statelyai 0.14.3 → 0.15.0

package/README.md

@@ -26,6 +26,10 @@ npm install --global statelyai
 
 | Goal                                | Command                                              | Writes                                                  |
 | ----------------------------------- | ---------------------------------------------------- | ------------------------------------------------------- |
+| Calibrate reviewed behavior | `statelyai behavior calibrate --config config.json --output .scratch/calibration` | Native setup receipt awaiting inspected adoption |
+| Run deterministic behavior evidence | `statelyai behavior check --config config.json --output .scratch/run` | Fresh local run artifact; checked baseline unchanged |
+| Adopt inspected red or green evidence | `statelyai behavior adopt .scratch/run --config config.json` | Immutable checked state and exact config baseline |
+| Test an application against a model | `statelyai test ./feature.campaign.mjs` | Local code execution, report and optional replay fixture |
 | Process a saved verification report | `statelyai verify --from-report report.json`         | Local report file                                       |
 | List local machines                 | `statelyai scan`                                     | Nothing                                                 |
 | Edit local machines visually        | `statelyai open [file]`                              | Local source, when you save in the editor               |
@@ -40,6 +44,14 @@ npm install --global statelyai
 | Compare two machines                | `statelyai diff <source> <target>`                   | Nothing                                                 |
 | Fail on differences in CI           | `statelyai diff <source> <target> --fail-on-changes` | Nothing                                                 |
 
+## Test application behavior
+
+<!-- trusted local campaign command from src/testCommand.ts -->
+
+`statelyai test ./feature.campaign.mjs` runs graph paths and property testing through `@statelyai/model-testing`. Install that package in the campaign project. The campaign is trusted JavaScript with ordinary local permissions. The child process is not a sandbox.
+
+Use `--report report.json --fixture regression.json` to save results and a concrete counterexample, then `--replay regression.json` after fixing the app. Exit codes are 0 for passed, 1 for a model/app discrepancy, and 2 for inconclusive execution. See [Test application behavior](../../docs/cli/test.md) for budgets and replay, or the [runnable browser example](../model-testing/README.md#verify-the-browser-example).
+
 ## Quick start
 
 <!-- project initialization and sync flow from InitCommand, LinkCommand, PushCommand, PullCommand, and projectConfig.ts -->
@@ -282,6 +294,13 @@ safely and require manual migration.
 
 <!-- command names, arguments, flags, defaults, and behavior from src/cli.ts#COMMANDS -->
 
+### `behavior calibrate` / `behavior check` / `behavior adopt`
+
+`check --config <path> --output <fresh-directory>` runs the selected tests and
+writes an evidence artifact. `adopt <artifact-directory> --config <path>`
+validates inspected red/green evidence and advances the exact checked baseline.
+See [Behavior evidence in CI](#behavior-evidence-in-ci) for setup and exit codes.
+
 ### `login`
 
 Store an OAuth credential or API key.
@@ -573,6 +592,63 @@ See [Verification reports](../../docs/cli/verify.md) for offline JSON/JUnit
 processing and the programmatic executor hook. Live verification requires a host
 adapter; no verification engine is bundled.
 
+## Behavior evidence in CI
+
+<!-- behavior check/adopt commands, config, statuses and artifacts from src/behaviorCi.ts -->
+
+Connect existing Vitest tests with the repository's
+[setup-behavior-ci skill](../../.agents/skills/setup-behavior-ci/SKILL.md).
+The skill discovers real test names, writes a reviewed specification and model
+estimate, selects source/test inputs, and generates a CI workflow. Users do not
+author XState or maintain hashes. CI runs ordinary checked commands without an
+LLM, authentication, or automatic Git writes.
+
+```sh
+statelyai behavior calibrate --config .behavior-ci/feature/config.json --output .scratch/calibration
+statelyai behavior adopt .scratch/calibration --config .behavior-ci/feature/config.json
+statelyai behavior check --config .behavior-ci/feature/config.json --output .scratch/behavior-run
+statelyai behavior adopt .scratch/behavior-run --config .behavior-ci/feature/config.json
+```
+
+`check` captures Git revision/dirty bytes, selected input hashes, runner identity,
+command argv, logs, raw native report and exact model/record references. It copies
+the checked public FileStore into the fresh output directory and evolves only
+that copy. First setup uses `baseline: null`; stage generated inputs before
+checking, inspect output, then explicitly adopt and commit the checked state.
+
+Exit codes describe **this run**. `0` means all retained active checks passed;
+`1` means a mapped test failure, skipped or missing evidence; `2` means invalid
+configuration, process, provenance, reporter or artifact evidence. Missing or
+duplicate mappings, unknown statuses, unmapped selected cases in schema 2, changed source,
+and exit/report disagreement cannot pass. Setup maps every selected native
+case, including existing tests. A native test failure can be an assertion, runtime exception or timeout; inspect its detail
+before diagnosing the cause. Historical failures remain readable
+and do not make every future run fail. Omitting a requirement/check retains it;
+removal requires an explicit retirement reason.
+
+`summary.md` shows current verdict, Git/model/spec/implementation bindings,
+changed and retained obligations, failures and historical core gaps. `run.json`,
+`manifest.json`, `process.json`, raw report/logs and `store/.statelyai` preserve
+the evidence. Producers may write optional diagnostics under `$STATELY_BEHAVIOR_OUTPUT/traces`; the child receives a unique absolute output path. Trace files remain in red/green artifacts and are checksum-validated on adoption. Invalid setup still writes a minimal error artifact when the
+output directory can be safely allocated. Existing/populated output is refused.
+
+`adopt` accepts inspected red and green artifacts against their exact checked
+parent. It validates checksums, record/model digests, selected Git input bytes
+(or unchanged local dirty bytes), path safety and immutable version conflicts,
+then updates the config baseline after copying files. Repeating the same
+adoption is idempotent. Divergent parents and error artifacts refuse adoption.
+For schema 2 it verifies native normalization with the current trusted repository adapter. It never executes artifact code. Download and authenticate remote run artifacts
+before adoption; checksums are corruption detection, not runner authentication.
+
+Schema 2 adds a cited discovery inventory for seven input categories and a checked native calibration receipt. Checked authoritative intent files are contract-bound, so requirement-document edits invalidate calibration even before the structured spec changes. Initial calibration executes against the unchanged app; unresolved mismatches block adoption. A cited implementation-defect disposition can retain a known red baseline without changing its verdict. Changed model/spec/plan/mappings, adapter, test producers or runner identity require inspected recalibration. Ordinary app implementation changes retain the calibrated contract. Receipt snapshots preserve original revision evidence. Selected-check gaps remain visible; generated paths and executed cases are separate evidence. Schema 1 remains compatible and reports setup assurance unavailable.
+
+Config and the single module-adapter protocol are documented in the
+[setup contract](../../.agents/skills/setup-behavior-ci/references/contract.md).
+Source scope is bounded to 200 files including checked state. It is a selected
+dependency scope, not complete transitive coverage. Core execution remains
+reported; review identity is unavailable. CI requires clean bound inputs and
+actual HEAD equal to `GITHUB_SHA`; environment metadata alone is not remote proof.
+
 ## Self-hosting
 
 <!-- self-host auth and URL behavior from LoginCommand, baseUrlFlags, OpenCommand, credential resolution, and projectConfig.ts -->

package/dist/bin.mjs

@@ -1,5 +1,5 @@
 #!/usr/bin/env node
-import { C as run } from "./cli-D8zy3z4n.mjs";
+import { C as run } from "./cli-hfhCVqEx.mjs";
 
 //#region src/bin.ts
 run();