Skip to content

AI-DLC Workflows 2.0 - Roadmap

Status as of 2026-09-01.

  • The current v2 version is 2.6.124 (origin/main tip 82d2e304). Version numbers describe the committed framework tree, not GitHub Releases.
  • AI-DLC Workflows 2.0 is GA on the default main branch. Use main for new installations and upgrades. The earlier implementation is maintained separately on v1.
  • Release publication is not yet aligned with main: GitHub still marks v1.0.1 as Latest, tracked by #635. The native distribution implementation for #722 remains under review in #756; no public v2 native release exists yet.
  • PR validation now includes the deterministic integration and end-to-end tiers in addition to smoke, unit, packaging, typecheck and lint (#791).

The version numbers below describe where work landed on main. Future themes and open pull requests are directional, not committed release promises.

North star reference

The seven functional goals of the AI-DLC Workflows 2.0 North Star, verbatim in intent:

  1. Mimic what we practice in the real world - a stage executed by a configurable ensemble (Owner, Collaborator, Verifier) with consistent semantics across harnesses.
  2. Customization of behaviour - encode new behaviours, policies, or constraints in no more than two targeted changes, reusable across harnesses without tool-specific rewrites.
  3. Adaptiveness of workflows - scale in (report triage to compact Fix, Test, PR) and scale out (decide next stages at boundaries); composition not hard-wired.
  4. Verifier as a true adversary - adversarial quality gate; may use a different LLM than the producer; validates against machine-checkable evidence; budgeted self-heal loop escalating to HITL.
  5. Support for cyclic, directional flows - forward progression plus governed, directional feedback loops.
  6. Preserve artefact traceability - downstream stages enrich upstream artefacts rather than spawning disconnected ones.
  7. Organizational, not project-local, artefact repository - shared org knowledge layer across projects, intents, and repos; six named scenarios.

Strategic delivery pillars

Two strategic pillars shape how the North Star reaches users and evolves:

  • Productization and distribution (#722) - make AI-DLC straightforward to install, configure, upgrade, release and roll back across supported harnesses.
  • Plugin ecosystem and marketplace (#723) - make trusted extensions discoverable, installable and reusable, with a clear path from external plugin to first-party capability.

Goal scorecard

# Goal Status Delivered by Remaining work
1 Real-world ensemble Shipped 2.5.0 independent collaborators and selectable topologies (#568), enforced reviewer receipts (#569), batch-parallel per-unit waves (#617), team-owned parallel Units (#879) Harness-native live-team transports remain an enhancement
2 Customization Shipped, with follow-ups 2.3.0 plugin seam, 2.3.5 content projection/selection (#550), deterministic rule delivery (#658), plugin scopes (#664), reusable plugin test kit (#792), plugin doctor extensions (#797), standalone authoring toolchain (#892) Stage-specific rules, when: evaluation, remote discovery and marketplace (#723)
3 Adaptiveness Shipped 2.2.0 composer, entropy-scored composition (#595), deterministic ARS (#644), unit-major Code Generation (#705), Classic/Express scopes and conditional protocol modules (#767), per-session workflow bindings (#858) Boundary changes remain human-approved by design
4 Verifier as adversary Shipped 2.4.0 adversarial evidence contract (#566), gate-and-completion enforcement (#569, #551), reviewer-class cost dial (#718), turn/recovery backstops (#613, #758), gate-bound blocking sensors (#836) Pull-request-level adversarial review is under development in #799
5 Cyclic flows Partial Within-stage review/revision loops, bounded recovery mechanics, explicit human-authorized forward/backward/redo stage jumps, and bounded Build & Test to Code Generation loop-back (#616) General governed cross-stage feedback loops remain unbuilt
6 Traceability Partial Artefact graph, upstream coverage, per-stage enforcement (#401), claim provenance (#647, #686), shared CodeKB safeguards (#670), domain/contract boundaries (#711), stale-result propagation (#716), source-bound and per-Unit review receipts (#646, #813) Progressive in-place enrichment and cross-unit discovery propagation (#299)
7 Org repository Shipped 2.1.0 spaces/intents/org-KB, declared multi-repo manifest and sync (#674), clone-safe active-space cursor (#709), DocumentKB indexing and citations (#731), summaries and tags (#894) Auditable supplemental-knowledge selection remains an active extension (#694)

Delivered

Version Capability Goal Key PRs
2.0.0 - 2.0.2 GA preview: reviewer mechanism, multi-harness core, agent roster 1, 4 v2 baseline
2.1.0 Per-intent workspace: spaces, intents, multi-repo, org-KB 7 #429
2.1.2 Per-unit for_each iteration 3 #444
2.1.3 - 2.1.8 Loop integrity and reviewer wiring across harnesses 1, 4, 5 #405, #443, #466, #482
2.2.0 - 2.2.19 Adaptive workflows, composer, scale-in and Construction hardening 3 #477, #491, #509-#512, #520-#522, #525
2.3.0 - 2.3.5 Plugin mechanism, agent tiers, install-time plugin selection and content projection 2, 4 #475, #546, #550
2.3.6 - 2.3.11 Phase progress, citation-aware upstream coverage, pinned lint and gate accounting 4, 6 #562, #563, #572, #573
2.4.0 Reviewer-as-verifier: adversarial, evidence-grounded review 4 #566
2.4.2 - 2.4.6 Whole-root packaging, native dispatcher/binaries, documentation parity and opencode harness 1, 2 #560, #571, #577, #578, #581
2.5.0 Three-role ensemble: independent collaborators, pipeline, mob and hub-and-spoke 1 #568
2.5.1, 2.5.25 Entropy-scored minimum workflow composition and deterministic ARS 3 #595, #644
2.5.2 Redacted /aidlc --doctor --export diagnostic bundle - #576
2.5.5, 2.5.39, 2.5.41, 2.5.54-2.5.55 Reviewer receipts, review freeze, plan-before-code guard, reviewer classes and authorization receipts 1, 4 #569, #677, #692, #702, #718
2.5.11, 2.5.38, 2.5.57-2.5.58 Claim provenance, pre-generation confirmation and project-language grounding 6 #647, #686, #703, #707
2.5.33 - 2.5.36 Deterministic steering delivery, plugin scopes, CodeKB preservation and workspace manifest/sync 2, 7 #658, #664, #670, #674
2.5.40, 2.5.53 Per-stage token/cost accounting, opt-in metrics and usage-tracking kill switch - #673, #720
2.5.56 Code Generation joins the unit-major Construction walk 3 #705
2.5.60 GitHub Copilot harness for Copilot CLI and VS Code agent mode 1, 2 #657
2.5.63 Cursor harness 1, 2 #661
2.5.67 Batch-parallel per-unit waves and foreground reviewers 1, 4 #617
2.5.71 - 2.5.75 Per-stage traceability enforcement, design-stage code boundaries, test-instruction ownership and first-class observability artifacts 4, 6 #401-#404
2.6.1 - 2.6.2 Domain/contract design restructuring, consolidated infrastructure design and follow-up guards 1, 6 #711, #751
2.6.8 - 2.6.9 Reviewer turn backstop and bounded stale-receipt recovery 4 #613, #758
2.6.12 - 2.6.14 Copilot continuation stability, gate authorship enforcement and audit timestamp normalization 1, 4 #749, #750, #759
2.6.15 DocumentKB S1 indexing and citation delivery 7 #731
2.6.16 Code Generation plans bound to the affirmed Testing Posture 4, 6 #772
2.6.17 Reusable plugin test kit and plugin-author testing tiers 2 #792
2.6.18 Classic and Express scopes, Classic implicit default and conditional protocol modules 2, 3 #767
2.6.20 Bounded Build & Test to Code Generation failure loop-back 5 #616
2.6.37 Code Generation review receipts bound to workspace source state 4, 6 #646
2.6.51 - 2.6.64 Continuation cursor, Kiro reliability, plugin-extensible doctor checks, stage-validity receipts and native Kiro IDE surfaces 1, 2, 4, 6 #822, #788, #797, #716, #824
2.6.69 Per-Unit source attribution for Code Generation review receipts 4, 6 #813
2.6.72 - 2.6.74 Gate-bound blocking sensors and binding quality-target verification 4, 6 #836, #842
2.6.80 Per-session workflow bindings for concurrent intents in one space 3, 7 #858
2.6.87 DocumentKB summaries and tags 7 #894
2.6.105 Standalone plugin create, validate, build and test toolchain 2 #892
2.6.107 Team-owned Units and parallel Construction across teams 1, 3 #879
2.6.114 No-DAG per-Unit review continuity 1, 4 #947
2.6.121 - 2.6.124 Immutable reviewer evidence, Git-independent source binding and portable workflow state paths 4, 6 #888, #904, #962

In flight

Selected open work is listed without version claims. Merge readiness changes frequently; each linked pull request is authoritative.

PR Work Theme
#756 Native distribution, six-command CLI, config policy and release hardening Installation and releases
#775 Unified Kiro distribution aligned to the agent harness Harness parity
#782 Product-discovery plugin (AI-PLC) Plugins and product discovery
#799 Adversarial AI pull-request review agent CI and verification
#969 Construction integration through pull requests Delivery workflow
#968 Devin CLI and Desktop harness Harness expansion
#907 mabl verification plugin Plugins and verification
#753 Evaluator integration Evaluation
#526 Product discovery in Ideation Product discovery

Directional themes

These themes are supported by open RFCs, issues or implementation pull requests, but do not yet have committed release versions.

Traceability and progressive enrichment

  • Per-stage upstream traceability enforcement shipped in #401. Source-bound review evidence shipped in #646, stale stage-result propagation shipped in #716, and per-Unit attribution shipped in #813.
  • Cross-unit discovery propagation remains open (#299/#300).
  • Preserve progressive enrichment as the North Star destination: downstream stages enrich upstream artefacts in place, with ADRs as a core design artefact.
  • Commit-level provenance remains an open design question; the current audit chain does not provide a durable reverse lookup from an arbitrary source commit to its intent and workflow.

Governed feedback loops

  • #616 shipped one bounded Build & Test to Code Generation return path for #611. It is an incremental loop, not a general cyclic graph engine.
  • General cross-stage backward edges still need engine-level governance, stale artefact handling and explicit human authorization.

Plugins and marketplace

  • The plugin mechanism, content projection, selection and plugin-contributed scopes are shipped; the plugin test kit and authoring tiers shipped in #792.
  • Plugin-extensible doctor checks shipped in #797. The offline plugin CREATE, VALIDATE, BUILD, and TEST authoring tiers ship as the standalone aidlc-plugin-create.ts, aidlc-plugin-validate.ts, aidlc-plugin-build.ts, and aidlc-plugin-test.ts tools. The top-level plugin validate and plugin build routes also ship. Top-level plugin create and plugin test routes remain proposed in #723. Remote discovery, trust, a first-party marketplace and a graduation path are also proposed in #723. Product discovery (#652, #782) and design (#527) are candidates for first-party plugins.
  • aidlc-plugin-test.ts exercises composition against a disposable copy of an install for external plugin authors.

Knowledge and documents

  • #731 shipped DocumentKB's first indexing and citation slice. Summaries and tags shipped in #894, completing that metadata slice of the tracked #714 RFC.
  • #694 tracks intent-aware discovery and auditable supplemental-knowledge delivery across stage topologies.

Product discovery

  • Core Ideation delivery remains under review in #526, with an external-handover contract in #586 and a plugin-shaped alternative in #652.
  • The delivery surface, core versus first-party plugin, is not yet settled.

Installation, upgrades and releases

  • The GA implementation and its active development line now live on main. The earlier implementation remains on v1.
  • #722 covers binary packaging, installers, release automation, rollback and post-install setup; its milestones 1-3 implementation is under review in #756. The earlier Bun dependency tracker #399 is closed as superseded by #722.
  • #636 tracks a first-class upgrade contract. The earlier implementation PR #535 closed without merging.
  • #635 tracks the mismatch between the v2 GA main branch and GitHub's Latest release still pointing at v1.0.1.

Harness expansion and parity

  • GitHub Copilot support shipped in #657, and its RFC #472 is closed.
  • Cursor support shipped in #661. A unified Kiro distribution is under review in #775. Native Kiro IDE surfaces shipped in #824, closing #555, and hook matcher hardening shipped in #788.
  • Antigravity setup is proposed in #690.

Evaluation and operations

  • #684 proposes repeatable benchmarks for measuring AI-DLC outcomes. Evaluator work is active in #753; the earlier harness-evaluation tracker #223 closed as not planned for v1.
  • Operations-phase steering remains a requested direction (#221, #473), not an active main implementation stream.

Known gaps

  • Stage-specific rules (aidlc-stage-<slug>.md) are reserved but unbuilt.
  • Plugin when: evaluation, remote discovery and marketplace trust remain open.
  • Write-fired sensors remain advisory; gate-bound sensors support blocking severity and human-backed override.
  • General cross-stage cycles and progressive in-place artefact enrichment remain North Star gaps.
  • The unified Kiro distribution remains under review in #775.
  • Older community PR #526 remains open and needs rebasing or disposition. PRs #432, #535, #552, #653 and #712 closed without merging.