Sley Simple Tools

Primitive Sley examples: seventeen native graph_authoring.v0 packages with bound portable artifacts and independently expected cases. These are primitive examples of single operations and small compositions — not substantial end-user tools. They exist to pin the verified 2.0.1 authoring baseline: one readable recipe per program, one executable artifact, and a round-trip check each.

This repo complements the adoption toolkit in sley-lang/tools (Rust sley-tools CLI, Python/TypeScript/MCP SDKs, Playground, Repair Lab, Data Pipe, importers, Review). That toolkit is the machinery; this repo holds small native programs built with it. It does not duplicate the N01–N10 catalog (release-readiness, resource-admission, benchmark trio, tier-classifier, range-predicate, bounded-clamp, quota-remaining, generated-payload-validator).

Corpus: 30 native examples — the seventeen 2.0.1 primitives below plus thirteen 2.0.3 compositions in composed/. Version-pinned authoring inputs, readable views, executable artifacts, boundary checks, and one worked modification per composed example. Counts, kept as separate categories: 233 external assertions across the corpus (78 primitive + 155 composed) and 155 advisory workbench checks in the composed collection. Native test-admitted maintenance remains unavailable through the documented route (PENDING_CORE_EXECUTOR / NOT_RUN); the split code-then-tests persistence does not close that limitation.

Primitive examples (17 total, 78 external assertions)

ID	Program	What runs natively	Cases
S01	programs/bool-and	bool_and on two Bool	4
S02	programs/bool-nor	bool_or + bool_not	4
S03	programs/uint32-minimum	greater_equal vs minimum 5	5
S04	programs/is-negative	less_than vs 0	5
S05	programs/equals-answer	equal vs 42	5
S06	programs/safe-add	int_add_checked → {sum: CheckedSInt64}	5
S07	programs/bool-status	bool_and/bool_or/bool_not → {both, either, neither}	4
S08	programs/is-positive	greater_than vs 0	5
S09	programs/safe-sub	int_sub_checked → {difference: CheckedSInt64}	5
S10	programs/bool-or	bool_or on two Bool	4
S11	programs/safe-mul	int_mul_checked → {product: CheckedSInt64}	5
S12	programs/safe-neg	int_neg_checked → {negated: CheckedSInt64}	5
S13	programs/not-equal	not_equal vs 42 → Bool	5
S14	programs/at-or-below	less_equal vs 0 → Bool	5
S15	programs/at-or-above	greater_equal vs 0 → Bool	5
S16	programs/bool-not	bool_not → Bool	2
S17	programs/wrap-ok	result_ok → {wrapped: CheckedSInt64}	5

Each package holds build/graph.json (authoring recipe), tests/cases.json (independently expected), program.json (manifest bound to the artifact by sley-tools artifact export --bind-manifest), and artifact/native-graph.json (portable native graph). The accepted SSMC1 graph is the program; the recipe is only the authoring input. Sley has no canonical source text. Later cores also accept optional text authoring (the familiar frontend from 2.0.5, the Python dialect from 2.1.1), which compiles to the same graph; these examples do not use it.

Toolchain pin

• Core: Sley 2.0.1 (sley-2.0.1-linux-x86_64.tar.gz, commit c748dda), binary sha 90f48df4….
• Builder: sley-lang/tools sley-tools backend, graph_authoring.v0.
• External execution assertions: 78 total across the seventeen programs, all PASS at build time; every artifact reimports into a fresh workspace and re-passes.
• Native test admission: PENDING_CORE_EXECUTOR on the selected 2.0.1 core (NOT_RUN: production native executor unavailable). This is reported separately and is not covered by the assertion counts above.

Rebuild everything

./scripts/rebuild.sh \
  --tools /path/to/sley-tools \
  --core /path/to/sley-2.0.1-linux-x86_64/bin/sley

The script re-runs init + test + bound artifact export + fresh-workspace artifact import + re-test for all seventeen programs and refreshes program.json and artifact/native-graph.json in place. It refuses to run with uncommitted changes unless --allow-dirty is passed.

License

This repository is licensed under the Apache License, Version 2.0. That covers the original examples, documentation, supporting scripts, and project-owned artifacts, except where another license or notice is expressly identified. See NOTICE for attribution.

Composed examples (C01–C13, sley-agent 2.0.3)

Thirteen composed programs authored with the official 2.0.3 sley-agent workbench, kept in composed/ strictly separate from the seventeen 2.0.1 primitives above. The 2.0.1 pin, the seventeen packages, and the rebuild script are unchanged by them.

ID	Example	Function	Workbench checks / External
C01	composed/retry-decision	retry_decision → Result<RetryState,RetryError>	11 / 11
C02	composed/invoice-line-total	line_total → Result<i64,InvoiceError>	12 / 12
C03	composed/duration-breakdown	split_duration → Result<(d,h,m,s),DurationError>	9 / 9
C04	composed/page-window	page_window → Result<(start,count,has_more),PageError>	16 / 16
C05	composed/ledger-post	post_ledger → Result<i64,LedgerError>	15 / 15
C06	composed/window-overlap	window_overlap → Result<(overlap,gap),WindowError>	15 / 15
C07	composed/quorum-decision	quorum_decision → Result<QuorumOutcome,QuorumError>	14 / 14
C08	composed/cooldown-gate	cooldown_gate → Result<(GateState,i64),GateError>	15 / 15
C09	composed/days-in-month	days_in_month → Result<i64,DateError>	15 / 15
C10	composed/clamp-value	clamp_value → Result<i64,ClampError>	10 / 10
C11	composed/safe-quotient	safe_quotient → Result<i64,QuotientError>	8 / 8
C12	composed/safe-absolute	abs_value → Result<i64,AbsError>	8 / 8
C13	composed/sign-class	sign_class → Sign	7 / 7

Each example holds exact authoring frames as run, a contract (parameter semantics, error precedence), independently expected cases, an external checker, a readable workbench view, the exported pack plus author name map, and a before/after record of one genuine program modification. composed/corrections.md retains one repaired authoring error and two independently justified expectation corrections. composed/admission-probe.md records the fresh-import edit/commit probe. composed/pilot.md records the example-assisted learning smoke test (methodology and aggregates only).

Reproduce without the pilot directory (needs the 2.0.3 agent binary):

./composed/reproduce.sh --agent /path/to/sley-2.0.3-linux-x86_64/bin/sley-agent

The workbench and scripts need writable scratch space. If the default temporary directory is small or quota-limited, point TMPDIR at a writable directory before running (the scripts and the agent honor it); no path is hardcoded and no cleanup infrastructure is provided.

Authoring workflow: composed/authoring.md is the default — pinned guide plus the frozen serializer/parse/recovery appendix first, with targeted consultation of a small relevant example only when a specific problem warrants it. composed/pilot.md preserves the learning smoke test including its unfavorable preload result (supplying examples raised input-token spend and did not help).

Persistence limitation: a combined code+tests commit is refused (TXN_TEST_EVIDENCE_UNSUPPORTED), so each example was committed split — code-only, then tests-after (transactions in catalog.json). Code-only plus tests-after is two transactions, not a test-admitted atomic commit. Native test admission is not run — unavailable in 2.0 — and no count in composed/ is test admission.