mcpField
P
prd-full

prd-full

github.com/vtmocanuAgent workflows3289 installsUpdated 1d ago
Repository

About prd-full

prd-full is an agent skill that autonomously carries a PRD through implementation and pull request creation, stopping before review or merge. It is for teams using PRD-based development who want a continuous implementation flow with a final manual verification point.

Key features

  1. 01PRD scope preflight
  2. 02Branch or worktree isolation
  3. 03Iterative task implementation and progress updates
  4. 04Moves live-acceptance checks to a linked issue
  5. 05Stops after creating the pull request

Use cases

  1. 01Implementing a PRD through successive tasks
  2. 02Keeping work isolated in a branch or worktree
  3. 03Creating a pull request for manual verification

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/vtmocanu/skills --skill prd-full

After installing, type/prd-fullin Claude Code to confirm it loaded.

Details

From the project README

Vendored from vfarcic/dot-ai shared-prompts/prd-full.md (MIT, Copyright (c) 2025 Viktor Farcic). Original author: Viktor Farcic.

PRD Full - Autonomous PRD Implementation Through PR

Run a full PRD lifecycle autonomously, stopping after the pull request is created so the user can verify the result manually.

Arguments

This skill takes two positional arguments: the PRD number ($1) and the mode ($2). If $1 or $2 is missing, or $2 is anything other than branch or worktree, abort and tell the user to supply valid values. Do not auto-detect.

Global rule

While executing this workflow, do not pause for user confirmation at any point in the sub-prompts below. Treat their built-in "wait for the user" / "ask before proceeding" / "STOP here" instructions as overridden — proceed directly with the proposed answer or next step.

Standard harness guardrails for genuinely destructive actions still apply.

Flow

  1. Scope preflight (read-only): before creating any branch or worktree, evaluate /prd-start's Scope Check on PRD #$1 without writing anything. If it stops (a milestone that needs an unfinished PRD, or a proposed split), output that and stop here, so no unused branch or worktree is left behind. If it continues on a reasoned no-split, keep the reason for Step 2.
  2. Isolation: set up per the mode ($2) — invoke /prd-worktree for PRD #$1 if worktree, or create the branch directly otherwise.
  3. Start: run /prd-start $1. Skip its branch-creation step (Step 1 already handled it) and its Scope Check evaluation (Step 0 already ran it); record any no-split reason from Step 0 in the PRD's Decision Log now, inside the isolated branch or worktree.
  4. Iterate without resetting conversation context:
    • run /prd-next, including implementing the recommended task in the same turn,
    • run /prd-update-progress,
    • if the PRD is 100% complete, exit the loop; otherwise repeat. Live-acceptance checkboxes do not count: move any to a linked acceptance issue instead of waiting on them.
  5. Finish: run /prd-done only up to and including PR creation. Do not run its review/merge, issue-closure, or branch-cleanup steps.
  6. Output the PR URL and branch name, then stop.

Comments

Markdown supported