Dev pilot
SaaS factory: Next.js starters + spec-kit extensions + Ralph++ autonomous loop. Production-ready in 48h.
npx -y skills add JimmyBlanquet/project-forge --skill dev-pilotAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Full autonomous pipeline orchestrator that loops over features.json and executes the complete development cycle for each feature: spec → plan → tasks → implement → validate → deploy → smoke test. Use this when the user wants to run the autonomous development pipeline, process all pending features, execute the full dev cycle end-to-end, go from features.json to deployed features automatically, or automate the entire SDLC for a project. Also triggers on "dev pilot", "dev-pilot", "run the pipeline", "process all features", "autonomous mode", "full cycle", or "pipeline run". Manages feature lifecycle in features.json (todo → specifying → implementing → validating → staging → done) with escalation on failures. This does NOT implement individual features (use ralph-loop), run individual spec steps (use speckit-* skills), or manage infrastructure. It is the top-level orchestrator that chains all other tools together.
SKILL.md
6.5 KB, as published. Nobody here has run it
User Input
$ARGUMENTS
You MUST consider the user input before proceeding (if not empty).
Overview
Dev-pilot is the top-level orchestrator for the Project-Forge autonomous pipeline. It reads features.json, processes each feature in priority order, and drives the entire cycle:
features.json → for each feature (by priority):
/speckit-feature → spec + plan + tasks + prd
tools/ralph → implementation (parallel stories)
tools/validate-feature → typecheck + lint + tests + build
tools/deploy-staging → push + Vercel preview
tools/smoke-test → verify routes
[HUMAN REVIEW] → approve + promote
features.json status=done
Prerequisites
Before running, verify:
- Project has
.speckit/features.jsonwith at least one featurestatus=todo tools/ralph,tools/validate-feature,tools/deploy-staging,tools/smoke-testexist and are executable- Git repo is clean (no uncommitted changes)
- Required CLIs available:
gh,vercel,claude(check withcommand -v)
If prerequisites fail, report what's missing and stop.
Step 1: Load and Sort Features
# Read features.json
cat .speckit/features.json
Sort features by:
- Priority: P1 before P2 before P3
- Within same priority: by ID (F01 before F02)
Filter to features with status=todo or status=specifying (resume interrupted work).
If no features to process, report "All features are done or in progress" and stop.
Step 2: Process Each Feature
For each feature in priority order:
2a. Specification Phase
If feature status is todo:
- Run
/speckit-feature {feature-id}(chains specify → plan → tasks → convert) - This updates status to
implementing - On failure: Log error, skip feature, continue to next
2b. Implementation Phase
If feature status is implementing:
- Find the ralph session from features.json
prdSession - Run
tools/ralph --session {session-id} --auto-merge - On failure (3 retries exhausted on a story):
- Update features.json:
status=blocked - Log: "Feature {id} blocked: story {N} failed after 3 retries"
- Continue to next feature
- Update features.json:
2c. Validation Phase
After ralph completes:
- Run
tools/validate-feature - Update status to
validating - On failure:
- Log validation errors
- Keep status as
implementing(don't advance) - Continue to next feature
2d. Staging Phase
After validation passes:
- Run
tools/deploy-staging - Update status to
staging - Read the preview URL from
.staging-url - On failure:
- Log deployment error
- Keep status as
validating - Continue to next feature
2e. Smoke Test Phase
After deployment:
- Run
tools/smoke-test - On failure:
- Log failed routes
- Keep status as
staging(deployed but not verified) - Continue to next feature
2f. Human Review Gate
After smoke tests pass:
- Report: "Feature {id} ({name}) is ready for review"
- Print preview URL
- Print summary of changes (stories completed, files modified)
- STOP and wait for human input
The human can:
approve→ runtools/promote-to-prod --auto-merge, set status todoneskip→ leave instaging, move to next featurereject→ set status toimplementing, move to next featureabort→ stop the entire pipeline
Step 3: Report
After all features processed (or pipeline aborted):
Dev-Pilot Pipeline Report
═════════════════════════
Features Processed: {N}
F01 auth done ✓ (promoted to prod)
F02 payments staging ⏳ (awaiting review)
F03 dashboard blocked ✗ (story 4 failed)
F04 settings todo — (not started)
Summary:
Completed: {N}
Staging: {N}
Blocked: {N}
Remaining: {N}
Next actions:
- Review F02 at https://preview-url.vercel.app
- Fix F03 story 4 manually, then re-run dev-pilot
Escalation Rules
| Scenario | Action | Status |
|---|---|---|
| speckit-feature fails | Skip feature, log error | remains todo |
| Ralph story fails 3x | Skip feature, log blocker | blocked |
| Validation fails | Don't advance, log errors | implementing |
| Deploy fails | Don't advance, log error | validating |
| Smoke test fails | Keep deployed, log routes | staging |
| Human rejects | Reset to implementing | implementing |
Mutation Testing (Optional)
After validation passes (step 2c) and before deploy (step 2d):
- If
stryker.config.jsonexists, runpnpm test:mutation - If mutation score < 50%: warn "Tests may be insufficient" but don't block
- Log mutation score in the report
Configuration
Dev-pilot reads configuration from features.json top level:
{
"project": "my-project",
"config": {
"autoMerge": true,
"skipMutationTesting": false,
"ralphModel": "sonnet",
"ralphBudget": "2.00"
},
"features": [...]
}
Defaults if config section missing:
autoMerge: false (require human approval)skipMutationTesting: true (opt-in)ralphModel: "sonnet"ralphBudget: "2.00"
Headless Mode
Dev-pilot can run in fully headless mode (no human review gate):
- Pass
--headlessflag - All features auto-promote after smoke tests pass
- Best for CI/CD environments
- Still stops on
blockedfeatures
Safety
- NEVER force-push or delete branches
- NEVER modify main directly
- Always create feature branches via ralph
- Always wait for CI before merge (auto-merge handles this)
- Preserve features.json as single source of truth