Taskrail

Taskrail is a deterministic execution harness for AI agents that turns goals into structured tasks, keeps work aligned to an authoritative state file, and advances execution through validation, verification, and follow-up.

Status

This repository is bootstrapping Taskrail while also dogfooding the workflow style Taskrail is meant to standardize.

• Current target release: v0.1.0
• Active spec: specs/v0.1.0.md
• Planning state: planning/STATE.md

What Taskrail Owns

• repo-native specs under specs/
• deterministic tracked work under planning/
• one authoritative planning/STATE.md
• task validation and dependency checks
• deterministic next-task selection
• explicit task transitions
• verification artifacts and follow-up tasks

What Taskrail Does Not Own

• container sandboxing
• worktree lifecycle orchestration
• network policy enforcement
• distributed multi-agent runtime
• hosted control planes
• built-in LLM provider integrations in v0.1.0

Repository Layout

.
├── AGENTS.md
├── CHANGELOG.md
├── README.md
├── cmd/
├── docs/
├── internal/
├── planning/
├── scripts/
├── skills/
└── specs/

Local Commands

• Build: go build ./cmd/taskrail
• Test: go test ./...
• Validate Taskrail structure: go run ./cmd/taskrail validate
• Select next task: go run ./cmd/taskrail next
• Show CLI help: go run ./cmd/taskrail --help
• Skill mirror check: ./scripts/check-skill-mirrors.sh

Read Next

1. specs/v0.1.0.md
2. specs/v0.2.0.md
3. specs/v0.3.0.md
4. planning/STATE.md
5. AGENTS.md

Notes

• The versioned specs in specs/ are normative.
• planning/STATE.md is the authoritative execution state for this repository.
• Temporary workflow scaffolding in skills/, .agents/skills/, .claude/skills/, and docs/workflow/ exists to dogfood the Taskrail style before the product replaces it.