sddl-eval-level-3

Runs the sdd-lite main orchestration flow end to end in a disposable fixture with human approval gates.

Updated Apr 2, 2026
One-click install
npx skills add https://github.com/nico0695/ai-tools --skill sddl-eval-level-3-nico0695
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: sddl-eval-level-3
Source: https://github.com/nico0695/ai-tools/tree/main/sdd/sdd-lite-evals/skills/sddl-eval-level-3
Command: npx skills add https://github.com/nico0695/ai-tools --skill sddl-eval-level-3-nico0695

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Validating changes to the sdd-lite main orchestration flow requires exercising the full pipeline—proposal, spec, design, plan, approvals, executor, and QA—without risking real workspaces, and this Skill automates that end-to-end release-confidence check. ## Core Features & Use Cases - Full-flow execution: Drives the installed sdd-lite CLI through every stage from proposal to QA inside a disposable fixture workspace. - Human approval loop: Pauses at each mandatory gate, presents run id, stage, scope, risks, and report path, and resumes only after explicit stage approval. - Evidence-based analysis: Fills an analysis template from observable evidence and attaches it via the harness so semantic conclusions enter the report and local history. - Use Case: After modifying the main orchestration flow, run a Level 3 eval against a reviewed case to confirm every mandatory gate still passes before release. ## Quick Start Ask the AI to run a Level 3 sdd-lite eval for a given workspace, case, and provider, previewing first and then executing with approvals at each stage.

Frequently Asked Questions about sddl-eval-level-3

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an end-to-end eval of the sdd-lite main flow?▼

Run the harness script with the level3 subcommand, providing a workspace id, case, and provider. Preview first without --execute, then repeat with --execute after confirming the cost and model implications.

How do I approve a stage when the eval returns awaiting_approval?▼

Review the persisted plan, state summary, assertions, and source delta, then resume with --resume-run and --approve-stage naming the exact stage. Approval must be explicit per stage and never inferred from the original task.

What prerequisites does the Level 3 eval require?▼

It requires a passing or non-blocking Level 1 eval and a reviewed case before starting. The default case excludes 4R, Judgment Day, delivery, and archive stages.

Does the Level 3 eval have an automatic cost cap?▼

No, the first child run uses the CLI default model with no automatic cost cap and runs only through planning. Always preview the command and explain this to the user before executing.

Why must each continuation recover from files instead of conversation history?▼

Each continuation launches a new CLI process with no shared memory, so all state must be recovered from persisted files on disk. This ensures the eval faithfully reproduces real orchestration behavior across process boundaries.