A Claude Code skill that runs a four-phase workflow to complete and ship a task end-to-end, with a confidence gate before commit.
When invoked (e.g. /ship-task, or "ship the next task"), the skill runs four phases in strict order:
- Complete the next task — from a task document or a beads (
br/beads_rust) backlog. - Confidence check (4-axis rubric) — score the change on four 0–25 axes: Requirements Fit, Functional Robustness, Verification Evidence, System Safety & Integration. Each axis demands concrete evidence (executed tests, probes, real-environment runs — not "I read the code"). Total must be ≥95 with no axis below 15, no rounding. Iterate until the gate passes. Optionally escalate to an external Grok second opinion (user relays the prompt and response) when confidence stalls on genuine uncertainty.
- Update documentation — only what actually changed or was learned this session.
- Commit — via the project's
/commitskill if present, otherwise a plaingitcommit. Does not push unless asked.
The confidence gate is the point: an honest 78% that leads to fixing three gaps is the value.
git clone https://github.com/shaal/skill-ship-task.git ~/.claude/skills/ship-taskNote: the repo is skill-ship-task but the skill folder is ship-task — the folder name should match the name: field in SKILL.md frontmatter, which is what Claude Code uses to load and invoke the skill.
git clone https://github.com/shaal/skill-ship-task.git <your-repo>/.claude/skills/ship-taskThen commit .claude/skills/ship-task/ into your project so collaborators get it automatically.
cd ~/.claude/skills/ship-task && git pullSKILL.md is a single self-contained file. Edit it freely — the frontmatter description is the trigger text Claude uses to decide when to invoke the skill, so keep that field accurate if you change the workflow.
MIT — see LICENSE.