Orchestrate multi-stage maintenance pipeline for linting plugin development

This skill orchestrates maintenance work on the brooks-lint repo itself, running a sequential subagent pipeline where each stage is a dedicated agent spawned with the model always set to opus. Phase 0 determines run mode - a partial re-run reusing prior notes if resuming part of an earlier run, a new run that archives the old workspace, or an initial run - keeping run notes and QA verdicts in a workspace folder while the real artifacts are the repo files themselves. Phase 1 classifies the request against a table to pick the minimal necessary stage set, whether that's adding a skill, editing content, editing the shared framework, refreshing the eval suite only, fixing trigger descriptions, cutting a release, or a full change plus release - the QA stage is never skipped regardless of which other stages run. Phase 2 runs the selected stages in order: skill-author creates or edits content, invoking a new-skill scaffold for brand-new skills; eval-curator adds paired happy-path and false-positive eval scenarios if risk codes or modes changed, then runs the eval suite; consistency-qa is a mandatory gate running validation, tests, and evals plus cross-document sync checks across manifests, the README badge, the changelog, and book or eval counts, producing a PASS or FAIL verdict that loops back to the author or eval-curator on failure, retried once before stopping and reporting to the maintainer; trigger-boundary-auditor runs only if a description field changed, read-only auditing all shipped skills' trigger surfaces for false-triggering or routing collisions and looping back on a real collision; release-manager runs only if a release was requested and only after QA passes, cutting the release through a companion release skill. Phase 3 reports the stages run, files changed, QA verdict, audit findings, and release URL, then invites maintainer feedback logged to a harness change-log table.

Enforced conventions include every Agent call using opus, consistency-qa running as a general-purpose agent since it executes npm scripts while the trigger auditor stays read-only, no slash commands being created since short forms auto-install via a session-start hook, and changes pushing directly to main without a pull request per repo convention, though a simplify-review-commit gate still applies to non-documentation edits. Error handling retries a failed stage once with its error as input, stops on a second failure rather than silently skipping it, never lets a QA failure proceed to release, and requires explicit maintainer authorization for high-risk git operations like force pushes or history rewrites.

Use it only for maintaining the brooks-lint plugin's own repository - adding or editing skills, refreshing evals, keeping manifests, README, and changelog in sync, auditing trigger boundaries, or cutting a release. It is not applicable outside this specific repo's maintenance workflow.

Given a maintenance request, it produces a classified stage plan, edited skill, eval, and manifest files, a QA PASS or FAIL verdict with cross-document sync confirmation, optional trigger-audit findings, and an optional release URL.

The Agent tool spawning dedicated subagents - skill-author, eval-curator, consistency-qa, trigger-boundary-auditor, and release-manager - defined in the repo's agent configuration, npm scripts for the QA gate, and a companion release skill for cutting releases.

Maintainers of the brooks-lint plugin who need a consistent, gated pipeline for adding or editing skills, keeping documentation and manifests in sync, and safely cutting releases without manual cross-checking.