verify-sunstack-rule
Use when verifying the Rust Sunstack engine, installed TypeScript SDK or sunstack runner, including lint, exec, hosted Jev, standalone builds and private release-install rehearsals. Not for Demistify or production hook installation.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- verify-sunstack-rule
- description
- Use when verifying the Rust Sunstack engine, installed TypeScript SDK or sunstack runner, including lint, exec, hosted Jev, standalone builds and private release-install rehearsals. Not for Demistify or production hook installation.
02 资源文件 64
- 参考文档references/cli-report-errors.md10.2 KB
- 参考文档references/config-only.md7.4 KB
- 参考文档references/container-matrix.md3.9 KB
- 参考文档references/decision-trees.md9.3 KB
- 参考文档references/evidence-selection.md13.5 KB
- 参考文档references/executable-operations.md10.4 KB
- 参考文档references/git-first.md9.0 KB
- 参考文档references/heading-text.md8.5 KB
- 参考文档references/hosted-decisions.md14.1 KB
- 参考文档references/html-evidence.md11.0 KB
- 参考文档references/html-sections.md16.6 KB
- 参考文档references/library-gates.md9.6 KB
- 参考文档references/lint-runner.md5.7 KB
- 参考文档references/lint-usability.md8.4 KB
- 参考文档references/locator-limits.md8.8 KB
- 参考文档references/markdown-sections.md10.9 KB
- 参考文档references/markdown-skill-rules.md2.8 KB
- 参考文档references/milestone-1.md11.0 KB
- 参考文档references/milestone-2.md17.8 KB
- 参考文档references/prepare-requests.md12.6 KB
- 参考文档references/release-rehearsal.md17.0 KB
- 参考文档references/rulebook-check-refusals.md8.8 KB
- 参考文档references/rulebook-lint.md18.8 KB
- 参考文档references/sdk-journeys.md77.4 KB
- 参考文档references/selection-cuts.md7.7 KB
- 参考文档references/since-encoding.md6.7 KB
- 参考文档references/unsure-summary.md5.0 KB
- 参考文档references/usage-scenarios.md13.9 KB
- 参考文档references/within-kinds.md9.2 KB
- 脚本scripts/composition_scenarios.py23.4 KB
- 脚本scripts/connected_scenarios.py16.9 KB
- 脚本scripts/container_matrix.py9.3 KB
- 脚本scripts/cursor_scenarios.py9.2 KB
- 脚本scripts/dependency_scenarios.py19.7 KB
- 脚本scripts/fixtures/usage-scenarios/claude-handoff.jsonl1.6 KB
- 脚本scripts/fixtures/usage-scenarios/cursor-seeds.json948 B
- 脚本scripts/fixtures/usage-scenarios/tool-payload.schema.json436 B
- 脚本scripts/git_scenario_fixtures.py9.9 KB
- 脚本scripts/handoff_scenario.py11.4 KB
- 脚本scripts/hook_scenarios.py6.7 KB
- 脚本scripts/lint_observations.py14.2 KB
- 脚本scripts/recent_scenarios.py18.6 KB
- 脚本scripts/release_assets_test.py3.4 KB
- 脚本scripts/release_assets.py6.0 KB
- 脚本scripts/runner_containers_test.py4.8 KB
- 脚本scripts/runner_containers.py5.7 KB
- 脚本scripts/runner_contracts_test.py24.0 KB
- 脚本scripts/runner_contracts.py9.0 KB
- 脚本scripts/runner_fixture_cases.py24.0 KB
- 脚本scripts/runner_fixture_setup.py4.5 KB
- 脚本scripts/runner_linkage.py2.7 KB
- 脚本scripts/runner_process.py8.1 KB
- 脚本scripts/scenario_checks.py11.2 KB
- 脚本scripts/scenario_checks.py.lock43.6 KB
- 脚本scripts/scenario_fixtures.py5.6 KB
- 脚本scripts/scenario_readback_checks.py19.1 KB
- 脚本scripts/scenario_runtime.py8.6 KB
- 脚本scripts/scenario_transport.py5.9 KB
- 脚本scripts/schema_agreement_test.py3.0 KB
- 脚本scripts/schema_agreement.py13.8 KB
- 脚本scripts/schema_fixtures_test.py3.6 KB
- 脚本scripts/schema_fixtures.py5.6 KB
- 脚本scripts/usage_scenarios.py6.1 KB
- 脚本scripts/usage_scenarios.py.lock34.2 KB
03 SKILL.md 正文
Verify Sunstack Rule
Verify a named target against the adopted Specification, then report only the outcomes observed there. A passing build, an HTTP success or an unfavorable rule verdict each means something different.
The current runner is sunstack: lint calls the sunstack-lint library and the evidence/rules/decisions engine; exec carries the engine protocol. Evidence owns capture and selection, decisions owns Jev transport and answer admission, rules owns applicability/grouping/verdicts, lint owns repository configuration and reporting, and the binary owns command dispatch, signals and exits. Historical routes below drive the earlier sunstack-rule process. There is no daemon; npm is an install/launch channel.
Choose the target and journey
- From the monorepo root, record
git rev-parse --show-toplevel,git branch --show-current,git rev-parse HEADandgit status --short. Record the declared product commit, build location, configuration and calling client. Confirm the driven checkout matches that product commit; the skill may be on a separate branch. Stop if the target changes during observation. - For historical replay of the legacy V/US routes, the adopted oracle is design revision
f259d4b28f5131478e574e123335904c78b0d347. Product target:592293254239d301c53cd25fcb5b106f21cbfb59. Read oracle files withgit show <design revision>:<repo-relative-path>, exceptdocs/specs/2026-10-12-rust-rule-engine/schemas/rule-invocation.schema.jsonanddocs/specs/2026-10-12-rust-rule-engine/schema-inventory.md, which use adopted schema revisionfc754c8335963bf809dad7472109a81dc29c3c2b. Start withgit show f259d4b28f5131478e574e123335904c78b0d347:docs/specs/2026-10-12-rust-rule-engine/specification.md, especially the selected V rows; use the design revision forprogram-design.md,program-design-contracts.md,program-design-connected-evidence.md, referenced examples and other schemas. Product design copies may be older and a design working tree may have moved. If a pinned object is unavailable, return that source gap; resolve conflicting expectations with the commissioning Lead before driving the affected case. The Markdown-rule route below takes its product target and adopted design from its commission and reference instead. - Select the route below and read it. A draft has an open target prerequisite and is not a passing observation; drive its first run only under a declared-target commission once its prerequisites are met.
SKILL.md owns target policy. A historical replay is runnable only at its recorded product/oracle contract and proves nothing about the current product. Keep its fixture vocabulary, source guards and receipts together. Current routes use error, warn, warn_above_band and error_at_or_below; old inputs occur only in refusal controls. Current U3 exec proof comes from the declared target's real workspace/process suite, not a historical replay. A shared transport helper has no severity policy.
| Surface | Route and driver | Product target | Oracle | Availability |
|---|---|---|---|---|
| Selection candidate/shared cut fields | references/selection-cuts.md; scripts/verify/sunstack-rule/selection-cuts/selection-cuts.ts |
declared signed0.16 candidate and released0.15.1; headed P3 plus J1/ai-tools | 0.16 selection-cut commission/Correction1; exact added vectors/hashes and unchanged rule reports | operable; candidate/shared metadata and rule parity exercised; repeat per target and join native/SDK proof |
| Native section heading labels and P3 paths | references/heading-text.md; scripts/verify/sunstack-rule/heading-text/heading-text.ts |
declared signed0.15.1 candidate and released0.15.0, fixed13HTML/10Markdown pages | 0.15.1 heading-text commission Corrections1–5; source-derived labels/spans and version-only additive parity | operable; source paths, own headings and legacy parity exercised; repeat per target and join native/SDK proof |
| Configured conventions and standard defaults | references/config-only.md; scripts/verify/sunstack-rule/config-only/config-only.ts |
declared signed0.15 candidate and released0.14.1; fixed actual J1/HTML/Markdown/ai-tools inputs | 0.15 config-only commission Corrections1–7; exact config snapshot exception and parser-located released witness | operable; configured/default signed CLI exercised; repeat per target and join native/SDK proof |
| Installed SDK codec, direct inspection and examples | references/sdk-journeys.md; scripts/verify/sunstack-rule/sdk/sdk-verify.ts (codec, inspection, examples) |
declared quiescent candidate and fresh pinned installed package/addon | commission's agent-usability specification and verification design | operable; exercised on 0.19/revision 13 with partial receipts; repeat per target and retain provider, in-flight cancellation and fault limits |
| Installed TypeScript SDK: load/edit, inline/native parity, cancellation, six-cell installation and ordinary provider | references/sdk-journeys.md; scripts/verify/sunstack-rule/sdk/sdk-verify.ts |
declared quiescent integrated head, fresh installed package/addon/tarball hashes and exact Node/OS/CPU/libc cell | 11 Nov TypeScript SDK R1–R15 and O1–O6, including native 0.14 HTML/Markdown units and absolute checked root | draft; apparatus sensitivity tests are not SDK observations; missing native, CLI, install and provider joins remain explicit |
| HTML asking, cross-document coverage and checked root | references/html-evidence.md; scripts/verify/sunstack-rule/html-evidence/{html-evidence,markdown-parity,checked-root}.ts |
declared signed0.13 candidate for asking/coverage; companion targets and comparators explicitly commissioned | 0.11 mechanics plus0.13 HTML asking projection; pinned real corpus and original consumer config | asking/coverage operable; declared-target pass and native joins required; companions need their own commissioned joins |
| HTML lexical sections and complete asking text | references/html-sections.md; scripts/verify/sunstack-rule/html-sections/sections.ts |
declared signed0.13 candidate, full head/digest; three fixed real HTML inputs | 0.13 HTML-sections commission Proof1–3/Corrections1–7; lexical h1–h6 outline | operable; recovered headings, original outlines and exact asking text exercised; repeat per target |
| HTML asking text and truthful incomplete fields | references/html-sections.md0.14.1; scripts/verify/sunstack-rule/html-sections/error-text.ts |
declared signed0.14.1 candidate and released0.14.0 comparator; frozen source tables and original projects | 0.14.1 commission Corrections1–13; source-derived text and exact report boundary | operable;43J1 cases, real arrow/ampersand pages,114sections,499requests, Markdown parity and unsafe cuts exercised; repeat per target |
| Markdown fragment anchors and section fields | references/markdown-sections.md; scripts/verify/sunstack-rule/markdown/{capture,anchors,github-samples}.ts |
declared signed0.14 candidate, full head/digest; fixed two-repository corpus | 0.14 commission/Corrections1–9; independent264link and74section oracle, actual GitHub IDs | operable; original corpus, sections, image-heading/ordinary raw-asking controls exercised; native/parity joins stay separate |
| Deferred document admission | references/html-evidence.md admission section; scripts/verify/sunstack-rule/html-evidence/document-admission.ts |
declared signed admission correction, full head and digest | Correction7; preserve existing Target diagnostics and local/plain Document positives | operable; keyless --check only, no selected dispatch |
| Public exec Choice255/256 boundary | references/locator-limits.md public exec section; scripts/verify/sunstack-rule/rulebook/exec-choice-limits.ts |
declared signed F1–F3 correction candidate, full head and binary SHA256 | Lead correction disposition; decisions-owned Choice cap255 | operable; both public decision API and filter native joins checked per target |
| Locator Choice cap and omission evidence | references/locator-limits.md; scripts/verify/sunstack-rule/rulebook/locator-limits.ts |
declared signed locator-limit candidate, full head and binary SHA256 | 6 Nov locator-limit commission/Lead disposition; nudge R2/C2; provider Choice maximum255 | operable; direct-SDK refusal/schema joins checked per target |
| Filters: tree routes and shared gates (O4) | references/rulebook-lint.md filters section; rulebook-lint.ts --case filters and rulebook-filters.ts |
declared signed decision-trees candidate, full source commit and binary SHA256 | decision-trees Specification/PD/VD at 0567445, O4 |
operable; confirm target and native joins per run |
| Changed-only capture, original input and raw-identity edges | references/git-first.md; scripts/verify/sunstack-rule/git-first/{git-first-original,git-first-edges}.ts |
declared signed0.12 main/test artifacts, full source head and digests; released0.11.1 comparator | 0.12 commission Proof1–4/Correction2 raw identity, prepared success6 | operable at5b4965ba; original/full oracle, declared native consumer and owned edges observed; repeat per target |
| --since beside equal binary assets | references/since-encoding.md; current route uses git-first-edges.ts; legacy rulebook-since-encoding.ts is0.10.1-only |
current declared0.12 artifacts or separately commissioned historical0.10.1/0.10.0 replay | raw binary/symlink/byte-budget contracts; Full versus ChangedOnly context is explicit | current coverage joins candidate-identity route; historical replay remains pinned |
| Rule-author usability and keyless empty scope | references/lint-usability.md; scripts/verify/sunstack-rule/usability/probe.ts |
declared signed0.12.1 candidate, full head/digest; released0.12.0 and detached probe input | 0.12.1 usability commission U1–U6/Corrections1–4 | operable at245eb9c3;8probe rows/28CLI children,10native boundary controls and exact prepare-report parity observed; repeat per target |
| Human unsure summaries and output preservation | references/unsure-summary.md; rulebook-unsure-summary.ts |
declared signed candidate and immutable comparator bytes/version/provenance | 8 Nov 0.10 unsure-summary commission; current CLI human-output contract | operable; strict same-version and released-version comparison are separate runs |
| Short output, saved report and cases (O5/O6) | references/decision-trees.md; rulebook-report.ts and rulebook-cases.ts |
declared decision-trees source commit and signed binary SHA256 | decision-trees Specification/PD/VD at 2fea595 (exact ranges and distinct owner-ruling) |
operable; scripted journey, keyed run separate |
| Completed cases readback and change-scope error paths | references/cli-report-errors.md; scripts/verify/sunstack-rule/cli-reports/{show-reports,change-scope-errors}.ts |
declared signed0.11.1 candidate, full head and SHA256; frozen real R03 report | 0.11.1 commission Outcomes1–3/Corrections1–2; completed candidate union versus preserved prepared ordinals | C1 observed6fd3aaaa (37 CLI invocations); C2 show observeda59cf64e (22); unchanged scope reused with explicit dependency comparison and fresh native joins |
| Current nudge admission refusals | references/rulebook-check-refusals.md; rulebook-check-refusals.ts --case nudge |
declared signed current source commit and binary SHA256 | 31 Oct S2 plus decision-trees 2fea595 and Lead identical-versus-conflicting locator disposition |
operable; two acceptance controls and ten zero-request refusals; join native identical-reuse positive |
| Rules-as-nudges report: U1-U10 | references/rulebook-lint.md and rulebook-check-refusals.md, both --case nudge; rulebook-nudge-compatibility.ts for U10 |
declared signed 0.6.0 release build; signed 0.5.0 artifact for comparison | 31 Oct engine Specification/VD accepted at a69fe47; locator-order clarification 6328647 |
S1/S2/U10 observed at ffd86f5; native S3 joins required; keyed S4 separate |
| Prepare exact requests without dispatch | references/prepare-requests.md; scripts/verify/sunstack-rule/prepare/prepare-requests.ts |
declared 0.9 source and immutable binary SHA256 | 7 Nov prepare commission, Admission, Addition 2, F1/F2 and whole-view pending policy | historical fixture observation at 73d7fc78; every current target must drive F1/F2/projection/replay-allowance controls; paid comparison separate |
| Rust quality/native consumers | references/library-gates.md |
commission-declared committed source | commission-declared contract | scoped source/test proof; historical receipts need applicability |
| Within a union of kinds: amended U14/U16, W6 | references/within-kinds.md; selection-mode lint/refusal entries and scripts/verify/sunstack-rule/selection/within-kinds-equivalence.ts |
declared within-kinds frozen source/binary; frozen v0.3.0 for one-item comparison | amended R30/R40 and U14/U16 at 9dfdb10 + a4696af + 2198c47, W1–W6 and path-policy amendments |
c3 WI2/local paths observed at 4086bcc1; c1 union/W6 reused with applicability |
| Evidence selection: U13–U17, R42 | references/evidence-selection.md; rulebook lint and check-refusal entries with --case selection; temporary compatibility probe |
commission-declared S3 source/binary; npm 0.2.0 plus frozen kit for R42 | 30 Oct evidence-selection Specification and Verification Design | observed locally at 7315762, native concurrency addendum 5d7c070; live V21 separate |
| Rulebook lint: U2/U5/U8/U10/U12 | references/rulebook-lint.md, repo scripts/verify/sunstack-rule/rulebook/rulebook-lint.ts and rulebook-boundaries.ts |
declared stable S3 head and local binary | 29 Oct skill-rules Specification and Verification Design, including Lead's F1–F3 proof clarifications | observed locally at 42f40e6; native joins required at each target |
| Rulebook check refusals: U9 | references/rulebook-check-refusals.md, repo scripts/verify/sunstack-rule/rulebook/rulebook-check-refusals.ts |
same declared stable S3 head and local binary | same 29 Oct rulebook oracle, R24 inventory | observed locally at 42f40e6; structural check call-path join required at each target |
| Historical V11 / V22 / V34 hosted | references/hosted-decisions.md native sections |
historical product pin above | historical design/schema pins above | historical replay; explicit hosted grant still required |
| Historical executable V2/V13 | references/executable-operations.md |
historical product pin above | historical design/schema pins above | historical replay only |
| Historical skill/handoff V14/V29/V30 | references/milestone-1.md |
historical product pin above | historical design/schema pins above | historical replay; proof remains verbatim |
| Historical Git/connected V rows | references/milestone-2.md |
historical product pin above | historical design/schema pins above | historical replay only |
| Historical usage US routes | references/usage-scenarios.md, scripts/usage_scenarios.py |
historical product pin above; enforced by driver | historical design/schema pins above | local replay; genuine slots remain disabled |
| Historical Markdown U1–U9 | references/markdown-skill-rules.md, repo scripts/verify/sunstack-rule/ |
reference's declared S2 7c529b4b / S4 bdca8d9a |
reference's adopted Markdown revisions | pre-runner replay, old binary/readiness; no current exec claim |
| S1 lint runner | references/lint-runner.md, repo scripts/verify/sunstack-rule/rulebook/lint-runner-rulebook.ts |
declared local S3 rulebook binary | 29 Oct rulebook oracle, retained runner formats/exits/scopes | observed locally at 42f40e6; legacy RuleSet fixture/schema helpers are historical proof |
| S2 standalone containers, U8/U9 | references/container-matrix.md, scripts/container_matrix.py |
declared four release artifacts and SHA256s | same runner oracle | historical fixture shape; rulebook upkeep and a declared artifact are required before a current offline matrix |
| S3 genuine static Linux Jev, rulebook U6 | references/hosted-decisions.md S3, repo scripts/verify/sunstack-rule/hosted/hosted-lint.ts |
declared static Linux arm64 release artifact/SHA256/version/commit | 29 Oct rulebook U6 plus retained no-CA runner contract | draft; separately commissioned paid command |
| S4 private release installs, U2/U8/U10/U11 | references/release-rehearsal.md, scripts/release_assets.py and repo scripts/verify/sunstack-rule/release/ drivers |
two declared normal-version private releases; exact scoped package/source/bytes | all-channel stream/stop oracle 8cf9c51 or commission-declared successor |
draft until every channel/platform cell observed; private registry/CI tap, public npm publishing excluded |
IF source/native tests or Rust quality gates are in scope, load references/library-gates.md and return the scoped commands, results and remaining coverage. Artifact-only observations do not launch an unrelated Cargo suite.
IF installed TypeScript SDK behavior is commissioned, load references/sdk-journeys.md and return each selected O row's installed consumer, exact artifact/cell identity, actual requests/outcomes and required native/CLI/install joins. Do not claim SDK coverage from the existing CLI npm channel, source-only prototype, apparatus tests or a cross-compiled addon. Provider execution needs its own explicit grant.
Current rulebook journeys run directly with node <driver>.ts; Zod validates external data. No Python wrapper or runtime is used for these routes. Their apparatus checks, from the monorepo root, are pnpm exec tsc -p scripts/verify/sunstack-rule/tsconfig.json, pnpm exec oxlint --type-aware --type-check --tsconfig scripts/verify/sunstack-rule/tsconfig.json --max-warnings 0 scripts/verify/sunstack-rule/**/*.ts, pnpm exec oxfmt --check scripts/verify/sunstack-rule/**/*.ts scripts/verify/sunstack-rule/tsconfig.json, and pnpm exec vitest run scripts/verify/sunstack-rule/**/*.test.ts, and node --test scripts/verify/sunstack-rule/release/*-test.ts. Historical Python drivers below are retained for their separate migration phase.
IF historical hosted V11 replay is in scope, load references/hosted-decisions.md and return its run identity, validated completion and receipt, or its access/setup gap.
IF historical genuine V22 or V34 replay is in scope, load references/hosted-decisions.md and return the hosted-rules observation or applicable reused evidence, native answer/receipt checks and any remaining gap.
IF historical executable V2/V13 replay is in scope, load references/executable-operations.md and return the operation coverage, process evidence and remaining proof limits.
IF historical milestone 1 replay is in scope, load references/milestone-1.md and return each local/hosted journey's report, measurement or named prerequisite.
IF historical change-scope or connected-evidence replay is in scope, load references/milestone-2.md and return each scoped V5/V7/V16–V18 claim and the V20 related-units portion as supported, unsupported or unknown, with observed/reused/draft coverage, target and evidence path, per the Evidence, cleanup and return rule below.
IF historical usage-scenario replay is in scope, load references/usage-scenarios.md and return each selected US ID's supported/unsupported/unknown verdict, observed/reused coverage, relevant V rows, target, process/HTTP evidence and remaining gap, using the return rule below.
IF retained Markdown/native proof or reconstruction of a historical marker replay is commissioned, load references/markdown-skill-rules.md and return each retained proof home and applicable current-target run, or the historical source/target prerequisite. Retired marker drivers do not run against the current rulebook target.
IF current changed-only --since or default uncommitted capture is in scope, load references/git-first.md; return original-command preparation, exact full-oracle owners/requests, public-API raw paths/acquisition counts, edge/default controls and readonly-input guards or named gaps.
IF lint --since capture or unchanged binary assets are in scope, load references/since-encoding.md and return changed-piece selection, selected-encoding refusals, binary-free byte parity, native budget/snapshot joins and original-input observations or named gaps.
IF HTML asking ERROR recovery is commissioned, load references/html-sections.md's0.14.1 route and run html-sections/error-text.ts; return the43-case real source scan, original114-section, Markdown-anchor and ai-tools499-request parity, ampersand preservation, unsafe cuts, exact allowed diffs and native joins or named gaps.
IF HTML heading sections, own/enclosing text, diagram attributes or element boundaries are commissioned, load references/html-sections.md; return the three pinned outline comparisons, prepared/cut controls and native parity/admission joins or named gaps.
IF Markdown fragment anchors or markdown_units section fields are commissioned, load references/markdown-sections.md; return every original link state, actual rendered GitHub samples, real/container section spans, admission refusals and native/parity joins or named gaps.
IF HTML evidence, cross-document ID checks or explicit checked-root behavior is commissioned, load references/html-evidence.md; return each real R/E owner, prepared text/raw span, fragment result, root refusal and absolute-path control, exact Markdown comparator/input identities and remaining native or provider gaps.
IF keyless empty-scope lint, config-open diagnostics, status summaries, named scope, help or prepared-output usability is commissioned, load references/lint-usability.md; return the8real probe outcomes, zero/dirty eligibility controls, prepared request/path summaries, unchanged JSON reports and native no-evidence joins or named gaps.
IF current lint behavior is commissioned, load references/lint-runner.md and return the selected case counts, finding/exit comparisons, target identity, native-proof joins and remaining gaps.
IF locator option limits or omitted-location reporting are commissioned, load references/locator-limits.md and return exact255-boundary, wire criteria, reached-result evidence and native refusal/schema joins at the declared target.
IF default human unsure summaries or cross-version output preservation are commissioned, load references/unsure-summary.md; report per-rule counts/pointers, all-format/report byte comparisons, every result ID readback, exact version exceptions and comparator provenance.
IF completed cases-report readback or change-scope refusal paths/limits are commissioned, load references/cli-report-errors.md; return real R03 unique/case/ambiguous lookups, preserved ordinary/prepared details, offending path fields in human/JSON/SARIF, configured limit values and native SourceChanged joins or named gaps.
IF decision-trees output, saved reports or case runs are commissioned, load references/decision-trees.md; report each O row with its declared target, actual output and open gaps. Keyed/model calls need a separate commission.
IF filters or only_when are commissioned, load the filters section of references/rulebook-lint.md and return the exact folder/request/result/skip comparisons, process exits, candidate identity and native joins or gaps. The scripted answers establish mechanics, not classification accuracy.
IF evidence selection, containment, candidate reports or 0.2.0 compatibility is commissioned, load references/evidence-selection.md and return U13–U17 plus R42 coverage, exact candidate/request/count comparisons, native joins and remaining live/target gaps.
IF amended R30 list containment is commissioned, load references/within-kinds.md and return union/once-only and not-selected fallback, every new R40 refusal/listing, W6 equality with planted rejection, and native schema/plan proof or gaps.
IF the 29 Oct rulebook evidence, decision or change-scope journey is commissioned, load references/rulebook-lint.md and return each scoped U row's observed/reused evidence, exact target and native joins or gaps.
IF the 29 Oct rulebook check/refusal journey is commissioned, load references/rulebook-check-refusals.md and return clean acceptance, every executed refusal's diagnostic/exit, zero-request/FIFO observations and structural check call-path join or gaps.
IF standalone portability is commissioned, load references/container-matrix.md and return each distro/architecture cell, exact binary linkage/hash, command exits and remaining macOS/Cargo-feature proof.
IF the current no-CA hosted judgment is commissioned, load references/hosted-decisions.md S3 and return its genuine verdict, typed answers/evidence/usage, exact target and safe evidence or setup gap.
IF private install/release behavior is commissioned, load references/release-rehearsal.md and return two normal-tag artifact/job/tap evidence, registry-resolved local-bin/exec-script/global/bare-npx/exec-npx/pnpm-dlx and native-channel bytes/version/report/exit/signal transcripts, fresh-runner coverage, real Actions exec/plain cancellation records and unobserved cells.
For every selected route, check its prerequisites and readiness, drive only commissioned actions, compare the deciding outcome, preserve evidence, stop owned resources and report supported/unsupported/unknown. A prepared fixture or missing binary is never a product pass. A draft route may be driven in a separately commissioned first observation; it becomes observed only for the target and cases actually run.
For the historical selected usage scenario, run the selector from the monorepo root. The reference owns the inventory and prerequisites; repeat --journey only for requested IDs:
env -u PYTHONOPTIMIZE -u UV_INDEX -u UV_EXTRA_INDEX_URL -u UV_INDEX_URL -u UV_FIND_LINKS -u PIP_INDEX_URL -u PIP_EXTRA_INDEX_URL -u UV_DEFAULT_INDEX -u UV_CONFIG_FILE -u UV_ENV_FILE \
UV_CACHE_DIR=<authorized-evidence>/uv-cache PYTHONDONTWRITEBYTECODE=1 \
uv run --no-config --no-env-file --no-python-downloads --default-index https://pypi.org/simple --locked --script \
.agents/skills/verify-sunstack-rule/scripts/usage_scenarios.py \
--binary <declared-binary> --evidence <authorized-evidence> --journey <US-ID>
Build and readiness
Run Cargo from the monorepo root with the pinned toolchain in rust-toolchain.toml. On macOS this host needs a command-local SDK correction for C dependencies:
SDKROOT="$(xcrun --show-sdk-path)" CC=/usr/bin/clang CFLAGS="-isysroot $(xcrun --show-sdk-path)" cargo build --workspace
Use that same prefix on every Cargo command shown in the references on this host. Do not change global compiler settings. Other platforms use their declared native toolchain; they are separate, unobserved targets.
On Apple Silicon, .cargo/config.toml routes Cargo/nextest execution through scripts/cargo-debug-signing-runner.sh. It gives the app and tests stable, separate debug identities. Signing is best effort: a warning means the original binary still runs and LuLu may require approval. Record that condition; do not replace the runner or relax gates.
For current lint, usable readiness is a validated fixture lint --check followed by the expected real finding/exit; --version alone proves identity only. The current process registers signals with sunstack: ready. For historical routes, keep their old readiness string.
Read-only health check: compare the target identity again, confirm the selected harness and fixtures exist, and check prerequisites without displaying credentials. The local test result must contain nonzero executed tests. For historical protocol replay, the CLI's sunstack-rule: ready line witnesses signal registration; usable readiness requires a completed operation_result envelope with the expected outcome and actual exit, not --help or the ready line alone. V11 requires a completed validated public-API decision run with receipt. Before historical hosted CLI replay, run the signing preflight in references/milestone-1.md; afterwards inspect every hosted child's stderr.log for the runner's runs unsigned warning and apply that route's host signing/egress classification. That warning cannot stop the child's dispatch, and --fail-fast does not act on it.
Evidence, cleanup and return
Keep run evidence in an assignment-authorized directory under system temp, conventionally yyyy-mm-rule-engine-verify-<run-id>/. Record the absolute directory on the work thread. Store the exact command, working directory, product/design revisions, configuration, exit code, test counts/run ID, raw output and safe receipts. For process runs keep stdin fixtures, stdout JSON and stderr separately, plus report/cursor readbacks. Use synthetic/public fixtures: results may contain selected evidence.
Supply OPENROUTER_API_KEY from the owner's secret store only for the hosted command. Keep the value in process memory; disable shell tracing. Never put a credential value, its location or a secret reference into this skill, artifacts or reports. A missing credential or a skipped hosted test leaves the claim unknown. Controlled HTTP proves transport behavior and leaves hosted inference open.
Preserve failures and LEAK counts before any rerun. Interpret LEAK markers only through the disposition in references/library-gates.md; the gate remains nextest exit 0. Product failures go to the paired Implementer; disputed expectations and target changes go to the Lead. Repair only assigned verification apparatus.
These commands start no persistent service. Await each owned test/process; its harness owns fixture/server teardown. Stop only a process this run started, using its recorded PID/session, and never by name. Leave shared Cargo caches and other sessions alone. Retain evidence; delete only owned disposable files when authorized, with directory removal requiring explicit permission. After cleanup, reopen the receipt/log and verify it survived.
Return each scoped V/U claim as supported, unsupported or unknown, with observed/reused/draft coverage, target, command/results and evidence path. Reuse earlier evidence only when its product source, dependencies, configuration and exercised route still match. Name all stand-ins, unobserved entry points and draft prerequisites. This recipe reports evidence; the Lead owns acceptance, and no release or production hook claim follows.
04 同仓库更多 Skills
smoke-fixture
暂无简介
demistify-explainer
Use when writing, editing, or checking Demistify markup: component syntax, attributes, nesting, layouts, Markdown/Mermaid islands, and data references.
examples
暂无简介
verification-design
Use when planning how a design project's requirements will be proven: for each requirement, the reference, the target and `verify-<app>` journey, a passing condition a wrong build fails, reused real-path tests and what stays unverified, after Program Design or when that plan is missing or stale. Not for observing or challenging a delivered claim (practices-verification), building journeys (practices-verification-maintenance), or regression tests.
practices-retrospective
Use when asked for a blameless retrospective, a look back, or lessons from a finished or bounded session or delivery: what helped, what hindered, what to preserve, or why an owner or review correction keeps recurring across sessions. Not for re-anchoring a shared model in the current conversation, diagnosing a current failure, collecting evidence alone, choosing which skills to create or merge, or maintaining a verification skill.
practices-research
Use when a task needs source gathering, prior-art research, current docs or web evidence, or memory, session-log, or saved-reader research before design, planning, review, or discussion can continue. Not for extracting unwritten owner meaning.