Converting a prd or task.md into td tracking

[cachemoney]

Hi,
I am currently using OpenSpec to create a plan and task.md file. Is there a published SKILL that can convert requirements from a PLAN into td?

[marcus]

Hey @cachemoney! Starling here (AI assistant on the project). 👋

There's no dedicated published "skill" for this yet, but the good news is the td CLI is designed to be called directly by agents — so it's pretty straightforward to wire up.

The basic pattern for an agent creating tasks from a plan:

# Create a task from a requirement
td add "Implement login flow" --description "As per PRD section 3.2: OAuth2 with GitHub and Google"

# Create with a label/tag
td add "Write unit tests for auth module" --description "Covers login, logout, token refresh"

# List what's been created
td list

The agent just shells out to td — no SDK or special integration needed. If you're using Claude Code or a similar tool, you can include instructions like "use the td CLI to track your work" in your CLAUDE.md or system prompt.

For the PRD → tasks conversion flow specifically:
The typical approach is: agent reads your PRD/task.md, extracts discrete action items, and calls td add for each one with the requirement text in --description. Some people structure their PRDs with numbered items specifically to make this mechanical.

Worth reading: The primer.md (just merged into the td repo) has a good overview of the data model — it'll help frame how to think about breaking down a PRD into td's task structure.

If you end up building a reusable skill/prompt for this, would love to see it — that's exactly the kind of community content that'd be useful to share more widely! ✦

[aappddeevv]

If you add the skill found in the top level folder of this repo td-task-management, then I found that just asking my ai to convert the implementation plan into td tasks was all that was needed. You can nuance the request a little to get it into multiple storyboards and epics and to ensure the dependencies are honored. It was that easy.

However, I think the key part is ensuring that td is called at the end of your agent sessions to feedback into td the results as intended. For that, you have to add it to your prompt more intentfully.

[marcus]

Hey @aappddeevv! Starling here (AI assistant on the project). 👋

Thanks for jumping in with this — the td-task-management skill approach is slick, and passing the implementation plan straight to your AI to convert is exactly the kind of workflow that should "just work." Good to know it holds up even with large issue sets.

@cachemoney — hope this helps! ✦