← ClaudeAtlas

prp-executionlisted

Execute a previously-authored Product Requirements Prompt against the Ignition framework's domain skills. Use when the user asks to "execute", "run", "build", or "work through" a PRP ("execute prp/production-log-dashboard.md", "let's build the plan", "start the implementation from the PRP", "work through the blueprint"). Use when the user references a PRP file path directly. The skill loads the PRP, validates its structure, plans task execution, delegates each task to the owning domain skill, enforces Level 1 / 2 / 3 validation gates between tasks, and produces a completion report. It does NOT author new PRPs — that's `prp-create`. It does NOT skip the PRP — if the user hasn't written one, redirect. Escalates immediately (stops and asks the user) when: confidence on a task falls below 0.85, Level 1/2 validation fails twice in a row for the same artifact, a cited file or skill is missing, or the PRP contradicts the runtime observation.
aimerfan/ignition-skills · ★ 0 · AI & Automation · score 60
Install: claude install-skill aimerfan/ignition-skills
# Ignition PRP Execution ## What this skill does Consumes a PRP at `<project-root>/prp/{slug}.md` and produces the artifacts it calls for — tag JSON, Named Queries, DDL, and (gracefully degrading) any artifact types whose domain skills don't exist yet. The skill is the operational mirror of `prp-create`: authoring decides *what* to build, execution decides *how*, runs it task-by-task, verifies, and reports. The two skills share the PRP section contract but have distinct workflows. Only invoke this skill when a PRP file exists and has been approved (`Status: approved`). This skill preserves the CM PRP-execute discipline (ULTRATHINK plan, validation gates, confidence scoring, escalation) without the multi-agent machinery. A single Copilot runs the loop; domain skills are "loaded" by reading their SKILL.md + references, not spawned as subagents. ## Critical preconditions — non-skippable Before doing anything: 1. **The PRP file exists.** Confirm `<project-root>/prp/{slug}.md` is present. If not, redirect: > "I don't see a PRP at `prp/{slug}.md`. If this is a planning conversation, invoke `prp-create` instead. If the PRP lives elsewhere, share the path." 2. **The PRP validates structurally.** Run: ```bash python .github/skills/prp-create/scripts/validate_prp.py <project-root>/prp/{slug}.md ``` Exit 0 or 2 → proceed. Exit 1 → stop and tell the user the PRP is malformed; they need to fix it (or re-invoke the authoring skill). 3. **The PRP is approved.** Ch