Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

3.0.2 - 2026-08-22

The iac_scan verification gate now implements the full canonical iac_validator contract surface: terraform init -backend=false + terraform validate for Terraform configs (run from the directory containing the .tf files), pulumi preview --non-interactive for Pulumi projects, and the security operation via tfsec or checkov (MEDIUM+ findings fail the gate). Missing tooling is reported unavailable with an explicit follow-up (never a fabricated pass); provider-bound terraform init is reported unavailable rather than performing network acquisition.
Discovery emits IAC-001 (missing Infrastructure-as-Code) under the canonical performance gap dimension; the ops_workstreams forward fixture exercises the Terraform scaffold end-to-end.
Added a terraform-validation CI job that generates the baseline through the runtime and proves terraform init + terraform validate pass; the generated baseline uses the hashicorp/null provider so the proof runs offline with no credentials.

Ported the IaC handler (_handle_iac) from NOT_PORTED to a deterministic PARTIAL scaffold: canonical Terraform baseline (terraform/main.tf with pinned hashicorp/aws provider, variables.tf, outputs.tf), additive only — existing *.tf or Pulumi projects are never overwritten. The canonical iac_validator contract (validate/lint/security/cost) is surfaced as the follow-up agent step. No execution handler remains NOT_PORTED.
Discovery emits CONT-001 / OBS-001 gaps (classified under the canonical performance / dx dimensions) and execution routes them by gap id to the ws_containers / ws_observability handlers; the ops_workstreams forward fixture exercises both through the full CLI lifecycle.
Refreshed dist/rup-skill-v3.0.0.zip to track current HEAD.

Complete RUP Protocol v3.0.0 synchronization.
Agent workflow projections in workflows/.
Strict schema extraction for state artifacts.
Deterministic Python 3.11+ runtime for Discovery, Planning, Execution, Verification, and Reporting.
Robust security jailing and prompt-injection detection.
Deep capability mapping and validation tracking.
Test suites covering integration and portability.

Universal jailed I/O primitives (iter_jailed_files, open_jailed_read, atomic_jailed_write, jailed_mkdir, jailed_unlink) for every target read/write, including execution handlers, state migration, tool detection, provenance hashing, and packaging (symlinked members rejected).
Default <target>/.rup state root resolved and containment-verified.
Pre-execution adversarial trust gate, --allow-exec, and --sandbox required|preferred|off (default required); scrubbed subprocess environment, bounded and secret-redacted output.
Secret scanning with structured coverage status (strict mode fails closed on incomplete coverage), gitleaks/trufflehog merged when installed, and expanded built-in token patterns (GitLab, npm, PyPI, Stripe, Google, AWS session).
Offline JS/TS tool resolution (local .bin, npm exec --offline, pnpm/yarn exec, npx --no-install); network acquisition refused.

Planning constraints propagate from plan-state.json to execution and run IDs; ExecutionPhase enforces the escalation guard itself.
Verification gates require explicit command_succeeded (rc==0) alongside semantic results.
Per-workstream capability classes (deterministic/partial/agent_native/ not_ported) that verification evidence can never upgrade.
Dependency closure in work selection; per-workstream checkpoint graph enforced per item by execution.
Transactional rollback: content baseline (HEAD + per-path hashes), baseline-dirty-path write refusal, content-addressed backups, per-item platform-neutral operations, and a rollback CLI phase.
Monorepo workspace package graph (npm/yarn/pnpm, lerna, nx, turborepo, cargo, go.work) with --workspace / --changed-packages scoping, per-package tooling, per-package writes, and aggregate reporting.
Provenance reports omissions separately from transfer passes with per-capability rationale; capability lineage carries transfer rationale.

Refactored SKILL.md to properly route sub-workflows and enforce schema validation.
Restructured runtime/ to strictly return typed outputs and adhere to Schema Draft 2020-12.
Ported the containerization (ws_containers) and observability (ws_observability) workstreams from NOT_PORTED to PARTIAL: canonical multi-stage Dockerfile / .dockerignore / Compose scaffolding (language-aware, never overwrites user files) and an observability baseline (JSON structured logging, standard metrics, OpenTelemetry tracing with W3C Trace Context). All 27 canonical capabilities now have implementation; 0 remain not_ported.
Discovery now flags missing container configuration (CONT-001, classified under the canonical performance gap dimension) and a missing observability baseline (OBS-001, under dx) since the canonical Gap category enum has no dedicated slots; execution routes those gap ids to the ws_containers / ws_observability handlers. New ops_workstreams forward fixture exercises both handlers through the full CLI lifecycle.