In developmentConnected product family

my-dev-kit Ecosystem

Graph-guided development tooling for coding-agent workflows.

A graph-guided development toolkit for coding-agent workflows.

The ecosystem combines repository intelligence, workflow orchestration, and validation tooling into a repeatable development loop. my-dev-kit builds codebase context, my-dev-kit-orchestrator structures implementation workflows, and my-dev-kit-lab evaluates the tooling across realistic projects.

Designed for

Developers using AI-assisted coding workflows
Technical project maintainers
Teams evaluating repeatable coding-agent practices

How the products are related

Vertical overview of my-dev-kit, my-dev-kit-orchestrator, and my-dev-kit-lab.

Context acquisition layer

my-dev-kit

Retrieves bounded project context through graph-guided workflows.

Index project
Project manifest + graphs
Search / Lookup / Slice / Source / Semantic view
Architecture context

Bounded context

Handoff artifact

Architecture Context Packet

Retrieval evidence + synthesized context packet

Handoff artifact

Workflow control layer

my-dev-kit-orchestrator

Organizes the staged design-to-code workflow after context acquisition.

Consume Architecture Context Packet
Stage workflow
Generate implementation prompts
Coding agent execution
Tests / Reports / Artifacts

artifacts and outcomes

Evaluation and visualization layer

my-dev-kit-lab

Evaluates prompt variants, agents, and outputs through experiments and visualizations.

Benchmark projects
Prompt variants
Agent adapters
Experiment runner
Reports / Charts / Demos / Gallery

v0.1.0Establish the baseline raw-full-file versus my-dev-kit-guided experiment pipeline with benchmark projects, fake and real agent adapters, deterministic scoring, reports, plots, screenshots, and gallery outputs.
v0.1.1Clean up public release hygiene, documentation, changelog, funding links, ignored artifacts, and release-facing metadata without changing the core architecture.
v0.1.2Handle dependency and security maintenance separately from feature work while preserving existing experiment behavior.
v0.1.3Harden Codex and Claude real-agent campaign behavior with better structured outputs, smaller presets, timeout handling, partial-result reporting, and clearer limitations.
v0.1.4Polish report and gallery UX with clearer artifact links, navigation, summary cards, metric notes, collapsible sections, and compatibility-preserving output improvements.
v0.2.0Refactor my-dev-kit-lab into a generic experiment-plugin framework with context-strategy-comparison as the first plugin, target-aware experiment execution, and plugin-aware reports.
v0.2.1Fortify automated security validation by extending security:validate with attack-scenario profiles, stronger adversarial checks, structured exploit evidence, and multi-option flags while remaining backward compatible.
v0.3.0Add the generic audit framework and the first audit detector family for code rot, focused on evidence-backed detection of stale references, docs/code mismatch, test rot, architecture drift, package drift, and workflow drift.
v0.3.1Add the code quality detector family to the audit framework for maintainability risks such as large files, complex functions, duplication, weak test coverage, poor module boundaries, and TypeScript/lint drift.
v0.3.2Integrate existing security validation into unified audit reports so audit output can include security findings while security:validate remains the focused automated security command.
v0.3.3Add the project-wide audit command behavior so npm run audit can combine code rot, quality, and security findings with profiles, issue deduplication, readiness verdicts, and stable JSON output.
v0.4.0Add a manual pentest framework beside automated security validation, generating human-usable plans, checklists, findings files, and reports for CLI, package, local-tool, and web-app profiles.
v0.4.1Add warm-index reuse experiment support to test the strongest my-dev-kit value case of indexing once and reusing the index across multiple related tasks.
v0.4.2Expand the warm-index benchmark suite with multiple localized, cross-module, broad-change, and negative-control tasks for medium and large projects.
v0.4.3Run warm-index real-agent campaigns with Codex and Claude while preserving structured partial outcomes and clear token and duration limitations.
v0.5.0Add index freshness and changed-file detection so the lab can identify whether a my-dev-kit index is fresh, stale, partially stale, or unknown after code changes.
v0.5.1Add affected-neighborhood experiment support to map changed files and symbols to graph neighborhoods and determine whether a future task overlaps stale context.
v0.5.2Add the incremental-change and staleness experiment plugin to compare stale index, refreshed index, and partial-refresh behavior after controlled code changes.
v0.5.3Add partial-refresh planning support so the lab can model full refresh, no refresh, changed-file refresh, and affected-neighborhood refresh treatments before my-dev-kit fully supports them.
v0.6.0Add the context-window scaling experiment plugin to measure when raw full-file context exceeds token budgets and whether my-dev-kit retrieval remains usable.
v0.6.1Add a deterministic synthetic large-repo benchmark generator for controlled scaling experiments across file counts, module depth, symbol count, imports, and task locality.
v0.6.2Add real-world and local repo experiment support with privacy-safe metadata, ignored-file handling, reproducibility metadata, and no target-source modification.
v0.7.0Add the retrieval precision and recall experiment plugin to evaluate whether my-dev-kit retrieves the expected files, symbols, and source slices without requiring real agents.
v0.7.1Add retrieval query strategy comparison across search, lookup, graph neighborhood, source, slice, data-model, model-view-lineage, and combined graph-guided workflows.
v0.7.2Add context-pack generation experiments to test compact task-specific context packs for coverage, size, source-slice relevance, and answer-key support.
v0.8.0Add the agent-success-rate experiment plugin to measure whether my-dev-kit-guided workflows improve real coding-agent outcomes by capturing diffs, running tests, and scoring task success.
v0.8.1Add edit-quality and blast-radius metrics to measure expected versus unexpected file edits, lines changed, regression risk, and whether changes stay focused.
v0.8.2Add multi-attempt repair experiments to separate single-shot benchmarking from repair-loop workflows with controlled feedback and cumulative outcome tracking.
v0.9.0Normalize provider and CLI telemetry across Codex, Claude, and future agents, including token reliability, duration sources, statuses, and non-comparable run warnings.
v0.9.1Add an agent campaign scheduler with queueing, resume mode, skip-completed behavior, one-case-at-a-time execution, timeout presets, and partial-result preservation.
v0.9.2Harden real-agent prompts with stricter schemas, shorter prompt modes, bounded tool-use guidance, max-command guidance, and per-agent prompt templates.
v1.0.0Release a stable experiment framework with multiple experiment plugins, stable artifact schemas, stable reports and gallery, deterministic fake demos, structured real-agent partial outcomes, and strong documentation.
v1.1.0Focus on incremental index and stale-context proof by strengthening changed-node detection, affected-neighborhood experiments, stale-index risk reporting, and reindex recommendations.
v1.2.0Focus on large-repo and external-repo scaling with external subject support, synthetic large repos, context-window campaigns, privacy-safe artifacts, and reproducibility metadata.
v1.3.0Focus on agent productivity and edit quality with stronger agent-success experiments, diff capture, test-pass scoring, blast-radius scoring, repair loops, and real-agent presets.
v1.4.0Turn experiment outputs into a publication and evidence portal with curated reports, public demo screenshots, release-linked evidence bundles, cross-experiment summaries, and responsible interpretation docs.

feedback for better prompts and workflows — back to my-dev-kit-orchestrator