safe best-practice cards with semantic recall           |
 | Escalation        | `src/delivery/escalation.ts`          | Stagnation-driven ladder returning loop directives                 |
+| Ideation Seeder   | `src/delivery/ideation.ts`            | Divergent strategy seeds (generated or from curated ideas)         |
+| HALO Tracer       | `src/delivery/halo-trace.ts`          | Run/turn spans for `uap harness analyze`                           |
+| Run Coordinator   | `src/delivery/run-coordinator.ts`     | `uap agent` registration/heartbeat + `uap deploy` commit queueing  |
 
 The model is reached through an OpenAI-compatible client
 (`src/models/openai-compat-client.ts`) — the local inference gateway,
@@ -401,6 +409,12 @@ uap deliver "..." --dry-run
 
 # Scope to a subset of gates, cap turns, target another project
 uap deliver "..." --gates build,test --max-turns 8 --project-root ../service
+
+# Everything on: exploration, critic, practices, escalation, ideation, HALO, coordination
+uap deliver "refactor the cache layer to LRU with TTL" --optimize
+
+# Divergent ideation seeds + queue a commit into the deploy batcher on success
+uap deliver "..." --ideate --candidates 4 --deploy
 ```
 
 ### Key flags
@@ -416,6 +430,12 @@ uap deliver "..." --gates build,test --max-turns 8 --project-root ../service
 | `--no-semantic`            | Use keyword (not embedding) practice recall                            |
 | `--escalate`               | Escalation ladder on stagnation                                        |
 | `--escalate-model <preset>`| Stronger model for the final escalation tier                          |
+| `--ideate`                 | Divergent ideation: task-specific strategy seeds (implies exploration) |
+| `--ideate-project <name>`  | Seed exploration from `projects/<name>` curated ideas (`uap ideate`)   |
+| `--halo`                   | Emit HALO spans; analyze with `uap harness analyze`                    |
+| `--coordinate`             | Register with `uap agent`: announce, heartbeat, overlap detection      |
+| `--deploy`                 | On success, queue a commit into the deploy batcher (`uap deploy`)      |
+| `--optimize`               | Enable every convergence aid (deploy excluded)                         |
 | `--endpoint <url>`         | Override the model endpoint (OpenAI-compatible `/v1`)                  |
 | `--dry-run` / `--json`     | Show the plan only / emit machine-readable result                     |