wsr

Behavioral test coverage

flowchart LR
    S[Reviewed source scope] --> B[Behavioral tests]
    B --> E[Observed revision and result]
    E --> G[Limits and gaps]

Current status — November 5, 2026

The October 26 coverage work landed through PR #16 in merged baseline a64136113d0a42f20c96a585ead5cc76dcbe0a94. PR #17 added design documentation at 031b583e60ce93b6b0af3b41afdec02b393aedff. No ready failing implementation task is recorded. Product execution remains scaffold. Dates below identify observed evidence, not fresh test runs.

The Java checkpoint owns the current worker handoff under the Java contract. The earlier test-preparation-only mandate is superseded. Review live checkout/remote state and owner stop/hold instructions before selecting work.

October 26 review provenance

The product scope below was reviewed against preparation base faf418eb0afd4cc567ad9b69cd6414a88bde1c47. Preserve its observation dates and limits.

Reviewed product scope

Source scope Tested observable behavior and tests Focused command Status and limits
wsr-cli, including run/cache/hook/daemon arguments Parser tests: command tree, required arguments, run defaults/options/formats, rejection of invalid values/options, Unix OS-string preservation for inspect and run. Run regression also covers a leading-dash filename after --. cargo test -p wsr-cli --test cli --locked Passed during October 26 preparation; available on main through PR #16. New regression already passes existing code; no red phase. Unix path checks are conditional. No file access or execution coverage.
wsr entrypoint, binary and all eight handler modules Executable tests: all eleven operational paths exit 1 with exact placeholder diagnostic; help/version exit 0; invalid arguments exit 2; disposable repository sentinels remain unchanged. cargo test -p wsr --test cli --locked Passing October 26. Placeholder checks protect scaffold behavior only; they are not implementation acceptance tests for future operations, and do not prove isolation.
wsr corpus bridge Invokes Python harness regressions and offline baseline from an unrelated working directory. Catalog integrity, pinned snapshot digests, owned Python native pass/failure checks. cargo test -p wsr --test corpus --locked Passing October 26. Requires Python 3.11+ and Git. Catalog, fixtures, snapshots, and examples are supporting inputs. No upstream execution/download or WSR execution. Python harness internals were not exhaustively reviewed here.
wsr-gha scaffold and example guards All example YAML workflows have only manual dispatch and nonempty jobs; exact shell workflow script produces native pass/failure markers and exit status with explicitly bound mode. cargo test -p wsr-gha --test workflow_examples --locked Passing October 26. Requires POSIX sh. Uses canonical shell and Python YAML. No GHA expression evaluation, translation, WSR engine or isolation evidence.
wsr-types No behavioral tests present. Early structs/serde derives, trait and errors inspected. cargo test -p wsr-types --locked Build/doctest baseline only. Canonical schema and provider interface are not settled; avoid freezing legacy serialization as an accepted future contract.

The remaining twelve library entrypoints were inspected and contain documentation-only placeholders: wsr-bench, wsr-cache, wsr-client, wsr-engine, wsr-expr, wsr-fs, wsr-git, wsr-resolver, wsr-sandbox, wsr-shell, wsr-tracing, wsr-wasix. Their source is crates/<name>/src/lib.rs; no behavioral tests exist. Each can be checked with cargo test -p <name> --locked, but an empty passing suite establishes only build compatibility. No execution, resolution, filesystem, security, or performance claim follows. All are included in the workspace baseline below.

November 6 static inspection slice

cargo test -p wsr --test inspection --locked exercises the real binary against temporary source files: static summaries, deterministic job order and source step order, unreadable/malformed YAML, invalid/unsupported fields, duplicate identifiers, and unevaluated-expression rejection. A malicious run script does not execute; an action reference is not opened; repository sentinels and the file inventory remain unchanged. Remaining placeholder assertions continue in tests/cli.rs; inspection assertions are replaced by actual behavior, not removed without coverage.

Observed red: the first acceptance test failed on error: wsr inspect: not implemented yet. Subsequent red/green checks exposed dynamic runner acceptance, empty jobs, and missing manual-trigger support. Focused tests and cargo xtask ci passed after implementation. The October audit table above remains dated evidence; its eleven-placeholder and GHA-scaffold observations no longer represent this revision. See supported subset. This is not full GHA lowering, executable canonical IR, Component-provider or workload execution evidence.

Repository tooling and areas not reviewed

Private tools/xtask has two parser unit tests in cli.rs, run with cargo test -p xtask --locked. They cover canonical/legacy command parsing and tool-group defaults. Other xtask modules, Python tooling/scaffold regression scripts, release workflows and policy correctness have not received a behavioral audit in this run. The full baseline running their Cargo paths does not make them reviewed. Preserved prototype code/stash was not restored or tested.

Native hosted evidence remains dated reference evidence. No workflows were dispatched during this review. Canonical example YAML must be reused for future real translation tests and later execution tests; do not fork local workflow copies.

Historical October 26 implementation-preparation proposal

This section preserves the October 26 proposal and its then-current instructions. It is not the active worker handoff. At that review, ready failing tasks: none; pending failing branch/PR: none. PRs #12 and #13 contain dependency updates, not implementation handoffs. No additional speculative implementation suite should be accumulated.

The smallest next decision is to approve a first translation-only observable contract: which public entrypoint and result/diagnostic schema should represent one manual-dispatch workflow containing one sh run step? Recommendation: accept the canonical shell example as the positive input, retain its manual mode input/environment binding and source provenance, and reject unsupported constructs explicitly without running commands. Settle the minimal result fields, expression treatment and diagnostic expectations before choosing assertions. This is a recommendation, not an approved interface or ready task. The legacy WorkflowProvider/WorkflowIR declarations do not settle it; PLAN.md D-003 through D-005 and its implementation gate remain authoritative.

Once approved, prepare exactly one ordinary Rust integration test in wsr-gha/tests/ using that real entrypoint and canonical YAML. Record its exact command and observed behavioral assertion failure here and on its actual work branch/draft PR. Compilation/setup failures or generic CLI placeholder errors do not qualify. Non-goals: engine/runtime, execution, isolation, backend selection, plugin host, new framework and broad IR design. Follow the handoff contract, including prerequisite decisions, negative fixture, regression commands and completion evidence. The then-current scheduled test-preparation scope prohibited production implementation. The November 1 Java worker authorization superseded that scope for settled, bounded TDD slices.

Historical October 26 validation and resumption

October 26 review branch: cargo xtask ci passed (format, locked compilation/Clippy, workspace tests including corpus/native references, and strict rustdoc); git diff --check passed. The focused product suites above are included in that run. No failing implementation test is concealed, ignored or wrapped as expected failure. Hosted CI for the base main revision 40261792393 succeeded; branch CI is separate and must be checked on its PR before any merge. Auto-merge remains disabled.

At that checkpoint, the next-run instruction was to inspect the preparation branch/PR and preserve dirty work. Keep useful passing regression work bounded while the translator contract is unsettled. Update this index only for material changes, and do not repeatedly notify the same blocker.

Java reference input preparation — November 2, 2026

The manual preparation checkpoint records a complete pinned junit-jupiter-starter-maven source subtree, upstream license/workflow, verified GitHub commit/tree/blob identities and SHA256 manifest under ignored target/wsr-tests/java/manual-preparation-20261102/. Original input commit: junit-team/junit-examples@5bd6594a2336f1eff52527733ba28682a3215d83. An explicitly derived manual Maven workflow is review input only; action metadata is captured, not qualified under Bun. Parent Astra orchestrated a requested Sol fixture/docs delegation; see the checkpoint for assignment verification limits and preserved branch/base.

cargo xtask ci passed (exit 0); git diff --check passed. Raw logs/result records are in that ignored evidence directory. Read-only docker info failed (exit 1, absent desktop-linux daemon socket); no daemon was started. Native Java pass/fail, JAR/Surefire verification, WSR translation and execution, isolation and cache behavior remain unexecuted. Download/setup success is not behavioral red/green. Next command: docker info; qualification and exact image/JDK/Maven artifact verification must precede any upstream execution. No product behavior or test changed.

Linux Java reference prerequisites — November 3, 2026

The qualification checkpoint records an available Docker Desktop Linux ARM64 daemon and digest-pinned BellSoft JDK25 image. Ignored evidence target/wsr-tests/java/linux-qualification-20261103/commands.json preserves exact commands; image/container inspection, JDK hashes, probe/cleanup logs and CI results accompany it. Two disposable UID10001 containers verified bounded CPU/memory/PID settings, read-only root, zero capabilities, no-new-privileges/seccomp, no bind/volume/socket/credential mounts, denied outbound networking and fresh private tmpfs. Containers were removed; final filtered list empty. This is reference-environment prerequisite evidence, not a hostile-code security audit or WSR sandbox implementation. cargo xtask ci and git diff --check passed. Requested Sol delegation and runtime-verification limits are recorded in the checkpoint. Maven distribution/dependency verification, native pass/fail and JAR/Surefire evidence remain pending. No upstream Maven/script, WSR translation/execution, action/Bun, caching or product red/green behavior was exercised.

Maven reference inputs — November 4, 2026

The Maven preparation checkpoint records exact Maven 3.10.0 ZIP acquisition, matching upstream SHA512/SHA1, and a valid detached signature against HTTPS-fetched official Maven KEYS (independent signer trust not established). Ignored target/wsr-tests/java/maven-inputs-20261104/ retains input provenance, archive inventory, GPG command/results and the restricted dependency acquisition plan. No archive extraction, Maven/upstream script execution or dependency resolution occurred. Reuse of the qualified image-contained JDK is reference-only. Next prerequisite is qualification of restricted acquisition and container input delivery; native pass/fail and all Java/WSR product acceptance lanes remain pending.

Validation for this slice: cargo xtask ci exit 0 (format, locked compilation/Clippy, workspace tests and strict docs); git diff --check passed. Logs/results remain in the ignored evidence root. This is repository regression evidence only; no Java behavioral red/green claim.

Java offline dependency preparation — November 4, 2026

The dependency checkpoint qualifies validated source/Maven transfer and wrapper bootstrap inside the pinned Linux ARM64 reference container. All fetched workload execution remained nonroot and network-none, with explicit private settings/repository and prior resource restrictions; no host Maven ran. Explicit fixed Central HTTPS acquisition retained 51 checksum-checked POM/JAR artifacts (2,745,759 bytes). This is a partial inventory, not a resolved dependency closure.

Ignored target/wsr-tests/java/dependency-preparation-20261104/ preserves exact scripts/argv, per-attempt failures, artifact URLs/hashes, delivery-integrity checks, runtime identities and cleanup. Wrapper --version passed; final offline clean verify exited 1 during resources-plugin descriptor resolution for maven-filtering 3.5.0 and plexus-utils 3.6.0. The checkpoint identifies their reviewed missing parents and exact resume command. Compilation, tests, native pass/fail, Surefire/JAR and all Java/WSR product acceptance lanes remain pending; setup failure is not behavioral red.

CARGO_NET_OFFLINE=true cargo xtask ci and git diff --check passed. This is repository regression and reference-prerequisite evidence only.

Static dependency-inspection iteration

Real-binary acceptance tests in crates/wsr/tests/inspection.rs now cover scalar/list needs, no-edge output preservation, empty lists, forward references, chains, diamonds, disconnected jobs, repeated edges, unknown targets, self-cycles and longer cycles. A cycle’s downstream dependents are excluded from its reported path. Malformed types and expressions fail explicitly before stdout. A 4,096-job chain exercises nonrecursive traversal; hostile scripts/local action references and repository/config/hook/cache sentinels verify that graph validation does not execute or acquire inputs.

Observed red/green: the shared valid fixture first failed on unknown needs; missing-target, self-cycle and longer-cycle tests each failed because invalid graphs initially returned success, then passed after their corresponding checks. The old unknown-field assertion for needs is replaced by a semantic unknown-target assertion. All other existing inspection and placeholder assertions remain. cargo xtask ci passes with 13 inspection acceptance tests; this is static source/graph evidence, not execution, security enforcement or complete GHA compatibility.

Explicit step-identity iteration

The real CLI acceptance suite now has 17 inspection tests. New red/green checks first observed unknown id rejection, then acceptance of invalid/duplicate IDs, and passed after decoding, validation and duplicate checks were implemented. Tests verify IDs on run/uses steps, unchanged unidentified-step output/order, malformed types, expressions, syntax/reserved-prefix rejection, 99/100-character boundaries, same-job duplicate locations, cross-job reuse and case preservation. The shared tests/fixtures/inspection/step-identities.yml includes a malicious command and an unopened local action reference; source bytes/file inventory remain unchanged. Existing graph, CLI parser and remaining-placeholder regressions still pass. cargo xtask ci passed. This is static source/identity evidence, not output binding, expression evaluation or execution support.

Literal runner-label iteration

The real CLI suite now has 20 inspection tests. Observed red: a valid list failed with the old string-only decoder; after list support, an empty list incorrectly succeeded. Both went green. New assertions cover scalar/list output, ordered/repeated/case-sensitive labels, quoting and control characters, strict types, indexed blank/expression diagnostics, unsupported mappings, steps/needs integration and unchanged source/config sentinels. A GHA public-API integration test checks scalar versus list source-summary variants independently of rendering. Existing tests remain. These establish source inspection only, not runner matching, provisioning or execution.

Manual-trigger input iteration

25 real CLI inspection tests (plus 5 CLI regressions) and two new public parser-summary tests cover dispatch mappings/input declarations. Observed red/green: mapping rejected by old string-only parser; missing choice options accepted before cross-field checks; malformed identifiers accepted before validation. New cases include five types/omitted type, strict typed defaults, options, empty/null declaration states, deterministic input order, duplicate keys, expressions, unknown metadata/events, indexed diagnostics, escaped text and unchanged repository/source sentinels. No existing assertions were removed. Source summaries retain declaration states independently of rendering. Numeric defaults are preserved rather than dropped; this is not complete input validation or event binding. Existing examples still require other source fields before end-to-end acceptance.