Skip to content

Blind-validate clownware-go-tools skills before calling them stable #11

Description

@chrispezza

House convention #5 (docs/MAINTAINING.md): new skills get a blind validation run before release — a fresh agent executes each skill without knowing the expected outcome. clownware-go-tools shipped at v0.1.0 (now v0.2.0) with probe-level verification only (all 18 pre-fetch probes tested in bash+zsh across hit/miss/unrelated contexts) but no blind execution pass.

Skills to validate, each in go-performance-starter (the hit context) plus one non-starter Go repo (graceful-degradation context):

  • templ-component-scaffold — scaffold a component; verify typed props + generate step
  • test-scaffold — stub a table-driven test that compiles green
  • sqlc-query-scaffold — add a query + repo seam; verify task db:generate round-trip
  • perf-budget-check — verify it orchestrates task test:performance and reports per-gate verdicts
  • pr-description — on a real branch; verify task ci checklist and master base detection

The skill-validate skill (code-tools) exists for exactly this — pin ground truth, run blind, score reproduction.

🤖 Generated with Claude Code

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions