RTL-ASS 1.1.0

RTL-ASS 1.1.0

RTL-ASS 1.1.0 strengthens the vendor-neutral Codex RTL workflow with an audited local corpus, six executable workflow classes, and stricter evidence identity. Codex remains the design and coding agent; RTL-ASS supplies RTL-specific guidance, retrieval, and deterministic open-source verification support.

Highlights

• Added a reproducible admission policy and file-level lock for 1,429 reviewed Verilog/SystemVerilog files across seven isolated namespaces. Upstream HDL remains local and is not redistributed in the release.
• Added six paired Codex workflow classes covering specification-to-RTL, FIFO repair, RTL/testbench attribution, SystemVerilog signed-width semantics, native FST localization, and OpenSTA-driven timing refinement.
• Bound evaluation reports to prompt, fixture, hidden grader, harness, Skill, runtime, combined payload, and sanitized report identities.
• Added explicit defined/undefined equivalence input domains and included the selected domain in evidence identity.
• Added hierarchical leaf-name waveform matching and portable, bounded VCD/FST first-divergence evidence with source/container identities.
• Hardened evaluation isolation, activation detection, timeout accounting, candidate/deliverable separation, packaging, and bytecode-independent payload identity.
• Expanded strict mypy coverage to every tracked Python module in src, tests, tools, and evals.

Reviewed workflow result

The audit ran five skill-off/skill-on pairs for each of six task classes: 60 Codex runs in total. Skill-on activation and complete structured evidence were 30/30; skill-off was 0/30 for both. External candidate correctness was 30/30 on versus 28/30 off, and timely task success was 29/30 on versus 23/30 off.

These heterogeneous, transparent cases show improved task completion and observable evidence behavior within the six tested classes. The completion gain comes with a cost tradeoff: one otherwise-correct skill-on FIFO run exceeded the 600-second budget, and skill-on used more input tokens in aggregate. This historical verification-oriented evaluation is distinct from the later retrieval-first comparisons.

Compatibility and upgrade

• Python 3.11 and 3.12 remain supported.
• The public CLI, evidence schema 1.0, knowledge-pack schema 1.0, and explicit database migration model remain compatible with the 1.x line.
• No proprietary simulator, synthesis, FPGA-vendor, or timing tool is required. Supported evidence is produced only by available open-source tools.
• Install rtl_ass-1.1.0-py3-none-any.whl and replace the installed rtl-ass Skill directory with the one from rtl-ass-skill-1.1.0.zip.

Release assets

• rtl_ass-1.1.0-py3-none-any.whl
• rtl_ass-1.1.0.tar.gz
• rtl-ass-skill-1.1.0.zip
• rtl-ass-sbom.spdx.json
• SHA256SUMS

Verify all downloaded assets with sha256sum --check SHA256SUMS. Yosys-native formal and equivalence remain bounded, OpenSTA evidence is not physical signoff, and local knowledge records are not trusted merely because they were imported.

RTL-ASS 1.0.0

RTL-ASS 1.0.0

Initial stable release.

• Added the Codex-first, vendor-neutral RTL skill and progressive RTL references.
• Added safe Verilog/SystemVerilog inspection and open-tool discovery.
• Added Verilator, Icarus Verilog, Yosys synthesis/formal/equivalence, and OpenSTA evidence adapters.
• Added bounded VCD and resource-bounded FST waveform query and first-divergence workflows.
• Added the transactional SQLite/FTS5 knowledge store, explicit migrations, namespace isolation, guarded lifecycle, atomic verification and observation, and tamper-evident audit chain.
• Added candidate derivation and strict portable knowledge-pack import/export.
• Added the first-party ready/valid RTL, TB, assertion, and engineering-card starter pack.
• Added JSON schemas, Python 3.11/3.12 CI, optimized-runtime tests, strict typing/linting, source/wheel/skill packaging, SBOM/checksum generation, and reproducible release auditing.

Verification scope: this initial release supplies the knowledge and evidence mechanisms. Subsequent paired workflow and retrieval observations are documented in the repository's current docs/evaluation.md, separately from the 1.0.0 release artifacts. Yosys-native formal and equivalence remain bounded; OpenSTA is not physical signoff.