Implements one milestone from a work brief until planner-authored blocking tests pass, then delivers a PR.
npx skills add https://github.com/kpiteira/devops-ai --skill kbuild-kpiteira
Please help me install this Agent Skill. Skill: kbuild Source: https://github.com/kpiteira/devops-ai/tree/main/skills/kbuild Command: npx skills add https://github.com/kpiteira/devops-ai --skill kbuild-kpiteira
diverged, and stops for planner triage. - Structured PR delivery: Produces one PR per milestone with the spec mapping line, blocking test output, and three mandatory sections — Decisions I made alone, For the human, Facts I corrected. - Use Case: Point it at docs/specs/reminders/briefs/M1-storage.md; it sets the spec row to in progress, implements until the blocking tests pass, runs a fresh-context self-review, and opens a PR ready for /kbabysit review rounds. ## Quick Start Ask the AI to run kbuild on a work brief such as docs/specs/reminders/briefs/M1-storage.md and implement the milestone until its blocking tests pass.High-intent search queries and answers about installing and using this skill.
Invoke kbuild with the brief path, for example docs/specs/<feature>/briefs/M2-history.md. The executor reads only that brief and the current code, runs a goal loop against the brief's Blocking acceptance tests, and delivers the milestone as a PR.
The executor never edits acceptance tests; they are read-only contract. It writes a divergence report from the bundled template, sets the milestone's Decomposition row to diverged with the report path, commits, and stops for a planner triage session.
No. Acceptance tests are writable only in planning sessions, and gates like make check are fixed by changing code, never by editing thresholds or contracts. A CI guard rejects brief or acceptance-test edits from impl/ branches.
The PR body needs the spec mapping line, the blocking commands with final green output, a Review scope section listing the brief's jobs, and three sections: Decisions I made alone, For the human, and Facts I corrected, each present even when empty.
Stop when a stated fact is false, a decision conflicts with the actual codebase, or a test contradicts its job and proceeding would require a workaround, self-grading, or a widened security boundary. A wrong annotation on an otherwise buildable requirement is only a fact-correction.
ttphatsGuides collaborative design exploration through clarifying questions, approach proposals, and approved spec documents.
ZaxbyHubResolve pull request feedback with verified source evidence and closure summaries.
trailofbitsRuns an autonomous review-and-fix loop over branch changes until PR review passes.
understudy-aiFetch GitHub issues, spawn sub-agents to fix them, and open pull requests.