AI Agent Workflow Collection

Machine-legible accuracy — write claims an agent can act on without judgment, and treat documentation as another form of code. Verifies signatures against real exports, paths against the filesystem, counts against a real enumeration, and commands against the manifest that defines them; prefers omission over invention, and never launders an unverified guess as a hedge. A wrong doc is a bug. Use when writing or updating machine-readable docs, registries, API references, or any doc whose claims an agent will act on. Invokable by users and agents.

Codebase architecture standards and best practices for {{project.longName}}. Covers file structure conventions, module patterns, naming conventions, dependency rules, and code organization. Use when enforcing or communicating architectural decisions.

Maintains documentation files (README.md, .claude/claude.md, .claude/agent.md, API docs) for Unite-Hub. Ensures docs stay in sync with codebase changes. Uses Haiku model for cost-effective documentation generation.

Human-readable documentation standard — scannable, plain-language docs derived from the agent docs but formatted for human consumption. Defines the project's human-doc file set and format rules. Use when creating or updating documentation intended for human stakeholders.

Security auditing for the {{project.name}} plugin. Covers secret detection, input validation, command injection prevention, API key handling, and .gitignore enforcement. Use when reviewing code for security vulnerabilities or setting up security guardrails.

Expo development workflow for {{project.name}} — Expo Go vs development builds, prebuild/CNG, running on simulators, emulators, and devices, and EAS Build/Submit/Update. Use when setting up, running, or shipping the app.

Synchronize AI swarms with GitHub Projects for visual task management and progress tracking. Use for project board automation, task synchronization, sprint management, and team coordination with GitHub Projects.

Execute the label→action mapping when a GitHub trigger label dispatches the CI harness: enrich an issue in place, or implement it end-to-end (branch → PR). Invoked by the waffle-label-hook workflow with a constant action token; also user-invocable to test the hook locally.

Use markdown's full formatting range — tables, callouts, task lists, collapsible sections, code fences, bold-lead bullets — to build tasteful richness in docs. Chooses the form from the shape of the content (parallel items become tables, sequences become ordered lists, optional depth becomes a collapsible) and holds every formatting choice to one test — does it speed up a scanning reader. Richness in service of scanning, never decoration. Use when writing or restructuring human-facing markdown. Invokable by users and agents.

Obsidian plugin development patterns, API usage, manifest configuration, settings management, testing, and build pipeline. Use when working on plugin architecture, lifecycle hooks, tests, or Obsidian API integration.

Triage the review findings on a pull request — read every review and review comment (from the adversarial-review bot or a human), score each finding on a four-dimension rubric (severity, validity, effort/risk, alignment), decide Implement / Defer / Decline with a recorded score and one-line reason, apply the accepted fixes, and post one reply per round summarizing the verdicts. Use when a PR has review feedback to answer, when the user says "respond to the review", "address the findings", or "triage the PR comments". Invokable by users and agents.

Check whether this repo's rendered wafflestack files still match the lock (drift detection). Use to diagnose a CI doctor failure or before committing; explains each modified/missing/stale-render finding.

Stop wafflestack managing one item (skill/agent/file) and hand it to the project. Confirm first — the file stays in place but becomes project-owned and stops receiving toolkit updates.

Scaffold a starter .waffle/waffle.yaml so wafflestack can render into this repo. Use in a repo that has never run wafflestack, before picking stacks and running /waffle-render.

Install wafflestack stacks or individual items into this repo — persists the refs to .waffle/waffle.yaml, then re-renders. Use to add a stack (e.g. github-workflow) or an item (skills/issue).

Re-render this repo's current wafflestack selection into harness-native files (.claude/ etc.) and refresh the lock. Use after editing .waffle/waffle.yaml or upgrading the toolkit.

Print the wafflestack setup playbook and full stack/skill/agent inventory for this repo, then help pick stacks and config. Use to plan an install or audit what the toolkit offers.

Move this repo's wafflestack pin to the toolkit's current version — runs migrations, re-renders, and doctors. Use to adopt a new toolkit release; review the diff before committing.

Validate wafflestack's own toolkit definitions — manifests, frontmatter, and placeholder declarations. Use when authoring or forking the toolkit itself, not for a plain consuming repo.

Security auditing for {{project.name}}, a browser-delivered web app. Covers dependency hygiene, XSS prevention, secret leakage in client bundles, validation of third-party data, CORS / mixed content, browser-storage safety, and CI workflow safety. Use when reviewing code for security vulnerabilities or setting up security guardrails.