pdd-prompting
The PDD Prompting guide. Use when you write, change or review a module prompt (*.prompt.md), a Story or test cases, when the person asks how to write prompts in PDD, or asks to set up real behavior verification for a repository. In PDD, prompts are the source and code follows them.
How do I install this agent skill?
npx skills add https://github.com/promptdriven/pdd --skill pdd-promptingIs this agent skill safe to install?
- Gen Agent Trust Hubpass
This skill provides a framework for Prompt-Driven Development (PDD), guiding the agent through writing, testing, and verifying AI-driven modules. While the skill is structured with security best practices (e.g., using sanitizers for HTML, isolating external effects via ports), it processes user-provided markdown files which creates a surface for indirect prompt injections. It also facilitates the execution of local repository code during testing phases.
- Socketpass
No alerts
- Snykfail
Risk: CRITICAL · 2 issues
What does this agent skill do?
PDD prompting
Each module's .prompt.md is the source; code implements it.
Core rules
- Read
RUNBOOK.mdfirst,guide.mdin full, andstory-guide.md; skiprunbook/*.mdchapters unless a workflow step or lint error points to one. - Finish Prompt Space (
pdd lint --local-> passingpdd test) before code; change no code and start no interactivepdd simulateunless asked. - One prompt per module: purpose,
## Domain decisionsand failure edges. Give each rule one owner prompt. Link shared rules. For a CLI, say if a rule belongs to the whole-CLI prompt or to one module's prompt. - Inject outside effects (DB, files, net, clock, RNG, IDs) via inputs or
<port>.<op>ports; never use prose counters or linkkind: native-onlyadapters under## Dependencies(RUNBOOK §21). - Linked
.ui.svgis the layout spec: parts, words, order, ids and states. Check withpdd lint --local; render PNGs only if asked (RUNBOOK §13). - Write Short JSON scenarios with self-contained step inputs (
seeds+$use/patch) for parallel steps; chain$steporworldonly for persistence, with1observable check per claim; never copy code output. - Show prompt, Story and cases first. Record actual chat approval with --said (RUNBOOK §2), never invent it.
- Do not run
pdd lintbeforepdd test; test includes lint. Report only real output.
Files
RUNBOOK.md: concise topic router and core commands (pdd test,pdd lint --local,pdd simulate).guide.md: the Prompting guide.story-guide.md: Story sizing and criteria rules.review.md,verification-setup.md,verification-recipe.md,autopilot-job-rules.md: check, code and Autopilot references.
How to write
- Name applicable
guide.mdrules and find the owning prompt. - Write or change the prompt. Use
0Stories +"No Story:"for internal changes, edit the existing Story, or write 1 cohesive standalone-loop Story in the person's words, with no function calls or type names. - Derive normal, limit and error cases (
<= 15steps; alignsimulation_overridesports) and runpdd files+pdd lint --local. - Show drafts and open choices.
- After changing prompt, Story and tests, run
pdd testas RUNBOOK.md says, and show the person thePrompt testsline(s) and the report link(s). Offerpdd simulate STORY. If a check fails, show it; never change the expected value yourself.
Without prompts, use code, callers and tests as evidence. Change no code until asked.
Workflow skills: pdd-small-change, pdd-bug, pdd-feature and pdd-product.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/promptdriven/pdd/pdd-prompting">View pdd-prompting on skillZs</a>