From sth-harness
Role-driven implementation loop that works one harness feature through its phases until blocked, a decision is needed, or the spec is done. Used by /sth-harness:implement-next-spec.
How this skill is triggered — by the user, by Claude, or both
Slash command
/sth-harness:harness-implementThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Drive one feature through the harness lifecycle, honoring minimal-context reading and durable state.
Drive one feature through the harness lifecycle, honoring minimal-context reading and durable state.
ai-harness/START-HERE.md once (if not already read this session).ai-harness/specs/<feature>/state.md: phase, next action, read budget, tasks.spec.md).ai-harness/roles.md (read only that role's section).Run task by task. For an implementation task:
doing in state.md.ai-harness/context/testing.md; if deferred, follow the DV plan and add a follow-up task.ai-harness/context/architecture.md.review, then done once its validation passes.state.md after every task.STOP and hand back when ANY of:
blocked, record blocker + what would unblock, set overall blocked);done (all tasks done/cancelled, requirements satisfied, validation run or scheduled).Update state.md (phase, last completed step, next action, task states). Report the stop reason and the single next action. The file is the truth — never rely on conversation memory.
Guides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.
npx claudepluginhub danielemiliogarcia/sth-harness-plugin --plugin sth-harness