Skip to content

v1.5.0 — Lifecycle Expansion + Hardening Pass

Choose a tag to compare

@VaiYav VaiYav released this 24 Apr 12:30
· 63 commits to main since this release

[1.5.0] — 2026-04-19

Minor release (additive). Expansion of the lifecycle with portfolio view,
brown-field entry, lite mode, full monorepo support, optional V-Model
integration, unit + integration test phases, and operational readiness
artifacts. Schema v3 for .forge-status.yml is additive over v2 — no
breaking change to existing features; a lazy migration stamps
schema_version: 3 on first write.

Added — 7 new commands

  • speckit.product-forge.portfolio — Cross-cutting portfolio view. Scans
    every features/*/.forge-status.yml, produces feature table, file-conflict
    matrix, dependency graph (Mermaid), and suggested merge order. Read-only.
  • speckit.product-forge.backfill — Brown-field entry. Reverse-engineers
    a feature folder from existing code: retro product-spec, plan, simplified
    tasks, .forge-status.yml with backfilled: true, and a gaps-report of
    missing artifacts.
  • speckit.product-forge.monitoring-setup (Phase 9.5, opt) — Produces
    real NewRelic-compatible dashboard JSON, alert rules, and SLI/SLO doc
    from plan NFRs and tracking-plan events. Wraps newrelic-dashboard-builder.
  • speckit.product-forge.migration-plan (Phase 5.5, opt, conditional) —
    Generates zero-downtime migration plan with forward.sql, rollback.sql,
    validation.sql, backfill plan, and risk matrix when plan.md introduces
    schema changes. Wraps db-migration-manager / mongodb-ops.
  • speckit.product-forge.i18n-harvest (post-bridge, opt) — Extracts
    user-facing strings from wireframes/spec, generates feature-namespaced
    locale keys, stubs TODO entries in every configured locale. Wraps
    i18n-workflow.
  • speckit.product-forge.experiment-design (Phase 9B, opt) — Forces a
    pre-registered A/B plan — hypothesis, MDE, sample size, exposure rules,
    guardrails, decision rule — before shipping a flag-gated feature. Wraps
    feature-flag-ab-testing.
  • speckit.product-forge.feature-flag-cleanup — Cross-cutting stale-flag
    audit. Scans every flags/registry.yml, cross-checks codebase, produces
    removal recipes for flags past cleanup_after. Read-only; wraps
    feature-flag-manager.

Added — Monorepo support (first-class)

  • codebase.paths block in project config, with workspace_type
    (pnpm | yarn | npm | turbo | nx | rush | lerna | none) driving test-
    runner resolution. Legacy codebase_path still supported.
  • scope.paths, scope.cross_workspace, scope.primary on
    .forge-status.yml — names which workspaces a feature touches.
  • Workspace-prefixed paths in tasks.md and task_log[].paths (e.g.
    backend:src/users.ts).
  • Per-workspace conflict matrix in /portfolio with grouped "By
    workspace" section.
  • Affected-only test execution in implement + test-run when running
    under nx / turbo / pnpm workspaces.
  • Cross-workspace change propagation — scope widens automatically
    when a change request touches code outside the original scope;
    recorded as gate condition.
  • Documented in runtime.md §9,
    config.md Paths.

Added — V-Model integration (optional extension)

  • optional_extensions block in extension.yml declaring
    leocamello/spec-kit-v-model
    (≥0.5.0) as an opt-in dependency.
  • feature_mode: v-model is now a real mode, not a stub. When
    selected, forge.md detects the V-Model extension and delegates the
    middle of the lifecycle (V1–V13) to its 14 commands: requirements,
    acceptance, system / architecture / module design paired with system /
    integration / unit test plans, trace, peer-review, test-results
    ingestion, audit report.
  • No silent fallback: if V-Model plugin is absent, v-model mode
    aborts with the install command. Regulated work must not degrade.
  • Domain selection via v-model-config.yml
    (iec_62304 | iso_26262 | do_178c | generic).
  • New docs/v-model-integration.md covers
    the full phase map, detection, fallback rules, status-file additions.

Added — Testing strategy (universal)

  • New docs/testing-strategy.md — universal
    (framework-agnostic) pyramid, when unit vs integration vs contract vs
    E2E, per-layer coverage criteria, anti-patterns, flaky-test handling,
    test-data management, monorepo-aware execution.
  • test-plan.md §5E Unit Tests (TC-UNIT-NNN) — derived from module
    boundaries + behavioural acceptance criteria. Framework-agnostic.
  • test-plan.md §5F Integration Tests (TC-INT-NNN) — beyond endpoint
    contracts: service↔DB, service↔cache, event emitter↔listener,
    middleware stacks, cross-workspace integration.
  • test-run.md §4E Unit and §4F Integration — non-browser
    execution with the same auto-fix loop; §4F handles testcontainers /
    docker-compose / in-memory / shared-DB isolation strategies.

Added — Structural capabilities

  • .forge-status.yml schema v3 — Additive over v2. New fields:
    feature_mode, backfilled, v2_native, phase started_at/completed_at/tokens_in
    /tokens_out/tool_calls/digest_path/skipped/skip_reason, task_log[]
    (renamed from early-draft tasks[] to avoid collision with phases.tasks)
    with size (XS/S/M/L/XL), paths, and commit_sha, gates[].approvals,
    gates[].skip_reason, dependencies.depends_on /
    depended_on_by, role_approvals.solo_mode /
    required_roles_per_phase. New status enum value not_applicable for
    out-of-mode and backfilled phases. Full spec in
    docs/schema/forge-status-v3.schema.yml.
    Migration rules in docs/schema/migration-v2-to-v3.md.
    Optional helper script scripts/migrate-status-v2-to-v3.ts.
  • State-lock protocol — .forge-status.yml.lock file-based lock with
    TTL-based takeover. Prevents concurrent-writer corruption between the
    orchestrator and sub-skills. Documented in
    docs/runtime.md §2.
  • Per-phase digests — Every major phase (research, product_spec,
    plan, tasks, implement, verify) now writes <phase>/digest.md.
    Runtime refuses to mark a phase completed without a digest. Downstream
    phases (verify-full, portfolio, retrospective) read digests first to keep
    context budgets small. Template:
    docs/templates/phase-digest.md.
  • Lite mode — New feature_mode field selects a 5-phase map
    (problem-discovery opt → product-spec → plan → implement → verify)
    for small features. Escalation to standard is append-only. Details in
    docs/policy.md §4.
  • Skip-reason policy — Skipping an optional phase now requires a
    free-text reason when require_skip_reason: true (default). Reason is
    persisted on both the phase and the gate entry. Enforced by all skippable
    commands. Details in docs/policy.md §3.
  • Learning loop — .product-forge/lessons.md append-only log. Written
    by retrospective at launch close, read by research as a new dimension
    ("Prior lessons that apply") scored by tag overlap. Format in
    docs/lessons-format.md.
  • Drift budget — sync-verify now categorizes drift as structural
    (always human-in-the-loop) or cosmetic (whitespace, ordering, stale
    dates). Opt-in sync_verify.auto_resolve.cosmetic auto-fixes whitelisted
    drift only; structural drift never auto-resolves. Budget warns when
    cosmetic count exceeds threshold.
  • Release-readiness becomes an artifact producer — Step 1D now invokes
    feature-flag-manager to produce flags/registry.yml; Step 3D invokes
    newrelic-dashboard-builder to produce monitoring/dashboard.json,
    alerts.yml, slo.md. Graceful fallback when provider skills are missing.

Changed

  • commands/forge.md refactored. Operating rules moved to
    docs/policy.md; runtime flow (config, state lock,
    detection, resume, pre-flight, sync, gate audit, digest enforcement,
    context budget) moved to docs/runtime.md; status
    schema narrative moved to docs/schema.md. The
    orchestrator file now focuses on phase delegation and the mode-
    resolution entry. No behavioral change; cross-references only.
  • config-template.yml — New keys: default_feature_mode,
    require_skip_reason, sync_verify.drift_budget.{cosmetic, structural},
    sync_verify.auto_resolve.cosmetic.
  • research.md Step 2.5 — Consults .product-forge/lessons.md for
    matching prior lessons and surfaces them in research/README.md.
  • retrospective.md Step 5 — Drafts lesson blocks, confirms with user,
    appends to lessons.md, records count in phases.retrospective.lessons_added.

Release-prep hardening

Surfaced by the plugin test plan dry-run and closed before release:

  • Broken historical refs removed from CHANGELOG.md, README.md,
    and docs/how-it-works-v2.md — pointed to archive directories that
    are not in this tree.
  • Policy.md §3 tightened — empty skip reasons under
    require_skip_reason: true now explicitly reject the gate and
    re-prompt; no skipped gate entry is written until a non-empty
    reason is supplied.
  • Enum validation added to commands/forge.md Mode Resolution and
    docs/runtime.md §4 pre-flight — invalid feature_mode,
    phases.<name>.status, or gates[].decision values abort with a
    clear message instead of falling through silently.
  • tasks.md Step 4.1 — hard structural checks for task-ID
    uniqueness and monorepo workspace-prefix validation. Duplicates and
    unknown workspace names now abort the phase with a pointer to the
    offending line instead of propagating into task_log[].
  • Test-plan self-references — smoke and Layer B criteria updated
    to exclude docs/qa/plugin-test-plan.md (which legitimately names
    the patterns being searched for).

Migration notes

  • No action required. Existing features continue to work. First time a
    v1.5.0-aware skill writes to a feature's .forge-status.yml, it stamps
    schema_version: 3 and may populate new optional fields as it runs.
  • Optional: run scripts/migrate-status-v2-to-v3.ts once to stamp
    every feature at once.
  • Monorepo support (GitHub issue #1) remains unchanged in this release.
    Schema v3 fields reserve room for it (dependencies, backfilled) but
    the codebase-path configuration remains single-path. Tracked for a
    follow-up release.