agentic-workflow
A risk-aware, evidence-based engineering lifecycle protocol for robust agentic task execution and safety.
skill install https://www.promptspace.in/skills/agentic-workflowWhat it does
The Agentic Workflow is a language-agnostic, risk-aware protocol designed to prevent the four most common AI agent failures: silent assumptions, scope creep, untested deployments, and credential leakage. It transforms vague requests into evidence-based execution by enforcing a multi-stage engineering lifecycle similar to a senior dev's mental model.
Why use this skill
Prompting an AI often results in "hallucinated" progress or unexpected side effects. This skill forces your agent to follow a structured discipline of research, planning, safety-gating, and verification. It is better than raw prompting because it includes built-in guardrails like cascading failure detection, secret scanning, and an absolute "never push" rule for production stability.
Supported tools
- Development Environments: Works inside or outside Git worktrees with filesystem fallback.
- Frameworks: Language-agnostic (Python, Go, Node, Rust, etc.).
- CI/CD: Comprehensive verification matrices and rollback planning.
- Safety: Integrated secret scanning and shell quoting safety protocols.
The Output
Instead of just code, you receive an audit-ready closure report including a Risk Class assessment, a Verification Matrix (Passed/Failed/Unstable), a Scope Check (Planned vs. Actual changes), and a Rollback Strategy. It ensures all work is staged, verified, and ready for human approval.
Use cases
- Enforce risk-gated execution for complex codebase refactors.
- Prevent credential leaks with mandatory safety-gate scanning during tasks.
- Generate comprehensive verification matrices for all code changes.
- Ensure structured, review-ready branch closures with rollback plans.
Example
Prompt
Output
## Verification - Passed: Unit tests, linting - Failed: 0 - Environment Delta: Node v18 -> v20 - Dependency Scan: Clean ## Summary Report - Commits pending push: [a1b2c3d, e4f5g6h] - Status: Verified & Awaiting human push. Agent did not push to remote.