util
claudeUtility skills for the orc pipeline: quick commit/review/PR, Dependabot triage, issue-status listing, and repo scaffolding.
4 skills · 7 agents
Issue-to-PR automation: discuss and spec issues interactively, then build them autonomously with parallel subagents, focused verification, and a folded-in AI review, for a GitHub-issue-driven workflow.
Plugin orc from marketplace hexbyte in joshevensen/hexbyte-plugins. Commands read from the repo's manifest — 21 days ago.
createDiscuss an idea at the product/scope level and produce one or more GitHub issues — no spec-writing. Splits work into separate issues when it's naturally independent. Ends only when every open question is resolved. Interactive and local.
discussEnter discussion mode: explore a topic freely using read-only tools, but make no code or file changes until the user explicitly says they are ready.
doRun the full orc pipeline — create, plan, build — for one simple issue in a single pass, skipping the separate handoff between stages. For work small enough that three commands would be pure overhead. Invoke as /orc:do {description}.
explainExplain what a file, a route, or a PR is doing in plain language, as a chat reply. Read-only — writes no files, posts no comments. Invoke as /orc:explain {file path | route | pr number | pr url} [--junior|--intermediate|--senior].
planTake a draft issue from /orc:create, explore the codebase, choose a technical approach, and write a build-ready spec onto the issue. Sets status:ready when the spec is complete. Invoke as /orc:plan {number}.
respondFetch unresolved PR comments — yours and everyone else's — and resolve them, no interactive per-item confirmation. Manually triggered, never watches or subscribes to a PR. Never merges. Invoke as /orc:respond [pr].
resumeContinue a build that stopped after its PR reached ready-for-review — re-checks review freshness, then CI and mergeability, without redoing implementation. Use for a blocked issue past the review gate, not mid-implementation. Invoke as /orc:resume {number}.
walkthruWalk the engineer through a finished build as a junior developer reporting to their senior — what was asked, the approach, the choices made and why, whether each acceptance criterion is met, and the code that needs a close look — then end with a merge decision. Interactive, one item at a time; repla
changelog-writerReads the branch diff and existing CHANGELOG.md, then writes a new changelog entry in the repo's existing changelog convention. Used by /orc:build after implementation and review, as part of finishing up before merge readiness.
ci-debuggerInvoked by build or push when a CI check is failing. Given the failing check name, log output, and diff, diagnoses the root cause and proposes a specific fix.
conflict-classifierInspects merge conflict markers in the working tree and classifies each conflicted file as simple (auto-resolvable) or complex (needs human judgment). Invoked by build and push after a rebase surfaces conflicts during their merge-readiness check.
deploy-risk-scannerScans a git diff for deployment risks — database migrations, env var additions, background job schema changes, webhook/API contract changes. Returns a risk level and categorized list of findings. Invoked during the build/push review step alongside the four review agents; a HIGH risk level is treated
finding-verifierTries to disprove one review finding against the diff and spec before it reaches the user. Invoked by /orc:build and /orc:resume for each BLOCKER or WARNING a review agent raised.
implementerImplements one task from a build's dependency-ordered task list, touching only the files it owns. Invoked by /orc:build for each task in a wave; build overrides the model per task (see build step 7).
issue-loaderFetches a GitHub issue by number and extracts structured metadata — title, labels, status, spec comment, branch slug, and body. Used by plan and build to load issue context at the start of each step.
pr-context-writerWrites the reviewer-facing context for a built PR — a concise PR description plus inline and file-level review comments explaining non-obvious choices, what needs scrutiny, and what can be skimmed. Invoked by /orc:build and /orc:resume after review. Reasons come only from the build's decision log, n
review-correctnessInvoked during the build/push review step. Checks that a diff satisfies every spec acceptance criterion, detects scope drift against the Out of Scope list, and identifies logic errors or unhandled edge cases.
review-impactInvoked during the build/push review step. Checks for breaking changes to public interfaces, unexpected side effects, performance regressions, and database query changes.
review-qualityInvoked during the build/push review step. Checks complexity, consistency with surrounding codebase patterns, and test coverage of critical paths.
review-securityInvoked during the build/push review step. Checks for injection vectors, auth gaps, data exposure, insecure defaults, and input validation problems.
spec-criticIndependently audits a drafted spec before it is posted — checks that cited paths exist, criteria are verifiable, and scope is consistent. Invoked by /orc:plan after spec-writer.
spec-writerResearches a codebase and writes an implementation spec from the shared template. Invoked by /orc:plan for one issue. Writes the spec directly to the output path in the prompt.
test-writerWrites tests for a spec's behaviors from the spec alone, without reading the implementation (initial mode), then fills specific coverage gaps build finds (gap-fill mode). Invoked by /orc:build in parallel with the first implementation wave, and again in step 8b.
This page updates itself from your repo. Add the badge to your README, or get it Featured to pin it above the organic results with a followed link to your site.