v0.7.0 — Multi-Agent Efficiency and SDLC Patterns

AgentFlow SDLC v0.7.0 — Multi-Agent Efficiency and SDLC Patterns

This minor release introduces formal patterns for multi-agent workflows, drastically improving efficiency while preserving durable compliance. It also expands the framework's documentation to explicitly map its core tenets and rule sets.

What's New

Multi-Agent Efficiency Patterns

• Dual-Write Handoffs: The Orchestrator now uses local scratch files (.agent-runs/scratch/SDD-.md) for heavy technical design documentation while simultaneously posting compliance summaries to the GitHub issue. This saves significant token context and time on network round trips.
• Architectural Sprints: Batch orchestration commands are now formalized as Architectural Sprints. The Architect phase evaluates the entire batch collectively before implementation begins.
• Opt-In Design Gates: Introduced an explicit pause step after the Architect phase for high-assurance work or explicit prompts, enforcing human review of the Solution Design Document before code is written.
• Sub-Agent Role Templates: Added isolated
  ole-architect.md and
  ole-implementer.md templates to provide strict boundaries during phase handoffs.

Documentation and Core Principles

• Core Tenets Matrix: The README now explicitly maps SDLC challenges to AgentFlow capabilities and solutions.
• Human/Agent Parity: Documented the strict rule that human and agent contributors must follow the exact same issue/PR contract.
• Unambiguous Issue Standards: Updated issue templates and standards to clearly separate Background & Problem from Technical Design.
• Lifecycle Flowchart: Added a Mermaid diagram to the README mapping the end-to-end issue workflow.

Upgrade Guidance

Existing projects can sync these updates and receive the new issue templates and orchestrator behaviors:

bash node /path/to/agentflow-sdlc/bin/cli.mjs sync --target .

v0.6.0 — Extension adoption automation and repository licensing

v0.6.0 — Extension adoption automation and repository licensing

Highlights

• Adds deterministic extension adoption automation for projects that use multi-agent-sdlc.
• Adds first-class multi-agent-sdlc extensions helper commands to list, inspect, enable, disable, and validate extension packs.
• Makes initial setup, doctor, and update automation extension-aware while preserving the rule that discovery is automatic and activation is explicit.
• Supports repo-local extension packs added after initial setup under extensions/ or contrib/, including duplicate-id detection and missing/invalid enabled-pack reporting.
• Documents the extension adoption lifecycle across onboarding, assisted update, deterministic update planning, and extension-pack reference docs.
• Adds an Apache-2.0 project license and package metadata so adopters and contributors have explicit reuse terms.

Included integrated work

• #84, #85, #86, #87, #88 — deterministic extension adoption automation workstream.
• #89 — Apache-2.0 project license.

Validation

• pnpm format:check
• pnpm test
• pnpm test:workflow
• pnpm validate:extensions
• node scripts/validate-release-versioning.mjs --current 0.5.0 --next 0.6.0 --bump minor --notes docs/releases/v0.6.0.md

Follow-up work

• None.

v0.5.0 — Capability-first orchestration and deterministic updates

v0.5.0 — Capability-first orchestration and deterministic updates

Highlights

• Adds a portable advanced-capability model for PLAN, WORKFLOW, LOOP, and SUB-AGENTS so skills can request reusable intents while each harness resolves native, package, emulated, manual, or unavailable support.
• Adds capability adapter documentation for Claude Code, Codex CLI, Pi, Agy, and manual/human workflows.
• Adds deterministic capability tooling with resolve-capability and validate-capability-evidence plus tests.
• Adds a deterministic assisted update proposal for already-adopted projects, including the future update-plan direction, approval gates, PR evidence, and LLM-session safety rules.
• Streamlines release closeout evidence and PR manifest guidance from the integrated v0.4.1 follow-up.

Included integrated work

• #61 — release closeout streamlining
• #64, #65, #66, #67, #68 — capability-first advanced agent orchestration workstream
• #70 — deterministic assisted update review

Follow-up work

• #71 — deterministic read-only update-plan command
• #72 — approval-gated assisted update apply mode
• #73 — deterministic assisted update PR evidence generation

v0.4.1 — Streamlined release closeout

Release v0.4.1

Summary

Patch release that makes release closeout safer and easier to verify. Maintainers now have explicit guidance and tooling to publish GitHub Releases after a release PR merges, and release notes are checked for user-facing capability wording instead of issue-list bookkeeping.

What maintainers can do

• Follow a deterministic post-merge release checklist from planning through gh release create, gh release view, and gh release list verification.
• Validate that a tag and GitHub Release point at the expected merge commit.
• Check release notes for capability-oriented wording that explains value to adopting projects and maintainers.
• Capture release closeout evidence in PR manifests with target commit, title, notes source, and verification commands.

Upgrade guidance

Projects that adopted the framework should update with the assisted update flow:

node bin/cli.mjs update-prompt --target /path/to/your-project
node bin/cli.mjs doctor --target /path/to/your-project

Then run sync only after reviewing the generated update plan.

Compatibility

This is a backwards-compatible patch release. It adds documentation, template fields, a release closeout validator, and tests. Existing release planning commands remain preview-first.

Validation

• pnpm test
• pnpm test:workflow
• pnpm format:check
• node scripts/verify-hooks.mjs
• node scripts/validate-role-routing.mjs
• node scripts/validate-release-versioning.mjs
• node scripts/validate-release-closeout.mjs --tag v0.4.0 --target 8281f0d9e70bb25a7cd0cfd1abc87b2e5084aee6 --notes .agent-runs/scratch/github-release-v0.4.0.md

References

• #60 — release closeout workflow and user-facing release-note guidance.

v0.4.0 — Deterministic multi-agent updates

Release v0.4.0

Summary

Minor release for additive workflow capabilities delivered after v0.3.0. This release focuses on contributor guidance, deterministic multi-agent execution evidence, and assisted update support for adopted projects.

Included issues

• #52 — canonical README.md#contribute section for human and agent contributors.
• #54 — deterministic execution targets for Claude, Agy, Pi, Codex, humans, and provider-backed models.
• #56 — deterministic multi-agent role alternation and role attribution matrix validation.
• #57 — assisted update workflow and update-prompt helper for already-adopted projects.

Highlights

• Added explicit execution-target vocabulary and preflight resolution for ambiguous requests such as with claude, with agy, and with pi.
• Added role-attribution validation so multi-agent claims show which SDLC roles were performed by which independent intelligences.
• Added an assisted update guide for consuming projects that already have agent-framework-lock.json.
• Added node bin/cli.mjs update-prompt --target <project> for deterministic, read-only, approval-gated update planning.
• Added contributor workflow guidance in README.md#contribute.

Validation

• pnpm test
• pnpm test:workflow
• pnpm format:check
• node scripts/verify-hooks.mjs
• node scripts/validate-role-routing.mjs
• node scripts/validate-release-versioning.mjs

Release decision

• Version: 0.4.0
• Tag: v0.4.0
• Bump: minor
• Rationale: additive backwards-compatible docs, CLI helper, templates, validators, and workflow evidence capabilities.

v0.3.0 — Release versioning strategy

Highlights

• Added configurable release versioning strategy support with the default main.minor.fix model.
• Defined release bump meanings: main for breaking/compatibility-impacting changes, minor for additive backwards-compatible capabilities, and fix for corrections.
• Added docs/release-versioning.md with setup choices, release evidence guidance, preview-first helper usage, validation examples, and promotion workflow.
• Added releaseVersioning configuration in agent-workflow.config.json docs for strategy, segments, tag format, package version source, explicit approval, and pre-release policy.
• Added read-only release preview command: node bin/cli.mjs release-plan.
• Added release consistency validator: node scripts/validate-release-versioning.mjs.
• Added release versioning library and tests.
• Added PR manifest ## Release decision section.
• Included the integration lifecycle fix from PR #50 so related Refs #... are ignored by automation; only implementation/closure keywords such as Implements and Closes drive issue integration.

Validation

• node bin/cli.mjs release-plan --target . --bump minor --json
• node scripts/validate-release-versioning.mjs --current 0.2.0 --next 0.3.0 --bump minor --json
• pnpm test
• pnpm test:workflow
• pnpm format:check

v0.2.0 — Assisted onboarding and QA tooling

Highlights

• Added assisted onboarding for existing projects, including a canonical guide, copy-paste agent handoff prompt, OS-specific launch examples, and a prompt-only onboarding-prompt CLI command.
• Added read-only environment validation guidance with docs/environment-tools.md, aligned with doctor-env and clarity-over-automation setup rules.
• Documented the optional qa-expert role and its environment/tool profile, including the opinionated default to use Vibium for browser-based exploratory QA.
• Added optional QA tooling guidance for browsers, evidence capture, Playwright/Cypress/Selenium, Axe/Lighthouse/devtools, curl/HTTPie/Postman/Bruno, simulators/emulators, and project-specific test data.
• Improved README and docs index discoverability for onboarding, environment tools, default skills, qa-expert, and project setup.

Validation

• node bin/cli.mjs onboarding-prompt --target .
• node bin/cli.mjs doctor-env --json
• pnpm test
• pnpm test:workflow
• pnpm format:check

v0.1.0 — First open release

This is the first open release of multi-agent-sdlc after the framework was extracted, reused, and refined across a few solo-entrepreneur projects.

It captures the workflow patterns that proved useful in practice: keeping one agent in control by default, making role transitions explicit, preserving durable GitHub evidence, and reducing the management overhead of agent-assisted delivery.

Highlights

• Role-based single-agent workflow with explicit analyst, architect, developer, tester, review, tech-writer, and PR-readiness passes.
• Optional role routing for teams that want to involve claude, codex, agy, or pi without making multi-agent orchestration mandatory.
• Project-configurable branch strategy, including this repository's development -> main integration flow with no staging branch.
• GitHub issue, handover, workflow-status, and PR manifest conventions for auditable delivery.
• Integration lifecycle automation that comments, labels, and closes implementation issues after PRs merge into development.
• Installer/sync/doctor tooling for adopting projects, including framework-owned docs, hooks, validators, templates, and GitHub workflows.
• README and docs improvements with setup guidance, product-management prompts, epic breakdown prompts, orchestration examples, and default skill provenance.

Intended audience

This release is aimed at solo maintainers and small teams using AI coding agents who want a manageable SDLC without carrying the whole workflow in memory: clear issue standards, predictable branch/PR behavior, review evidence, and lightweight automation around integration and release tracking.

Notes

This is an early open release. The framework is intentionally opinionated and should be adopted by syncing it into an existing project, then adjusting agent-workflow.config.json, AGENTS.md, and project-specific conventions.