← ClaudeAtlas

spec-driven-qalisted

Validates code quality through spec-driven QA validation with structural anti-skip enforcement. Replicates all 6 phases of the DevForgeAI QA workflow using the Execute-Verify-Gate pattern at every step. Designed to prevent token optimization bias through lean orchestration, fresh-context subagent delegation, and binary CLI gate enforcement. Enforces test coverage (95%/85%/80% strict thresholds), detects anti-patterns, validates spec compliance, and analyzes code quality metrics. Use when validating implementations, ensuring quality standards, or preparing for release. Always use this skill when the user runs /qa or mentions QA validation, quality checks, or coverage analysis.
bankielewicz/DevForgeAI · ★ 4 · Testing & QA · score 66
Install: claude install-skill bankielewicz/DevForgeAI
# Spec-Driven QA Validation Validate story implementations through strict 6-phase QA workflow while enforcing coverage thresholds, anti-pattern detection, and spec compliance. **Context files are THE LAW:** tech-stack.md, source-tree/, dependencies.md, coding-standards.md, architecture-constraints.md, anti-patterns.md. If ambiguous or conflicting → HALT and use AskUserQuestion. --- ## Anti-shortcut policy (ADR-062 / Tier 1.5 / D7) When encountering obstacles, do not use destructive actions as a shortcut. For example, don't bypass safety checks (e.g. `--no-verify`) or discard unfamiliar files. Specifically for QA gates: - **Do not bypass conviction gates by manually invoking hook scripts via Bash.** The PostToolUse `conviction-postrecord.sh` is designed to fire automatically when the framework dispatches the `phase-record` Bash call. Piping synthetic JSON to the hook directly forges the evidence trail. - **Do not `Edit()` subagent worklog files post-hoc** to inject required schema fields (`subagent_id`, `phase_state_checksum`, etc.). Re-invoke the subagent so it produces a compliant worklog naturally. - **Do not write placeholder content to satisfy substance checks.** Phrases like "placeholder", "to satisfy", "to bypass", "TODO", "TBD", and "filler" are detected and rejected by `conviction.py:_validate_subagent_worklog()` (D3). Writing fake content to satisfy a check that catches fake content is the antipattern the check exists to prevent. - **Avoid focusing on passing t