[pranshuchittora]

Adds an optional agent-qa skill for natural-language web, Android, and iOS testing with vostride's advanced memidex memory and self-improvement. The skill covers setup, authoring, execution, and review of recorded evidence, with two task-specific references and a generated README catalog entry. It distinguishes preparation from application results and preserves run identity, secret placeholders, and knowledge-correction permissions.

Validation:

• All 99 native unit tests pass under Python 3.12. On macOS the initial run hit /var versus /private/var fixture-path mismatches; rerunning with a canonical temporary directory passed without changing tests or implementation.
• README generation check, gh skill publish --dry-run, generic skill validation, and git diff --check pass. Native ZIP packaging contains exactly the three intended skill files, matching their source bytes.
• Native gh skill install --from-local installs the skill at project scope for Claude Code and Codex; content matches after the installer's documented tracking-metadata/formatting transformations.
• Against published agent-qa 0.2.3, discovery, validation, dry-run selection, and completed-run listing pass. The exact MCP example, with its two path placeholders substituted, launches from an unrelated directory, discovers 20 tools, and returns an empty completed-run list from an initialized fixture.

The integration check covers installation, startup, and supported preparation/read operations. No model-backed application journey, completed-run evidence walkthrough, or automatic host activation was exercised. No runtime code or public API is changed.