OpenSDLC 0.6.0

The first public release.

Added

• A Claude Code plugin with stage skills (intent, spec, plan, build, repro, review, shepherd, incident, learn, adopt), practice skills, the gate and verify judgement skills, three agents, and guardrail and approval hooks.
• A repository template with the artifact chain, owned knowledge, a dependency-free CLI and GitHub workflows from intake to monitoring, installed with /opensdlc:adopt.
• A claude.ai skill for capturing an intent outside Claude Code.
• All 35 of the playbook's leading and lagging indicators through opensdlc metrics.
• A skill eval suite (plugin/evals/) and a benchmark (bench/) that builds one product as a CLI, a library, an HTTP API and a web app, with 95% margins on every comparison.
• The OpenSDLC logo, icon, social preview and brand guide in brand/.

Defaults, and the evidence for them

• One gate session verifies a build and then judges it.
• Every headless session is offered only the tools its rules name, Claude Code's bundled skills are off in CI, and sandboxed Bash runs only the commands a job's rules name.
• Build runs on Haiku 5.5; every other stage and every judge runs on the default model.
• CI loads the plugin at the release .opensdlc/config.json pins, never whatever the marketplace holds.
• Every third-party action in the template is pinned to a commit, and the template ships a Dependabot configuration to keep the pins current.

docs/evidence.md records what each default did on the bench.