DevSpecs is a local-first command-line tool for AI-assisted brownfield development scenarios. Its core goal is to preserve a persistent "intent trail" for the work produced by AI coding agents in a repository, so that developers or agents returning after some time away can still understand why changes exist, what state they are currently in, and what to do next.
Purpose
DevSpecs solves the problem of "losing the trail" in the work of AI coding agents. Git can show what code changed, but it cannot explain why it changed or what to focus on next. DevSpecs adds a local trail layer on top of Git and rg: why changes exist, what lifecycle state they are in, superseded work, and the next entry point for continuing. It is aimed at developers who need to collaborate with existing coding agents and want to preserve recoverable context.
Main capabilities
- Local intent trail: attach a persistent trail of AI coding work to the repository, recording what the agent changed, why it exists, and what should be reviewed next.
- Context recovery: help restore active trails, lifecycle states, and possible next-step context before reviewing or continuing work.
- Evidence linking: read repository files, documents, specs, plans, commit records, and task records to keep evidence, decisions, checkpoints, and bounded handoffs connected.
- Task slicing and checkpoints: can create lightweight task specs, including packaged context, gates, iteration slices, and checkpoints.
- Collaboration with existing coding agents: does not replace existing agents, but outputs context they can use.
- Local indexing: writes local task files and a local SQLite index.
How to use
DevSpecs is provided as a command-line tool, and after installation uses the ds command. Official materials mention that you can try it in a repository within five minutes:
- When you feel unfamiliar with the context upon returning, you can first run ds recent.
- When the goal is already clear, you can use ds task.
- On the first run in a large repository, DevSpecs will show indexing progress while staying quiet and keeping JSON output clean.
- It is recommended to try it in a disposable checkout of a real repository, for example using FastAPI to experience the trail recovery and task handoff loop without touching your own code first; when done, you candelete that checkout.
Suitable scenarios
Suitable for developers working with existing coding agents who need to review, inspect, or continue work after the agent runs; suitable for teams that want to preserve a local AI work layer without replacing transactional trackers such as Jira, Linear, GitHub Issues, or Notion. It can also serve as a local navigation layer for intent already scattered throughout the repository.
Limitations and notes
- Does not call an LLM: the CLI is local-first and deliberately kept simple, only indexing local artifacts and outputting context for existing agents to use.
- No cloud sync: all processing is done locally.
- Does not replace transactional trackers: Jira, Linear, GitHub Issues, or Notion should still be retained; DevSpecs only tracks the local AI work layer beneath them.
- Relationship with spec frameworks such as OpenSpec: partially similar but broader in scope; it can both create lightweight task specs with packaged context, gates, iteration slices, and checkpoints, and serve as a local navigation layer.
- Official materials do not mention information such as fees, pricing, or a specific platform support list, so it is not provided for now.
