ur-agent 1.21.0 → 1.22.1

package/CHANGELOG.md

@@ -1,5 +1,100 @@
 # Changelog
 
+## 1.22.1
+
+### Added
+- **Rich task decomposition.** `src/services/agents/decomposer.ts` splits large
+  goals into atomic subtasks with `goal`, `filesTouched`, `risk` (low/medium/high),
+  `testsRequired`, and `rollbackPoint`. `ur crew create|run|plan ... --decompose`
+  uses it; a deterministic fallback keeps the command offline and fast, while a
+  headless subagent path produces structured JSON.
+- **Parallel specialized subagents.** New `parallel` multi-agent pattern in
+  `src/services/agents/patterns.ts`: bug finder, patch writer, test writer,
+  security auditor, and style reviewer run concurrently, then a synthesizer
+  merges their outputs. Run it with
+  `ur pattern parallel "<task>" --execute [--dry-run]`.
+- **AgentKernel abstraction.** `src/services/agents/kernel.ts` is a pure
+  orchestrator with seven roles: planner, executor, verifier, critic, memory,
+  router, and guard. `ur spec run|verify <name> --kernel` routes through kernel
+  stages while the legacy path stays the default.
+- **Spec verification / verifier kernel role.** `ur spec verify <name>` runs
+  deterministic `.ur/verify.json` project gates, then a read-only deep
+  verification subagent that must prove compile/test/lint/diff/runtime before
+  `VERDICT: PASS`. Persists `.ur/specs/<name>/verification.md` and records the
+  result in `spec.json`.
+
+### Changed
+- **Version bump.** Updated from 1.22.0 to 1.22.1 across `package.json`,
+  `bunfig.toml`, the VS Code extension manifest, and the bundled CLI.
+
+### Verified
+- Added `test/decomposition.test.ts`, `test/parallelPattern.test.ts`, and
+  `test/kernel.test.ts` covering decomposition metadata, parallel pattern
+  workflow execution, and all seven kernel roles.
+
+## 1.22.0
+
+### Added
+- **Agent execution metrics in `ur eval`.** `ur eval run` now captures
+  cost, input/output tokens, model used, API duration, files changed,
+  insertions/deletions, command failures, human-edit heuristics, and optional
+  per-case `testCommand` results. Metrics are written by the headless child
+  via `UR_EVAL_METRICS_FILE`, so future parallel eval runs stay safe.
+- **Richer eval dashboard.** `ur eval dashboard` and
+  `ur eval report <suite> --dashboard` generate local HTML dashboards with
+  summary cards (pass rate, test pass rate, cost, tokens, files changed, command
+  failures, human edits, duration) and a per-case timeline showing model, time,
+  cost, tokens, diffs, test result, and output preview.
+- **Per-case run metrics persistence.** `ur eval run <suite> --metrics` writes
+  each case's metrics to `.ur/evals/.runs/<suite>/<case>.json` for downstream
+  analysis.
+- **Benchmark-style reporting.** `formatEvalReport` prints aggregate cost,
+  tokens, files changed, command failures, human edits, duration, and test pass
+  rate alongside the pass-rate summary.
+- **Spec verification / verifier kernel role.** `ur spec verify <name>` runs
+  deterministic project gates from `.ur/verify.json`, then invokes a read-only
+  deep verification subagent that must demonstrate compile/test/lint/diff/runtime
+  proof before emitting `VERDICT: PASS`. Results are persisted to
+  `.ur/specs/<name>/verification.md` and recorded in `spec.json` so `ur spec status`
+  shows the latest verdict.
+- **AgentKernel abstraction.** Added `src/services/agents/kernel.ts`, a pure
+  orchestrator with seven roles: planner, executor, verifier, critic, memory,
+  router, and guard. `ur spec run <name> --kernel` and
+  `ur spec verify <name> --kernel` route through kernel stages while the legacy
+  loop remains the default. `src/services/agents/kernelSpec.ts` provides
+  spec-to-stage adapters.
+- **Rich task decomposition.** Added `src/services/agents/decomposer.ts`. `ur crew
+  create|run|plan ... --decompose` splits a goal into atomic subtasks with
+  `goal`, `filesTouched`, `risk` (low/medium/high), `testsRequired`, and
+  `rollbackPoint` metadata. A deterministic fallback keeps it offline and fast;
+  the model path asks a headless subagent for structured JSON.
+- **Parallel specialized subagents.** Added `parallel` multi-agent pattern in
+  `src/services/agents/patterns.ts`. Bug finder, patch writer, test writer,
+  security auditor, and style reviewer run in parallel, then a synthesizer merges
+  the results. `ur pattern parallel "<task>" --execute` runs it via the workflow
+  executor with concurrency matching the number of parallel agents.
+
+### Changed
+- **`makeCliEvalRunner` resets cost state per case**, reads child metrics,
+  gathers `git diff --stat` totals, runs `expect.testCommand` when present, and
+  returns an `EvalRunMetrics` object with best-effort command-failure and
+  human-edit counts.
+- **Version bump.** Updated from 1.21.0 to 1.22.0 across `package.json`,
+  `bunfig.toml`, and the VS Code extension manifest.
+
+### Verified
+- Added `test/evalMetrics.test.ts` and `test/evalDashboard.test.ts` covering
+  child metrics serialization, report aggregation, dashboard HTML rendering,
+  HTML escaping, and run-metrics persistence.
+- Added `test/specVerify.test.ts` covering dry-run verification, gate fast-fail,
+  verification record persistence, and verifier prompt construction.
+- Added `test/kernel.test.ts` covering all seven kernel roles and the spec
+  run/verify kernel adapters.
+- Added `test/decomposition.test.ts` covering deterministic decomposition, risk
+  heuristics, LLM-driven dry-run decomposition, and crew metadata persistence.
+- Added `test/parallelPattern.test.ts` covering the `parallel` pattern stages,
+  workflow compilation, and dry-run execution through the workflow executor.
+
 ## 1.21.0
 
 ### Added

package/README.md

@@ -124,7 +124,7 @@ as first-class subcommands in the shipped CLI.
 | --- | --- |
 | `ur` | Start an interactive terminal agent session in the current workspace. |
 | `ur -p` | Run a non-interactive prompt with text, JSON, or stream-JSON output. |
-| `ur spec` | Create requirements, design, and task documents under `.ur/specs/`, then run the task list. |
+| `ur spec` | Create requirements, design, and task documents under `.ur/specs/`, run the task list, and verify with strict proof gates. |
 | `ur escalate` | Plan or run work with fast and oracle model tiers selected from local model capabilities. |
 | `ur arena` | Run multiple agents on the same task in isolated worktrees and surface the winning diff. |
 | `ur ci-loop` | Run a build or test command, hand failures to a fix agent, and retry with a bounded budget. |
@@ -135,7 +135,8 @@ as first-class subcommands in the shipped CLI.
 | `ur worktree` | List, inspect, and clean up UR agent worktrees. |
 | `ur automation` | Store and run project-local scheduled automation specs under `.ur/automations/`. |
 | `ur workflow` | Define, validate, graph, run, and resume declarative agent workflows. |
-| `ur crew` | Run a lead and worker subagent crew over a shared task board. |
+| `ur crew` | Run a lead and worker subagent crew over a shared task board; `--decompose` auto-splits tasks with risk/tests/rollback metadata. |
+| `ur pattern` | Run multi-agent collaboration patterns (PEER, DOE, concurrent, handoff, debate, parallel); `--execute` runs them as a workflow. |
 | `ur goal` | Track long-horizon objectives that persist across sessions. |
 | `ur repo-edit` | Build a repo edit index, plan AST-aware renames, preview patches, and apply with rollback. |
 | `ur code-index` | Build, query, or watch a local semantic code index using Ollama embeddings. |
@@ -144,6 +145,9 @@ as first-class subcommands in the shipped CLI.
 | `ur artifacts` | Capture diffs, test runs, notes, and review feedback under `.ur/artifacts/`. |
 | `ur claim-ledger` | Map generated claims to file, web, MCP, tool, or user sources. |
 | `ur browser-qa` | Validate and smoke-run browser QA replay fixtures. |
+| `ur eval run` | Run an eval suite and grade outputs; optionally capture cost/tokens/files/test metrics. |
+| `ur eval report` | Show a saved eval report; `--dashboard` writes a single-suite HTML timeline. |
+| `ur eval dashboard` | Generate the local-first HTML dashboard across all saved reports. |
 | `ur eval bench` | Import local SWE-bench, Terminal-Bench, or Aider Polyglot exports. |
 | `ur model-doctor` | Inspect Ollama models and report likely agent capabilities. |
 | `ur model-route` | Recommend a local model for a task by capability fit. |
@@ -168,6 +172,12 @@ Examples:
 ```sh
 ur spec init checkout --goal "1. add cart 2. add payment 3. add receipt"
 ur spec run checkout --all --dry-run
+ur spec run checkout --all --kernel
+ur spec verify checkout --kernel
+ur crew create parser-crew --goal "fix the flaky parser test" --decompose --dry-run
+ur crew plan parser-crew --goal "fix the flaky parser test" --decompose
+ur crew run parser-crew --workers 3 --decompose --dry-run
+ur pattern parallel "refactor login without changing behavior" --execute --dry-run
 ur arena "implement a debounce helper" --agents 2 --dry-run
 ur escalate run "refactor the cache layer" --force-oracle --dry-run
 ur test-first detect
@@ -187,6 +197,9 @@ ur artifacts capture-tests --command "bun test"
 ur agent-task pr --create --dry-run
 ur acp serve --port 8123
 ur exec "add tests for the parser" --concurrency 4 --json
+ur eval run starter --metrics --json
+ur eval report starter --dashboard
+ur eval dashboard
 ```
 
 ## Project Context
@@ -296,6 +309,7 @@ CLI. Rebuild it after source, version, or macro changes.
 - [Configuration](docs/CONFIGURATION.md)
 - [Agent Feature Expansion](docs/AGENT_FEATURES.md)
 - [Agent Trend Coverage](docs/AGENT_TRENDS.md)
+- [1.22.0 Upgrade Notes](docs/AGENT_UPGRADE_1.22.0.md)
 - [1.20.0 Upgrade Notes](docs/AGENT_UPGRADE_1.20.0.md)
 - [1.19.0 Upgrade Notes](docs/AGENT_UPGRADE_1.19.0.md)
 - [1.18.0 Upgrade Notes](docs/AGENT_UPGRADE_1.18.0.md)