bp Blueprint Planner-Agent

Plan first. Route smarter. Ship with agents.

Terminal AI planning harness that turns vague software ideas into dependency graphs, model-routed handoffs, and validation contracts for coding agents.

$ blueprintAI Agents & Agentic Workflows

              planner gpt-5.5 / high reasoning
              providers OpenAI Codex · Claude Code · Gemini CLI
              output .blueprint/tasks/*.md
            

Problem

AI agents execute fast. Planning still breaks.

Large AI-assisted coding projects fail when the first prompt is vague and the work is not sliced cleanly.

Context drifts across chats and tools.
Tasks overlap or miss dependencies.
Expensive models get used for cheap work.
Weak models get assigned to high-risk architecture.
Outputs are hard to audit before code changes begin.

Solution

The planning brain before worker agents start.

Blueprint runs an investigative terminal chat, validates missing requirements, routes work to exact models, and writes handoffs that other agents can execute.

Chat-first planningBrainstorms and fills gaps before architecture is locked.

Compact contextReads project inventory instead of dumping the entire repo.

Task graphCreates dependency-aware work units with clear blocking rules.

Worker handoffsGenerates XML task prompts, model choices, and acceptance contracts.

Workflow

One terminal surface from setup to handoff.

The harness keeps configuration, brainstorming, preview, and artifact generation in one flow.

1Onboard project, providers, models, and reasoning effort.

2Brainstorm until scope, constraints, and validation are clear.

3Preview dependency graph and model assignments.

4Generate architecture docs, XML tasks, and integration guide.

Product Experience

Planning chat, not a rigid form.

Model Routing

Routes by exact model, not just provider.

Risk & Fit

Architecture, security, and multi-file work require stronger reasoning.

Cost & Latency

Small documentation or formatting tasks can use cheaper, faster models.

Context & Effort

Model context window and reasoning settings are part of routing.

previewtask -> model -> rationale

              task-001 docs polish -> gemini flash-lite
              task-004 routing contract -> gpt-5.5
              task-006 integration validation -> gpt-5.5 or opus
            

Artifacts

Versionable handoffs for worker agents.

MVP Status

Working planner-only harness.

Version 1 is focused on planning and handoff generation. It does not execute worker agents automatically.

TypeScript CLI/TUI with Ink and Commander.
OpenAI/Codex, Claude Code, and Gemini CLI integration paths.
Provider registry and selected model pool.
Deterministic and LLM planning engines.
Handoff linting, export, and smart revise.
CI passing with 111 tests.

Roadmap

From planner to supervisor.

Blueprint 2.0 can run workers from the same terminal harness, track progress, handle failures, and integrate results.

visionsingle AI development cockpit

              1.0 plan -> route -> handoff
              2.0 supervise -> run workers -> integrate
              demo marcellopps283.github.io/cli-planner-agent
              repo github.com/marcellopps283/cli-planner-agent
            